-
Notifications
You must be signed in to change notification settings - Fork 45
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Use a single thread within the SDK #187
Comments
Hi. Rewriting the SDK to use async semantics instead of threads is something other customers have suggested, and we're not ruling it out— but it would be a full rewrite; there is no way to patch the current SDK code to be single-threaded, without entirely removing major functionality that developers expect (like streaming updates and database integrations). However, I'm puzzled by the symptom of extremely large memory usage that you're seeing. It's hard for me to see how it could be entirely due to the use of worker threads as you're suggesting, so I'm wondering if there may be something else going on that we should look into. I say that for two reasons:
About that last point: if you have a very large amount of flag/segment data, the SDK has to load, parse, and store that data; and if you are generating a very large number of analytics events, the SDK has to buffer those events before they can be sent. A figure like "more than 1Gb of memory per instance" would be highly abnormal unless one or both of those things was the case. But if one of those things is the case, then I would tend to think the resource consumption would be fairly similar whether we used threads or not. So, I'm wondering if you could say a little more about 1. what kind of patches you've been doing, 2. what impact on memory usage you saw from them, and 3. the approximate size of your flag/segment data (you can get this by doing |
I'll preface this with I did my testing inside a docker container running with platform When I was testing, I put in a breakpoint and monitored memory consumption side by side and as I stepped through the code every time it launched a new thread I saw around a 75-100 MB increase in memory consumption for the process. In total I patched out 6 of the threads used by the sdk and saw ~600MB reduction in total memory consumption compared to no patch. As for the patch itself, it is below but I'll preface it with I took a path of changing as little as I could within the sdk (I agree it would take a significant rewrite of the sdk to cleanly reduce the number of threads) to achieve this so similar effect could be achieve much more cleanly with some more changes inside the SDK.
|
Using one scheduler thread for all repeating tasks is a good idea that I think we should adopt. That still isn't going to get you anywhere near having only a single thread for the whole SDK, though. And I don't think it's possible to use a similar approach to merge the SDK's other worker threads; they are just doing completely different things. I am still very puzzled as to how simply starting a thread in Python, especially very simple ones with almost no data of their own like the timer threads, could cause such a large jump in memory usage. And in order to try to reproduce this, I would need to know a bit more about your operating environment beyond just the host OS and the fact that you're using Docker. What version of Python is this? And is it the standard CPython runtime, or a variant? Also, what mechanism are you using to measure memory usage? |
I'm sure most of this can use a wider variety of versions, but dumping anything that might be helpful: Host OS: Mac 12.6 python version is the built in for the image with no special configurations/variations |
* make verify_ssl=False turn off certificate verification too (#129) * add more TLS config options and collect HTTP/HTTPS config options in a class (#130) * make stream retry/backoff/jitter behavior consistent with other SDKs + improve testing (#131) * streams shouldn't use the same read timeout as the rest of the SDK (#132) * implement our own retry logic & logging for event posts, don't use urllib3.Retry (#133) * remove support for indirect/patch and indirect/put * remove unused logic for individual flag/segment poll for indirect/patch * Ehaisley/84082/remove python2 (#136) * remove all references to six and remove queue fallback imports * remove NullHandler logger backwards compat * update circleci config to remove python 2.7 tests * remove ordereddict backwards compat * update setup.py to no longer list python 2.7 as compatible * no longer inherit from object for python 2 backwards compat * update readme and manifest to reflect python 2.7 removal * remove unicode type compatibility * remove 2.7 support from circleci * Allow authenticating with proxy This commit allows for authenticating with a proxy configured with the `http_proxy` environment variable. Authentication requires passing a header, and is not parsed by urllib3 from the proxy_url. * reimplement proxy tests for DRY and add test of proxy auth params * doc comment on auth params in proxy URL * add type hints to some of the public facing api. update some docs * Revert "add type hints to some of the public facing api." This reverts commit c35fa61. * Ehaisley/ch86857/type hints (#138) * add typehints to the public API * validate typehints in the public api and tests with mypy * remove all current deprecations (#139) * remove all currently deprecated classes, methods, arguments, and tests * also update semver usage to remove calls to deprecated functions and classes * remove global set_sdk_key, make SDK key required in Config (#140) * Removed the guides link * Pinning mypy and running it against different python versions (#141) * fix time zone mishandling that could make event debugging not work (#142) * fix 6.x build (#143) * fix time zone mishandling that could make event debugging not work (6.x) (#144) * prepare 6.13.3 release (#154) * Releasing version 6.13.3 * [ch99756] Add alias events (#145) * add support for experiment rollouts * fix unit test * address PR comments * use Releaser v2 config * Use newer docker images (#147) * Updates docs URLs * Add support for 3.10 (#150) * started work on FlagBuilder in as part of test data source implementation * finished FlagBuilder implementation and added FlagRuleBuilder implementation * added initial TestData interface and updated tests to not rely on test data internals * started data source implementation * changed FlagBuilder to public class; changed FlagBuilder attributes to be initialized in __init__ and eliminated use of try ... except: pass for handling empty attributes * (big segments 1) add public config/interface types * added implementation of test data source * docstring * formatting * ensure property doesn't return None * (big segments 2) implement evaluation, refactor eval logic & modules * linting * (big segments 3) implement big segment status tracking, wire up components * typing fixes * typing fixes * implement SSE contract tests * fix CI * fix CI again * fix CI * disable SSE tests in Python 3.5 * make test service port configurable * better SSE implementation that fixes linefeed and multi-byte char issues * fix constructor parameters in test service * comment * test improvements * rm obsolete default config logic * (big segments 4) implement big segment stores in Redis+DynamoDB, refactor db tests (#158) * converted ldclient.integrations module from file to directory; started moving public classes out of ldclient.impl.integrations.test_data* and instead into ldclient.integrations.test_data*; started adding TestData documentation * removed setup/teardown functions leftover from test scaffold * added TestData, FlagBuilder, and FlagRuleBuilder documentation; minor adjustments to implementation details * removed warning supression from TestData tests * fix big segments user hash algorithm to use SHA256 * update mypy version * updates to tests and related bug fixes * always cache Big Segment query result even if it's None * fix test assertion * lint * fix big segment ref format * fix big segments cache TTL being set to wrong value * fixed structure of fallthrough variation in result of FlagBuilder.build() * moved __test__ attribute into TestData class definition to prevent mypy from complaining about a missing class attribute * minor doc comment fix * Apply suggestions related to Sphinx docstring formatting from code review Co-authored-by: Eli Bishop <[email protected]> * fixed errors in the implementation of FlagBuilder's fallthrough_variation and off_variation when passing boolean variation values; updated tests to assert the expected behavior * added missing value_for_all_users() method to FlagBuilder class * Fix operator parsing errors (#169) * identify should not emit event if user key is empty (#164) * secondary should be treated as built-in attribute (#168) * URIs should have trailing slashes trimmed (#165) * all_flags_state should always include flag version (#166) * output event should not include a null prereqOf key (#167) * Account for traffic allocation on all flags (#171) * Add SDK contract tests (#170) * misc fixes to test data docs + add type hints * more type hints * remove some methods from the public test_data API * can't use "x|y" shortcut in typehints in older Pythons; use Union * fix misc type mistakes because I forgot to run the linter * update CONTRIBUTING.md and provide make targets * fixed a bug with flag rule clause builder internals; added unit test to verify rule evaluation * added ready argument to _TestDataSource class and indicated ready upon start to avoid delays in TestData initialization * Update contract tests to latest flask version (#176) Our contract tests depend on flask v1, which in turn depends on Jinja 2. Both of these are terribly dated and no longer supported. Jinja depends on markupsafe. markupsafe recently updated its code to no longer provide soft_unicode which in turn broke Jinja. Updating to the latest flask keeps all transitive dependencies better aligned and addresses this mismatch. * Adds link to Relay Proxy docs * Handle explicit None values in test payload (#179) The test harness may send explicit None values which should be treated the same as if the value was omitted entirely. * Fix "unhandled response" error in test harness (#180) When we return a `('', 204)` response from the flask handler, [Werkzeug intentionally removes the 'Content-Type' header][1], which causes the response to be created as a chunked response. The test harness is likely seeing a 204 response and isn't trying to read anything more from the stream. But since we are re-using connections, the next time it reads from the stream, it sees the `0\r\n\r\n` chunk and outputs an error: > 2022/04/20 14:23:39 Unsolicited response received on idle HTTP channel starting with "0\r\n\r\n"; err=<nil> Changing this response to 202 causes Werkzeug to return an empty response and silences the error. [1]: https://github.com/pallets/werkzeug/blob/560dd5f320bff318175f209595d42f5a80045417/src/werkzeug/wrappers/response.py#L540 * Exclude booleans when getting bucketable value (#181) When calculating a bucket, we get the bucketable value from the specified bucket by attribute. If this value is a string or an int, we can use it. Otherwise, we return None. Python considers a bool an instance of an int, which isn't what we want. So we need to add an explicit exclusion for this. * master -> main (#182) * Loosen restriction on expiringdict (#183) Originally this was pinned to a max version to deal with the incompatibility of Python 3.3 and the `typing` package. See [this PR][1]. Now that we now only support >=3.5, we can safely relax this restriction again. [1]: launchdarkly/python-server-sdk-private#120 * Fix mypy type checking (#184) A [customer requested][original-pr] that we start including a py.typed file in our repository. This would enable mypy to take advantage of our typehints. Unfortunately, this didn't completely solve the customers issue. A [second pr][second-pr] was opened to address the missing step of including the py.typed file in the `Manifest.in` file. However, this change alone is not sufficient. According to the [documentation][include_package_data], you must also include the `include_package_data=True` directive so that files specified in the `Manifest.in` file are included in distribution. [original-pr]: #166 [second-pr]: #172 [include_package_data]: https://setuptools.pypa.io/en/latest/userguide/datafiles.html#include-package-data * Add support for extra Redis connection parameters (#185) * Include wheel artifact when publishing package (#186) * remove warn-level logging done for every Big Segments query * skip tests that use a self-signed TLS cert in Python 3.7 * (U2C 1) drop EOL Python versions (#189) * drop EOL Python versions * misc cleanup, show Python version in CI * add Python 3.11 CI job * add Python 3.11 to package metadata * (U2C 2) remove alias event functionality (#187) * (U2C 3) remove inline users in events (#188) * (U2C 4) remove deprecated things (#192) * remove warn-level logging done for every Big Segments query (#190) * remove warn-level logging done for every Big Segments query * skip tests that use a self-signed TLS cert in Python 3.7 * implement context model * fix exports * specify exports * add copy constructor * minimal changes for SDK methods & evaluator to accept Context * update tests, add subscript method * lint * in type hints, must use Dict[A, B] rather than dict[A, B] for Python <3.9 * support context kind in clauses + enable v2 contract tests * misc fixes * misc fixes * support contextTargets * support contextKind in rollouts/experiments * support includedContexts/excludedContexts in segment * comment copyedit Co-authored-by: Matthew M. Keeler <[email protected]> * comment fixes * rm unused Co-authored-by: Matthew M. Keeler <[email protected]> * fix create_multi to support flattening * lint * use custom classes for flag/segment data model * use store adapter for safety * misc cleanup * misc fixes for persistent stores * more database store fixes * support attribute reference lookups in evaluations * pass logger from client * context kind logic for big segments + enable big segment contract tests * formatting fixes + test cleanup * prerequisite cycle detection * segment recursion * define custom classes for event data * add module init file * linting * fix prereq stack logic * (U2C 17) U2C changes for events, not including private attributes (#205) * private attribute redaction * move a lot of code out of top-level modules * TestData changes for contexts * general doc comment fixes for 8.0 * U2C configuration updates * update release metadata * store flag/segment target lists as sets * fix type hint * preprocess clause values for time/regex/semver operators * fix type checking for matches operator Co-authored-by: Eli Bishop <[email protected]> Co-authored-by: LaunchDarklyCI <[email protected]> Co-authored-by: Elliot <[email protected]> Co-authored-by: Gabor Angeli <[email protected]> Co-authored-by: Elliot <[email protected]> Co-authored-by: Ben Woskow <[email protected]> Co-authored-by: LaunchDarklyCI <[email protected]> Co-authored-by: hroederld <[email protected]> Co-authored-by: Robert J. Neal <[email protected]> Co-authored-by: Robert J. Neal <[email protected]> Co-authored-by: Ember Stevens <[email protected]> Co-authored-by: ember-stevens <[email protected]> Co-authored-by: Matthew M. Keeler <[email protected]> Co-authored-by: charukiewicz <[email protected]> Co-authored-by: LaunchDarklyReleaseBot <[email protected]> Co-authored-by: Christian Charukiewicz <[email protected]> Co-authored-by: Matthew M. Keeler <[email protected]>
* implement our own retry logic & logging for event posts, don't use urllib3.Retry (#133) * remove support for indirect/patch and indirect/put * remove unused logic for individual flag/segment poll for indirect/patch * Ehaisley/84082/remove python2 (#136) * remove all references to six and remove queue fallback imports * remove NullHandler logger backwards compat * update circleci config to remove python 2.7 tests * remove ordereddict backwards compat * update setup.py to no longer list python 2.7 as compatible * no longer inherit from object for python 2 backwards compat * update readme and manifest to reflect python 2.7 removal * remove unicode type compatibility * remove 2.7 support from circleci * Allow authenticating with proxy This commit allows for authenticating with a proxy configured with the `http_proxy` environment variable. Authentication requires passing a header, and is not parsed by urllib3 from the proxy_url. * reimplement proxy tests for DRY and add test of proxy auth params * doc comment on auth params in proxy URL * add type hints to some of the public facing api. update some docs * Revert "add type hints to some of the public facing api." This reverts commit c35fa61. * Ehaisley/ch86857/type hints (#138) * add typehints to the public API * validate typehints in the public api and tests with mypy * remove all current deprecations (#139) * remove all currently deprecated classes, methods, arguments, and tests * also update semver usage to remove calls to deprecated functions and classes * remove global set_sdk_key, make SDK key required in Config (#140) * Removed the guides link * Pinning mypy and running it against different python versions (#141) * fix time zone mishandling that could make event debugging not work (#142) * fix 6.x build (#143) * fix time zone mishandling that could make event debugging not work (6.x) (#144) * prepare 6.13.3 release (#154) * Releasing version 6.13.3 * [ch99756] Add alias events (#145) * add support for experiment rollouts * fix unit test * address PR comments * use Releaser v2 config * Use newer docker images (#147) * Updates docs URLs * Add support for 3.10 (#150) * started work on FlagBuilder in as part of test data source implementation * finished FlagBuilder implementation and added FlagRuleBuilder implementation * added initial TestData interface and updated tests to not rely on test data internals * started data source implementation * changed FlagBuilder to public class; changed FlagBuilder attributes to be initialized in __init__ and eliminated use of try ... except: pass for handling empty attributes * (big segments 1) add public config/interface types * added implementation of test data source * docstring * formatting * ensure property doesn't return None * (big segments 2) implement evaluation, refactor eval logic & modules * linting * (big segments 3) implement big segment status tracking, wire up components * typing fixes * typing fixes * implement SSE contract tests * fix CI * fix CI again * fix CI * disable SSE tests in Python 3.5 * make test service port configurable * better SSE implementation that fixes linefeed and multi-byte char issues * fix constructor parameters in test service * comment * test improvements * rm obsolete default config logic * (big segments 4) implement big segment stores in Redis+DynamoDB, refactor db tests (#158) * converted ldclient.integrations module from file to directory; started moving public classes out of ldclient.impl.integrations.test_data* and instead into ldclient.integrations.test_data*; started adding TestData documentation * removed setup/teardown functions leftover from test scaffold * added TestData, FlagBuilder, and FlagRuleBuilder documentation; minor adjustments to implementation details * removed warning supression from TestData tests * fix big segments user hash algorithm to use SHA256 * update mypy version * updates to tests and related bug fixes * always cache Big Segment query result even if it's None * fix test assertion * lint * fix big segment ref format * fix big segments cache TTL being set to wrong value * fixed structure of fallthrough variation in result of FlagBuilder.build() * moved __test__ attribute into TestData class definition to prevent mypy from complaining about a missing class attribute * minor doc comment fix * Apply suggestions related to Sphinx docstring formatting from code review Co-authored-by: Eli Bishop <[email protected]> * fixed errors in the implementation of FlagBuilder's fallthrough_variation and off_variation when passing boolean variation values; updated tests to assert the expected behavior * added missing value_for_all_users() method to FlagBuilder class * Fix operator parsing errors (#169) * identify should not emit event if user key is empty (#164) * secondary should be treated as built-in attribute (#168) * URIs should have trailing slashes trimmed (#165) * all_flags_state should always include flag version (#166) * output event should not include a null prereqOf key (#167) * Account for traffic allocation on all flags (#171) * Add SDK contract tests (#170) * misc fixes to test data docs + add type hints * more type hints * remove some methods from the public test_data API * can't use "x|y" shortcut in typehints in older Pythons; use Union * fix misc type mistakes because I forgot to run the linter * update CONTRIBUTING.md and provide make targets * fixed a bug with flag rule clause builder internals; added unit test to verify rule evaluation * added ready argument to _TestDataSource class and indicated ready upon start to avoid delays in TestData initialization * Update contract tests to latest flask version (#176) Our contract tests depend on flask v1, which in turn depends on Jinja 2. Both of these are terribly dated and no longer supported. Jinja depends on markupsafe. markupsafe recently updated its code to no longer provide soft_unicode which in turn broke Jinja. Updating to the latest flask keeps all transitive dependencies better aligned and addresses this mismatch. * Adds link to Relay Proxy docs * Handle explicit None values in test payload (#179) The test harness may send explicit None values which should be treated the same as if the value was omitted entirely. * Fix "unhandled response" error in test harness (#180) When we return a `('', 204)` response from the flask handler, [Werkzeug intentionally removes the 'Content-Type' header][1], which causes the response to be created as a chunked response. The test harness is likely seeing a 204 response and isn't trying to read anything more from the stream. But since we are re-using connections, the next time it reads from the stream, it sees the `0\r\n\r\n` chunk and outputs an error: > 2022/04/20 14:23:39 Unsolicited response received on idle HTTP channel starting with "0\r\n\r\n"; err=<nil> Changing this response to 202 causes Werkzeug to return an empty response and silences the error. [1]: https://github.com/pallets/werkzeug/blob/560dd5f320bff318175f209595d42f5a80045417/src/werkzeug/wrappers/response.py#L540 * Exclude booleans when getting bucketable value (#181) When calculating a bucket, we get the bucketable value from the specified bucket by attribute. If this value is a string or an int, we can use it. Otherwise, we return None. Python considers a bool an instance of an int, which isn't what we want. So we need to add an explicit exclusion for this. * master -> main (#182) * Loosen restriction on expiringdict (#183) Originally this was pinned to a max version to deal with the incompatibility of Python 3.3 and the `typing` package. See [this PR][1]. Now that we now only support >=3.5, we can safely relax this restriction again. [1]: launchdarkly/python-server-sdk-private#120 * Fix mypy type checking (#184) A [customer requested][original-pr] that we start including a py.typed file in our repository. This would enable mypy to take advantage of our typehints. Unfortunately, this didn't completely solve the customers issue. A [second pr][second-pr] was opened to address the missing step of including the py.typed file in the `Manifest.in` file. However, this change alone is not sufficient. According to the [documentation][include_package_data], you must also include the `include_package_data=True` directive so that files specified in the `Manifest.in` file are included in distribution. [original-pr]: #166 [second-pr]: #172 [include_package_data]: https://setuptools.pypa.io/en/latest/userguide/datafiles.html#include-package-data * Add support for extra Redis connection parameters (#185) * Include wheel artifact when publishing package (#186) * remove warn-level logging done for every Big Segments query * skip tests that use a self-signed TLS cert in Python 3.7 * (U2C 1) drop EOL Python versions (#189) * drop EOL Python versions * misc cleanup, show Python version in CI * add Python 3.11 CI job * add Python 3.11 to package metadata * (U2C 2) remove alias event functionality (#187) * (U2C 3) remove inline users in events (#188) * (U2C 4) remove deprecated things (#192) * remove warn-level logging done for every Big Segments query (#190) * remove warn-level logging done for every Big Segments query * skip tests that use a self-signed TLS cert in Python 3.7 * implement context model * fix exports * specify exports * add copy constructor * minimal changes for SDK methods & evaluator to accept Context * update tests, add subscript method * lint * in type hints, must use Dict[A, B] rather than dict[A, B] for Python <3.9 * support context kind in clauses + enable v2 contract tests * misc fixes * misc fixes * support contextTargets * support contextKind in rollouts/experiments * support includedContexts/excludedContexts in segment * comment copyedit Co-authored-by: Matthew M. Keeler <[email protected]> * comment fixes * rm unused Co-authored-by: Matthew M. Keeler <[email protected]> * fix create_multi to support flattening * lint * use custom classes for flag/segment data model * use store adapter for safety * misc cleanup * misc fixes for persistent stores * more database store fixes * support attribute reference lookups in evaluations * pass logger from client * context kind logic for big segments + enable big segment contract tests * formatting fixes + test cleanup * prerequisite cycle detection * segment recursion * define custom classes for event data * add module init file * linting * fix prereq stack logic * (U2C 17) U2C changes for events, not including private attributes (#205) * private attribute redaction * move a lot of code out of top-level modules * TestData changes for contexts * general doc comment fixes for 8.0 * U2C configuration updates * update release metadata * store flag/segment target lists as sets * fix type hint * preprocess clause values for time/regex/semver operators * fix type checking for matches operator * Add application info support (#214) * Add application info support (#214) (#215) * Upgrade pip to fix failing CI build (#216) The CI build was failing because pip had an outdated list of available wheels for installation. Since it couldn't find a match, it was trying to build a package from source, which requires the rust compiler, which in turn isn't present on some of the docker images. By updating pip we get the updated list of available wheels, thereby allowing us to bypass source building and the need for the rust compiler entirely. * prepare 7.6.0 release (#192) * comment * add end-to-end unit tests for proxy config * indents * add 3.8 build * image name * fail on SyntaxWarning * typo * command syntax * pin expiringdict dependency for Python 3.3 compatibility * add Windows CircleCI job * periods are no longer valid in CircleCI job names * syntax fix * install Python in Windows * set path * move command * turn off debug logging * Py3 in Windows * config param * rm redundant step * choco switch * refactor Linux jobs using CircleCI 2.1 features * set log level before anything else * rm Azure config * use yaml.safe_load() to avoid code execution vulnerability in file data source * Initial work on wrapper_name, wrapper_version, diagnostic config options and start of diagnostic config event creation. * Python 2 compat changes. * More event generation code and starting to integrate tracking diagnostic values. * Add minimum diagnostic recording interval. Fix diagnostic.py to be importable. Add more diagnostic event fields. * don't let user fall outside of last bucket in rollout * fixing conditional logic * Add docstrings for diagnostic configuration options. * fix off-by-1 error * avoid redundant dict lookups * add unit tests for basic bucketing logic and edge case * Stream init tracking. Feeding of accumulator object through SDK. Various fixes. * Track events in last batch. * Fix sdk version field, some stylistic improvements. * Last of diagnostic configuration object fields. * Fill out rest of platform fields. * Cleanup and failed stream initialization tracking. * Add diagnostic config option test. * Add tests for diagnostics.py * Testing rest of diagnostic fields. * Test that streaming update processor records successful and unsuccessful connection attempts in the diagnostic accumulator when available. * Improvements to testability of event processor. * Rest of event processor tests. * Remove janky reflection. * Test change to filesource optional test requirements. * [ch61092] Add event payload ID on event requests. * normalize data store type and OS name in diagnostic events * gitignore * copyedit to diagnostic event config property comment * fix spurious error after sending diagnostic event * make verify_ssl=False turn off certificate verification too (#129) * add more TLS config options and collect HTTP/HTTPS config options in a class (#130) * make stream retry/backoff/jitter behavior consistent with other SDKs + improve testing (#131) * streams shouldn't use the same read timeout as the rest of the SDK (#132) * implement our own retry logic & logging for event posts, don't use urllib3.Retry (#133) * remove support for indirect/patch and indirect/put * remove unused logic for individual flag/segment poll for indirect/patch * Ehaisley/84082/remove python2 (#136) * remove all references to six and remove queue fallback imports * remove NullHandler logger backwards compat * update circleci config to remove python 2.7 tests * remove ordereddict backwards compat * update setup.py to no longer list python 2.7 as compatible * no longer inherit from object for python 2 backwards compat * update readme and manifest to reflect python 2.7 removal * remove unicode type compatibility * remove 2.7 support from circleci * Allow authenticating with proxy This commit allows for authenticating with a proxy configured with the `http_proxy` environment variable. Authentication requires passing a header, and is not parsed by urllib3 from the proxy_url. * reimplement proxy tests for DRY and add test of proxy auth params * doc comment on auth params in proxy URL * add type hints to some of the public facing api. update some docs * Revert "add type hints to some of the public facing api." This reverts commit c35fa61. * Ehaisley/ch86857/type hints (#138) * add typehints to the public API * validate typehints in the public api and tests with mypy * remove all current deprecations (#139) * remove all currently deprecated classes, methods, arguments, and tests * also update semver usage to remove calls to deprecated functions and classes * remove global set_sdk_key, make SDK key required in Config (#140) * Removed the guides link * Pinning mypy and running it against different python versions (#141) * fix time zone mishandling that could make event debugging not work (#142) * fix 6.x build (#143) * fix time zone mishandling that could make event debugging not work (6.x) (#144) * prepare 6.13.3 release (#154) * Releasing version 6.13.3 * [ch99756] Add alias events (#145) * add support for experiment rollouts * fix unit test * address PR comments * use Releaser v2 config * Use newer docker images (#147) * Updates docs URLs * Add support for 3.10 (#150) * started work on FlagBuilder in as part of test data source implementation * finished FlagBuilder implementation and added FlagRuleBuilder implementation * added initial TestData interface and updated tests to not rely on test data internals * started data source implementation * changed FlagBuilder to public class; changed FlagBuilder attributes to be initialized in __init__ and eliminated use of try ... except: pass for handling empty attributes * (big segments 1) add public config/interface types * added implementation of test data source * docstring * formatting * ensure property doesn't return None * (big segments 2) implement evaluation, refactor eval logic & modules * linting * (big segments 3) implement big segment status tracking, wire up components * typing fixes * typing fixes * implement SSE contract tests * fix CI * fix CI again * fix CI * disable SSE tests in Python 3.5 * make test service port configurable * better SSE implementation that fixes linefeed and multi-byte char issues * fix constructor parameters in test service * comment * test improvements * rm obsolete default config logic * (big segments 4) implement big segment stores in Redis+DynamoDB, refactor db tests (#158) * converted ldclient.integrations module from file to directory; started moving public classes out of ldclient.impl.integrations.test_data* and instead into ldclient.integrations.test_data*; started adding TestData documentation * removed setup/teardown functions leftover from test scaffold * added TestData, FlagBuilder, and FlagRuleBuilder documentation; minor adjustments to implementation details * removed warning supression from TestData tests * fix big segments user hash algorithm to use SHA256 * update mypy version * updates to tests and related bug fixes * always cache Big Segment query result even if it's None * fix test assertion * lint * fix big segment ref format * fix big segments cache TTL being set to wrong value * fixed structure of fallthrough variation in result of FlagBuilder.build() * moved __test__ attribute into TestData class definition to prevent mypy from complaining about a missing class attribute * minor doc comment fix * Apply suggestions related to Sphinx docstring formatting from code review Co-authored-by: Eli Bishop <[email protected]> * fixed errors in the implementation of FlagBuilder's fallthrough_variation and off_variation when passing boolean variation values; updated tests to assert the expected behavior * added missing value_for_all_users() method to FlagBuilder class * Fix operator parsing errors (#169) * identify should not emit event if user key is empty (#164) * secondary should be treated as built-in attribute (#168) * URIs should have trailing slashes trimmed (#165) * all_flags_state should always include flag version (#166) * output event should not include a null prereqOf key (#167) * Account for traffic allocation on all flags (#171) * Add SDK contract tests (#170) * misc fixes to test data docs + add type hints * more type hints * remove some methods from the public test_data API * can't use "x|y" shortcut in typehints in older Pythons; use Union * fix misc type mistakes because I forgot to run the linter * update CONTRIBUTING.md and provide make targets * fixed a bug with flag rule clause builder internals; added unit test to verify rule evaluation * added ready argument to _TestDataSource class and indicated ready upon start to avoid delays in TestData initialization * Update contract tests to latest flask version (#176) Our contract tests depend on flask v1, which in turn depends on Jinja 2. Both of these are terribly dated and no longer supported. Jinja depends on markupsafe. markupsafe recently updated its code to no longer provide soft_unicode which in turn broke Jinja. Updating to the latest flask keeps all transitive dependencies better aligned and addresses this mismatch. * Adds link to Relay Proxy docs * Handle explicit None values in test payload (#179) The test harness may send explicit None values which should be treated the same as if the value was omitted entirely. * Fix "unhandled response" error in test harness (#180) When we return a `('', 204)` response from the flask handler, [Werkzeug intentionally removes the 'Content-Type' header][1], which causes the response to be created as a chunked response. The test harness is likely seeing a 204 response and isn't trying to read anything more from the stream. But since we are re-using connections, the next time it reads from the stream, it sees the `0\r\n\r\n` chunk and outputs an error: > 2022/04/20 14:23:39 Unsolicited response received on idle HTTP channel starting with "0\r\n\r\n"; err=<nil> Changing this response to 202 causes Werkzeug to return an empty response and silences the error. [1]: https://github.com/pallets/werkzeug/blob/560dd5f320bff318175f209595d42f5a80045417/src/werkzeug/wrappers/response.py#L540 * Exclude booleans when getting bucketable value (#181) When calculating a bucket, we get the bucketable value from the specified bucket by attribute. If this value is a string or an int, we can use it. Otherwise, we return None. Python considers a bool an instance of an int, which isn't what we want. So we need to add an explicit exclusion for this. * master -> main (#182) * Loosen restriction on expiringdict (#183) Originally this was pinned to a max version to deal with the incompatibility of Python 3.3 and the `typing` package. See [this PR][1]. Now that we now only support >=3.5, we can safely relax this restriction again. [1]: launchdarkly/python-server-sdk-private#120 * Fix mypy type checking (#184) A [customer requested][original-pr] that we start including a py.typed file in our repository. This would enable mypy to take advantage of our typehints. Unfortunately, this didn't completely solve the customers issue. A [second pr][second-pr] was opened to address the missing step of including the py.typed file in the `Manifest.in` file. However, this change alone is not sufficient. According to the [documentation][include_package_data], you must also include the `include_package_data=True` directive so that files specified in the `Manifest.in` file are included in distribution. [original-pr]: #166 [second-pr]: #172 [include_package_data]: https://setuptools.pypa.io/en/latest/userguide/datafiles.html#include-package-data * Add support for extra Redis connection parameters (#185) * Include wheel artifact when publishing package (#186) * skip tests that use a self-signed TLS cert in Python 3.7 * remove warn-level logging done for every Big Segments query (#190) * remove warn-level logging done for every Big Segments query * skip tests that use a self-signed TLS cert in Python 3.7 * update release metadata * Add application info support (#214) * Upgrade pip to fix failing CI build (#216) The CI build was failing because pip had an outdated list of available wheels for installation. Since it couldn't find a match, it was trying to build a package from source, which requires the rust compiler, which in turn isn't present on some of the docker images. By updating pip we get the updated list of available wheels, thereby allowing us to bypass source building and the need for the rust compiler entirely. --------- Co-authored-by: Eli Bishop <[email protected]> Co-authored-by: LaunchDarklyCI <[email protected]> Co-authored-by: Ben Woskow <[email protected]> Co-authored-by: Gavin Whelan <[email protected]> Co-authored-by: Elliot <[email protected]> Co-authored-by: Gabor Angeli <[email protected]> Co-authored-by: Elliot <[email protected]> Co-authored-by: Ben Woskow <[email protected]> Co-authored-by: LaunchDarklyCI <[email protected]> Co-authored-by: hroederld <[email protected]> Co-authored-by: Robert J. Neal <[email protected]> Co-authored-by: Robert J. Neal <[email protected]> Co-authored-by: Ember Stevens <[email protected]> Co-authored-by: ember-stevens <[email protected]> Co-authored-by: Matthew M. Keeler <[email protected]> Co-authored-by: charukiewicz <[email protected]> Co-authored-by: LaunchDarklyReleaseBot <[email protected]> Co-authored-by: Christian Charukiewicz <[email protected]> Co-authored-by: Matthew M. Keeler <[email protected]> * Releasing version 7.6.0 --------- Co-authored-by: Eli Bishop <[email protected]> Co-authored-by: LaunchDarklyCI <[email protected]> Co-authored-by: Elliot <[email protected]> Co-authored-by: Gabor Angeli <[email protected]> Co-authored-by: Elliot <[email protected]> Co-authored-by: Ben Woskow <[email protected]> Co-authored-by: LaunchDarklyCI <[email protected]> Co-authored-by: hroederld <[email protected]> Co-authored-by: Robert J. Neal <[email protected]> Co-authored-by: Robert J. Neal <[email protected]> Co-authored-by: Ember Stevens <[email protected]> Co-authored-by: ember-stevens <[email protected]> Co-authored-by: Matthew M. Keeler <[email protected]> Co-authored-by: charukiewicz <[email protected]> Co-authored-by: LaunchDarklyReleaseBot <[email protected]> Co-authored-by: Christian Charukiewicz <[email protected]> Co-authored-by: Matthew M. Keeler <[email protected]> Co-authored-by: Ben Woskow <[email protected]> Co-authored-by: Gavin Whelan <[email protected]>
* reimplement proxy tests for DRY and add test of proxy auth params * doc comment on auth params in proxy URL * add type hints to some of the public facing api. update some docs * Revert "add type hints to some of the public facing api." This reverts commit c35fa61. * Ehaisley/ch86857/type hints (#138) * add typehints to the public API * validate typehints in the public api and tests with mypy * remove all current deprecations (#139) * remove all currently deprecated classes, methods, arguments, and tests * also update semver usage to remove calls to deprecated functions and classes * remove global set_sdk_key, make SDK key required in Config (#140) * Removed the guides link * Pinning mypy and running it against different python versions (#141) * fix time zone mishandling that could make event debugging not work (#142) * fix 6.x build (#143) * fix time zone mishandling that could make event debugging not work (6.x) (#144) * prepare 6.13.3 release (#154) * Releasing version 6.13.3 * [ch99756] Add alias events (#145) * add support for experiment rollouts * fix unit test * address PR comments * use Releaser v2 config * Use newer docker images (#147) * Updates docs URLs * Add support for 3.10 (#150) * started work on FlagBuilder in as part of test data source implementation * finished FlagBuilder implementation and added FlagRuleBuilder implementation * added initial TestData interface and updated tests to not rely on test data internals * started data source implementation * changed FlagBuilder to public class; changed FlagBuilder attributes to be initialized in __init__ and eliminated use of try ... except: pass for handling empty attributes * (big segments 1) add public config/interface types * added implementation of test data source * docstring * formatting * ensure property doesn't return None * (big segments 2) implement evaluation, refactor eval logic & modules * linting * (big segments 3) implement big segment status tracking, wire up components * typing fixes * typing fixes * implement SSE contract tests * fix CI * fix CI again * fix CI * disable SSE tests in Python 3.5 * make test service port configurable * better SSE implementation that fixes linefeed and multi-byte char issues * fix constructor parameters in test service * comment * test improvements * rm obsolete default config logic * (big segments 4) implement big segment stores in Redis+DynamoDB, refactor db tests (#158) * converted ldclient.integrations module from file to directory; started moving public classes out of ldclient.impl.integrations.test_data* and instead into ldclient.integrations.test_data*; started adding TestData documentation * removed setup/teardown functions leftover from test scaffold * added TestData, FlagBuilder, and FlagRuleBuilder documentation; minor adjustments to implementation details * removed warning supression from TestData tests * fix big segments user hash algorithm to use SHA256 * update mypy version * updates to tests and related bug fixes * always cache Big Segment query result even if it's None * fix test assertion * lint * fix big segment ref format * fix big segments cache TTL being set to wrong value * fixed structure of fallthrough variation in result of FlagBuilder.build() * moved __test__ attribute into TestData class definition to prevent mypy from complaining about a missing class attribute * minor doc comment fix * Apply suggestions related to Sphinx docstring formatting from code review Co-authored-by: Eli Bishop <[email protected]> * fixed errors in the implementation of FlagBuilder's fallthrough_variation and off_variation when passing boolean variation values; updated tests to assert the expected behavior * added missing value_for_all_users() method to FlagBuilder class * Fix operator parsing errors (#169) * identify should not emit event if user key is empty (#164) * secondary should be treated as built-in attribute (#168) * URIs should have trailing slashes trimmed (#165) * all_flags_state should always include flag version (#166) * output event should not include a null prereqOf key (#167) * Account for traffic allocation on all flags (#171) * Add SDK contract tests (#170) * misc fixes to test data docs + add type hints * more type hints * remove some methods from the public test_data API * can't use "x|y" shortcut in typehints in older Pythons; use Union * fix misc type mistakes because I forgot to run the linter * update CONTRIBUTING.md and provide make targets * fixed a bug with flag rule clause builder internals; added unit test to verify rule evaluation * added ready argument to _TestDataSource class and indicated ready upon start to avoid delays in TestData initialization * Update contract tests to latest flask version (#176) Our contract tests depend on flask v1, which in turn depends on Jinja 2. Both of these are terribly dated and no longer supported. Jinja depends on markupsafe. markupsafe recently updated its code to no longer provide soft_unicode which in turn broke Jinja. Updating to the latest flask keeps all transitive dependencies better aligned and addresses this mismatch. * Adds link to Relay Proxy docs * Handle explicit None values in test payload (#179) The test harness may send explicit None values which should be treated the same as if the value was omitted entirely. * Fix "unhandled response" error in test harness (#180) When we return a `('', 204)` response from the flask handler, [Werkzeug intentionally removes the 'Content-Type' header][1], which causes the response to be created as a chunked response. The test harness is likely seeing a 204 response and isn't trying to read anything more from the stream. But since we are re-using connections, the next time it reads from the stream, it sees the `0\r\n\r\n` chunk and outputs an error: > 2022/04/20 14:23:39 Unsolicited response received on idle HTTP channel starting with "0\r\n\r\n"; err=<nil> Changing this response to 202 causes Werkzeug to return an empty response and silences the error. [1]: https://github.com/pallets/werkzeug/blob/560dd5f320bff318175f209595d42f5a80045417/src/werkzeug/wrappers/response.py#L540 * Exclude booleans when getting bucketable value (#181) When calculating a bucket, we get the bucketable value from the specified bucket by attribute. If this value is a string or an int, we can use it. Otherwise, we return None. Python considers a bool an instance of an int, which isn't what we want. So we need to add an explicit exclusion for this. * master -> main (#182) * Loosen restriction on expiringdict (#183) Originally this was pinned to a max version to deal with the incompatibility of Python 3.3 and the `typing` package. See [this PR][1]. Now that we now only support >=3.5, we can safely relax this restriction again. [1]: launchdarkly/python-server-sdk-private#120 * Fix mypy type checking (#184) A [customer requested][original-pr] that we start including a py.typed file in our repository. This would enable mypy to take advantage of our typehints. Unfortunately, this didn't completely solve the customers issue. A [second pr][second-pr] was opened to address the missing step of including the py.typed file in the `Manifest.in` file. However, this change alone is not sufficient. According to the [documentation][include_package_data], you must also include the `include_package_data=True` directive so that files specified in the `Manifest.in` file are included in distribution. [original-pr]: #166 [second-pr]: #172 [include_package_data]: https://setuptools.pypa.io/en/latest/userguide/datafiles.html#include-package-data * Add support for extra Redis connection parameters (#185) * Include wheel artifact when publishing package (#186) * remove warn-level logging done for every Big Segments query * skip tests that use a self-signed TLS cert in Python 3.7 * (U2C 1) drop EOL Python versions (#189) * drop EOL Python versions * misc cleanup, show Python version in CI * add Python 3.11 CI job * add Python 3.11 to package metadata * (U2C 2) remove alias event functionality (#187) * (U2C 3) remove inline users in events (#188) * (U2C 4) remove deprecated things (#192) * remove warn-level logging done for every Big Segments query (#190) * remove warn-level logging done for every Big Segments query * skip tests that use a self-signed TLS cert in Python 3.7 * implement context model * fix exports * specify exports * add copy constructor * minimal changes for SDK methods & evaluator to accept Context * update tests, add subscript method * lint * in type hints, must use Dict[A, B] rather than dict[A, B] for Python <3.9 * support context kind in clauses + enable v2 contract tests * misc fixes * misc fixes * support contextTargets * support contextKind in rollouts/experiments * support includedContexts/excludedContexts in segment * comment copyedit Co-authored-by: Matthew M. Keeler <[email protected]> * comment fixes * rm unused Co-authored-by: Matthew M. Keeler <[email protected]> * fix create_multi to support flattening * lint * use custom classes for flag/segment data model * use store adapter for safety * misc cleanup * misc fixes for persistent stores * more database store fixes * support attribute reference lookups in evaluations * pass logger from client * context kind logic for big segments + enable big segment contract tests * formatting fixes + test cleanup * prerequisite cycle detection * segment recursion * define custom classes for event data * add module init file * linting * fix prereq stack logic * (U2C 17) U2C changes for events, not including private attributes (#205) * private attribute redaction * move a lot of code out of top-level modules * TestData changes for contexts * general doc comment fixes for 8.0 * U2C configuration updates * update release metadata * store flag/segment target lists as sets * fix type hint * preprocess clause values for time/regex/semver operators * fix type checking for matches operator * Add application info support (#214) * Add application info support (#214) (#215) * Upgrade pip to fix failing CI build (#216) The CI build was failing because pip had an outdated list of available wheels for installation. Since it couldn't find a match, it was trying to build a package from source, which requires the rust compiler, which in turn isn't present on some of the docker images. By updating pip we get the updated list of available wheels, thereby allowing us to bypass source building and the need for the rust compiler entirely. * prepare 7.6.0 release (#192) * comment * add end-to-end unit tests for proxy config * indents * add 3.8 build * image name * fail on SyntaxWarning * typo * command syntax * pin expiringdict dependency for Python 3.3 compatibility * add Windows CircleCI job * periods are no longer valid in CircleCI job names * syntax fix * install Python in Windows * set path * move command * turn off debug logging * Py3 in Windows * config param * rm redundant step * choco switch * refactor Linux jobs using CircleCI 2.1 features * set log level before anything else * rm Azure config * use yaml.safe_load() to avoid code execution vulnerability in file data source * Initial work on wrapper_name, wrapper_version, diagnostic config options and start of diagnostic config event creation. * Python 2 compat changes. * More event generation code and starting to integrate tracking diagnostic values. * Add minimum diagnostic recording interval. Fix diagnostic.py to be importable. Add more diagnostic event fields. * don't let user fall outside of last bucket in rollout * fixing conditional logic * Add docstrings for diagnostic configuration options. * fix off-by-1 error * avoid redundant dict lookups * add unit tests for basic bucketing logic and edge case * Stream init tracking. Feeding of accumulator object through SDK. Various fixes. * Track events in last batch. * Fix sdk version field, some stylistic improvements. * Last of diagnostic configuration object fields. * Fill out rest of platform fields. * Cleanup and failed stream initialization tracking. * Add diagnostic config option test. * Add tests for diagnostics.py * Testing rest of diagnostic fields. * Test that streaming update processor records successful and unsuccessful connection attempts in the diagnostic accumulator when available. * Improvements to testability of event processor. * Rest of event processor tests. * Remove janky reflection. * Test change to filesource optional test requirements. * [ch61092] Add event payload ID on event requests. * normalize data store type and OS name in diagnostic events * gitignore * copyedit to diagnostic event config property comment * fix spurious error after sending diagnostic event * make verify_ssl=False turn off certificate verification too (#129) * add more TLS config options and collect HTTP/HTTPS config options in a class (#130) * make stream retry/backoff/jitter behavior consistent with other SDKs + improve testing (#131) * streams shouldn't use the same read timeout as the rest of the SDK (#132) * implement our own retry logic & logging for event posts, don't use urllib3.Retry (#133) * remove support for indirect/patch and indirect/put * remove unused logic for individual flag/segment poll for indirect/patch * Ehaisley/84082/remove python2 (#136) * remove all references to six and remove queue fallback imports * remove NullHandler logger backwards compat * update circleci config to remove python 2.7 tests * remove ordereddict backwards compat * update setup.py to no longer list python 2.7 as compatible * no longer inherit from object for python 2 backwards compat * update readme and manifest to reflect python 2.7 removal * remove unicode type compatibility * remove 2.7 support from circleci * Allow authenticating with proxy This commit allows for authenticating with a proxy configured with the `http_proxy` environment variable. Authentication requires passing a header, and is not parsed by urllib3 from the proxy_url. * reimplement proxy tests for DRY and add test of proxy auth params * doc comment on auth params in proxy URL * add type hints to some of the public facing api. update some docs * Revert "add type hints to some of the public facing api." This reverts commit c35fa61. * Ehaisley/ch86857/type hints (#138) * add typehints to the public API * validate typehints in the public api and tests with mypy * remove all current deprecations (#139) * remove all currently deprecated classes, methods, arguments, and tests * also update semver usage to remove calls to deprecated functions and classes * remove global set_sdk_key, make SDK key required in Config (#140) * Removed the guides link * Pinning mypy and running it against different python versions (#141) * fix time zone mishandling that could make event debugging not work (#142) * fix 6.x build (#143) * fix time zone mishandling that could make event debugging not work (6.x) (#144) * prepare 6.13.3 release (#154) * Releasing version 6.13.3 * [ch99756] Add alias events (#145) * add support for experiment rollouts * fix unit test * address PR comments * use Releaser v2 config * Use newer docker images (#147) * Updates docs URLs * Add support for 3.10 (#150) * started work on FlagBuilder in as part of test data source implementation * finished FlagBuilder implementation and added FlagRuleBuilder implementation * added initial TestData interface and updated tests to not rely on test data internals * started data source implementation * changed FlagBuilder to public class; changed FlagBuilder attributes to be initialized in __init__ and eliminated use of try ... except: pass for handling empty attributes * (big segments 1) add public config/interface types * added implementation of test data source * docstring * formatting * ensure property doesn't return None * (big segments 2) implement evaluation, refactor eval logic & modules * linting * (big segments 3) implement big segment status tracking, wire up components * typing fixes * typing fixes * implement SSE contract tests * fix CI * fix CI again * fix CI * disable SSE tests in Python 3.5 * make test service port configurable * better SSE implementation that fixes linefeed and multi-byte char issues * fix constructor parameters in test service * comment * test improvements * rm obsolete default config logic * (big segments 4) implement big segment stores in Redis+DynamoDB, refactor db tests (#158) * converted ldclient.integrations module from file to directory; started moving public classes out of ldclient.impl.integrations.test_data* and instead into ldclient.integrations.test_data*; started adding TestData documentation * removed setup/teardown functions leftover from test scaffold * added TestData, FlagBuilder, and FlagRuleBuilder documentation; minor adjustments to implementation details * removed warning supression from TestData tests * fix big segments user hash algorithm to use SHA256 * update mypy version * updates to tests and related bug fixes * always cache Big Segment query result even if it's None * fix test assertion * lint * fix big segment ref format * fix big segments cache TTL being set to wrong value * fixed structure of fallthrough variation in result of FlagBuilder.build() * moved __test__ attribute into TestData class definition to prevent mypy from complaining about a missing class attribute * minor doc comment fix * Apply suggestions related to Sphinx docstring formatting from code review Co-authored-by: Eli Bishop <[email protected]> * fixed errors in the implementation of FlagBuilder's fallthrough_variation and off_variation when passing boolean variation values; updated tests to assert the expected behavior * added missing value_for_all_users() method to FlagBuilder class * Fix operator parsing errors (#169) * identify should not emit event if user key is empty (#164) * secondary should be treated as built-in attribute (#168) * URIs should have trailing slashes trimmed (#165) * all_flags_state should always include flag version (#166) * output event should not include a null prereqOf key (#167) * Account for traffic allocation on all flags (#171) * Add SDK contract tests (#170) * misc fixes to test data docs + add type hints * more type hints * remove some methods from the public test_data API * can't use "x|y" shortcut in typehints in older Pythons; use Union * fix misc type mistakes because I forgot to run the linter * update CONTRIBUTING.md and provide make targets * fixed a bug with flag rule clause builder internals; added unit test to verify rule evaluation * added ready argument to _TestDataSource class and indicated ready upon start to avoid delays in TestData initialization * Update contract tests to latest flask version (#176) Our contract tests depend on flask v1, which in turn depends on Jinja 2. Both of these are terribly dated and no longer supported. Jinja depends on markupsafe. markupsafe recently updated its code to no longer provide soft_unicode which in turn broke Jinja. Updating to the latest flask keeps all transitive dependencies better aligned and addresses this mismatch. * Adds link to Relay Proxy docs * Handle explicit None values in test payload (#179) The test harness may send explicit None values which should be treated the same as if the value was omitted entirely. * Fix "unhandled response" error in test harness (#180) When we return a `('', 204)` response from the flask handler, [Werkzeug intentionally removes the 'Content-Type' header][1], which causes the response to be created as a chunked response. The test harness is likely seeing a 204 response and isn't trying to read anything more from the stream. But since we are re-using connections, the next time it reads from the stream, it sees the `0\r\n\r\n` chunk and outputs an error: > 2022/04/20 14:23:39 Unsolicited response received on idle HTTP channel starting with "0\r\n\r\n"; err=<nil> Changing this response to 202 causes Werkzeug to return an empty response and silences the error. [1]: https://github.com/pallets/werkzeug/blob/560dd5f320bff318175f209595d42f5a80045417/src/werkzeug/wrappers/response.py#L540 * Exclude booleans when getting bucketable value (#181) When calculating a bucket, we get the bucketable value from the specified bucket by attribute. If this value is a string or an int, we can use it. Otherwise, we return None. Python considers a bool an instance of an int, which isn't what we want. So we need to add an explicit exclusion for this. * master -> main (#182) * Loosen restriction on expiringdict (#183) Originally this was pinned to a max version to deal with the incompatibility of Python 3.3 and the `typing` package. See [this PR][1]. Now that we now only support >=3.5, we can safely relax this restriction again. [1]: launchdarkly/python-server-sdk-private#120 * Fix mypy type checking (#184) A [customer requested][original-pr] that we start including a py.typed file in our repository. This would enable mypy to take advantage of our typehints. Unfortunately, this didn't completely solve the customers issue. A [second pr][second-pr] was opened to address the missing step of including the py.typed file in the `Manifest.in` file. However, this change alone is not sufficient. According to the [documentation][include_package_data], you must also include the `include_package_data=True` directive so that files specified in the `Manifest.in` file are included in distribution. [original-pr]: #166 [second-pr]: #172 [include_package_data]: https://setuptools.pypa.io/en/latest/userguide/datafiles.html#include-package-data * Add support for extra Redis connection parameters (#185) * Include wheel artifact when publishing package (#186) * skip tests that use a self-signed TLS cert in Python 3.7 * remove warn-level logging done for every Big Segments query (#190) * remove warn-level logging done for every Big Segments query * skip tests that use a self-signed TLS cert in Python 3.7 * update release metadata * Add application info support (#214) * Upgrade pip to fix failing CI build (#216) The CI build was failing because pip had an outdated list of available wheels for installation. Since it couldn't find a match, it was trying to build a package from source, which requires the rust compiler, which in turn isn't present on some of the docker images. By updating pip we get the updated list of available wheels, thereby allowing us to bypass source building and the need for the rust compiler entirely. --------- Co-authored-by: Eli Bishop <[email protected]> Co-authored-by: LaunchDarklyCI <[email protected]> Co-authored-by: Ben Woskow <[email protected]> Co-authored-by: Gavin Whelan <[email protected]> Co-authored-by: Elliot <[email protected]> Co-authored-by: Gabor Angeli <[email protected]> Co-authored-by: Elliot <[email protected]> Co-authored-by: Ben Woskow <[email protected]> Co-authored-by: LaunchDarklyCI <[email protected]> Co-authored-by: hroederld <[email protected]> Co-authored-by: Robert J. Neal <[email protected]> Co-authored-by: Robert J. Neal <[email protected]> Co-authored-by: Ember Stevens <[email protected]> Co-authored-by: ember-stevens <[email protected]> Co-authored-by: Matthew M. Keeler <[email protected]> Co-authored-by: charukiewicz <[email protected]> Co-authored-by: LaunchDarklyReleaseBot <[email protected]> Co-authored-by: Christian Charukiewicz <[email protected]> Co-authored-by: Matthew M. Keeler <[email protected]> * Releasing version 7.6.0 * Fix double spaces in changelog (#217) * Fix unsafe access to flag's trackEvents field (#218) When using the test data source, the trackEvents may not be required. Additionally, when receiving this information from LD, there is no guarantee that field will be provided. Instead of assuming that field always exists on the flag model, we should instead retrieve it with a default value to prevent index errors. * prepare 7.6.1 release (#195) * add 3.8 build * image name * fail on SyntaxWarning * typo * command syntax * pin expiringdict dependency for Python 3.3 compatibility * add Windows CircleCI job * periods are no longer valid in CircleCI job names * syntax fix * install Python in Windows * set path * move command * turn off debug logging * Py3 in Windows * config param * rm redundant step * choco switch * refactor Linux jobs using CircleCI 2.1 features * set log level before anything else * rm Azure config * use yaml.safe_load() to avoid code execution vulnerability in file data source * Initial work on wrapper_name, wrapper_version, diagnostic config options and start of diagnostic config event creation. * Python 2 compat changes. * More event generation code and starting to integrate tracking diagnostic values. * Add minimum diagnostic recording interval. Fix diagnostic.py to be importable. Add more diagnostic event fields. * don't let user fall outside of last bucket in rollout * fixing conditional logic * Add docstrings for diagnostic configuration options. * fix off-by-1 error * avoid redundant dict lookups * add unit tests for basic bucketing logic and edge case * Stream init tracking. Feeding of accumulator object through SDK. Various fixes. * Track events in last batch. * Fix sdk version field, some stylistic improvements. * Last of diagnostic configuration object fields. * Fill out rest of platform fields. * Cleanup and failed stream initialization tracking. * Add diagnostic config option test. * Add tests for diagnostics.py * Testing rest of diagnostic fields. * Test that streaming update processor records successful and unsuccessful connection attempts in the diagnostic accumulator when available. * Improvements to testability of event processor. * Rest of event processor tests. * Remove janky reflection. * Test change to filesource optional test requirements. * [ch61092] Add event payload ID on event requests. * normalize data store type and OS name in diagnostic events * gitignore * copyedit to diagnostic event config property comment * fix spurious error after sending diagnostic event * make verify_ssl=False turn off certificate verification too (#129) * add more TLS config options and collect HTTP/HTTPS config options in a class (#130) * make stream retry/backoff/jitter behavior consistent with other SDKs + improve testing (#131) * streams shouldn't use the same read timeout as the rest of the SDK (#132) * implement our own retry logic & logging for event posts, don't use urllib3.Retry (#133) * remove support for indirect/patch and indirect/put * remove unused logic for individual flag/segment poll for indirect/patch * Ehaisley/84082/remove python2 (#136) * remove all references to six and remove queue fallback imports * remove NullHandler logger backwards compat * update circleci config to remove python 2.7 tests * remove ordereddict backwards compat * update setup.py to no longer list python 2.7 as compatible * no longer inherit from object for python 2 backwards compat * update readme and manifest to reflect python 2.7 removal * remove unicode type compatibility * remove 2.7 support from circleci * Allow authenticating with proxy This commit allows for authenticating with a proxy configured with the `http_proxy` environment variable. Authentication requires passing a header, and is not parsed by urllib3 from the proxy_url. * reimplement proxy tests for DRY and add test of proxy auth params * doc comment on auth params in proxy URL * add type hints to some of the public facing api. update some docs * Revert "add type hints to some of the public facing api." This reverts commit c35fa61. * Ehaisley/ch86857/type hints (#138) * add typehints to the public API * validate typehints in the public api and tests with mypy * remove all current deprecations (#139) * remove all currently deprecated classes, methods, arguments, and tests * also update semver usage to remove calls to deprecated functions and classes * remove global set_sdk_key, make SDK key required in Config (#140) * Removed the guides link * Pinning mypy and running it against different python versions (#141) * fix time zone mishandling that could make event debugging not work (#142) * fix 6.x build (#143) * fix time zone mishandling that could make event debugging not work (6.x) (#144) * prepare 6.13.3 release (#154) * Releasing version 6.13.3 * [ch99756] Add alias events (#145) * add support for experiment rollouts * fix unit test * address PR comments * use Releaser v2 config * Use newer docker images (#147) * Updates docs URLs * Add support for 3.10 (#150) * started work on FlagBuilder in as part of test data source implementation * finished FlagBuilder implementation and added FlagRuleBuilder implementation * added initial TestData interface and updated tests to not rely on test data internals * started data source implementation * changed FlagBuilder to public class; changed FlagBuilder attributes to be initialized in __init__ and eliminated use of try ... except: pass for handling empty attributes * (big segments 1) add public config/interface types * added implementation of test data source * docstring * formatting * ensure property doesn't return None * (big segments 2) implement evaluation, refactor eval logic & modules * linting * (big segments 3) implement big segment status tracking, wire up components * typing fixes * typing fixes * implement SSE contract tests * fix CI * fix CI again * fix CI * disable SSE tests in Python 3.5 * make test service port configurable * better SSE implementation that fixes linefeed and multi-byte char issues * fix constructor parameters in test service * comment * test improvements * rm obsolete default config logic * (big segments 4) implement big segment stores in Redis+DynamoDB, refactor db tests (#158) * converted ldclient.integrations module from file to directory; started moving public classes out of ldclient.impl.integrations.test_data* and instead into ldclient.integrations.test_data*; started adding TestData documentation * removed setup/teardown functions leftover from test scaffold * added TestData, FlagBuilder, and FlagRuleBuilder documentation; minor adjustments to implementation details * removed warning supression from TestData tests * fix big segments user hash algorithm to use SHA256 * update mypy version * updates to tests and related bug fixes * always cache Big Segment query result even if it's None * fix test assertion * lint * fix big segment ref format * fix big segments cache TTL being set to wrong value * fixed structure of fallthrough variation in result of FlagBuilder.build() * moved __test__ attribute into TestData class definition to prevent mypy from complaining about a missing class attribute * minor doc comment fix * Apply suggestions related to Sphinx docstring formatting from code review Co-authored-by: Eli Bishop <[email protected]> * fixed errors in the implementation of FlagBuilder's fallthrough_variation and off_variation when passing boolean variation values; updated tests to assert the expected behavior * added missing value_for_all_users() method to FlagBuilder class * Fix operator parsing errors (#169) * identify should not emit event if user key is empty (#164) * secondary should be treated as built-in attribute (#168) * URIs should have trailing slashes trimmed (#165) * all_flags_state should always include flag version (#166) * output event should not include a null prereqOf key (#167) * Account for traffic allocation on all flags (#171) * Add SDK contract tests (#170) * misc fixes to test data docs + add type hints * more type hints * remove some methods from the public test_data API * can't use "x|y" shortcut in typehints in older Pythons; use Union * fix misc type mistakes because I forgot to run the linter * update CONTRIBUTING.md and provide make targets * fixed a bug with flag rule clause builder internals; added unit test to verify rule evaluation * added ready argument to _TestDataSource class and indicated ready upon start to avoid delays in TestData initialization * Update contract tests to latest flask version (#176) Our contract tests depend on flask v1, which in turn depends on Jinja 2. Both of these are terribly dated and no longer supported. Jinja depends on markupsafe. markupsafe recently updated its code to no longer provide soft_unicode which in turn broke Jinja. Updating to the latest flask keeps all transitive dependencies better aligned and addresses this mismatch. * Adds link to Relay Proxy docs * Handle explicit None values in test payload (#179) The test harness may send explicit None values which should be treated the same as if the value was omitted entirely. * Fix "unhandled response" error in test harness (#180) When we return a `('', 204)` response from the flask handler, [Werkzeug intentionally removes the 'Content-Type' header][1], which causes the response to be created as a chunked response. The test harness is likely seeing a 204 response and isn't trying to read anything more from the stream. But since we are re-using connections, the next time it reads from the stream, it sees the `0\r\n\r\n` chunk and outputs an error: > 2022/04/20 14:23:39 Unsolicited response received on idle HTTP channel starting with "0\r\n\r\n"; err=<nil> Changing this response to 202 causes Werkzeug to return an empty response and silences the error. [1]: https://github.com/pallets/werkzeug/blob/560dd5f320bff318175f209595d42f5a80045417/src/werkzeug/wrappers/response.py#L540 * Exclude booleans when getting bucketable value (#181) When calculating a bucket, we get the bucketable value from the specified bucket by attribute. If this value is a string or an int, we can use it. Otherwise, we return None. Python considers a bool an instance of an int, which isn't what we want. So we need to add an explicit exclusion for this. * master -> main (#182) * Loosen restriction on expiringdict (#183) Originally this was pinned to a max version to deal with the incompatibility of Python 3.3 and the `typing` package. See [this PR][1]. Now that we now only support >=3.5, we can safely relax this restriction again. [1]: launchdarkly/python-server-sdk-private#120 * Fix mypy type checking (#184) A [customer requested][original-pr] that we start including a py.typed file in our repository. This would enable mypy to take advantage of our typehints. Unfortunately, this didn't completely solve the customers issue. A [second pr][second-pr] was opened to address the missing step of including the py.typed file in the `Manifest.in` file. However, this change alone is not sufficient. According to the [documentation][include_package_data], you must also include the `include_package_data=True` directive so that files specified in the `Manifest.in` file are included in distribution. [original-pr]: #166 [second-pr]: #172 [include_package_data]: https://setuptools.pypa.io/en/latest/userguide/datafiles.html#include-package-data * Add support for extra Redis connection parameters (#185) * Include wheel artifact when publishing package (#186) * skip tests that use a self-signed TLS cert in Python 3.7 * remove warn-level logging done for every Big Segments query (#190) * remove warn-level logging done for every Big Segments query * skip tests that use a self-signed TLS cert in Python 3.7 * update release metadata * Add application info support (#214) * Upgrade pip to fix failing CI build (#216) The CI build was failing because pip had an outdated list of available wheels for installation. Since it couldn't find a match, it was trying to build a package from source, which requires the rust compiler, which in turn isn't present on some of the docker images. By updating pip we get the updated list of available wheels, thereby allowing us to bypass source building and the need for the rust compiler entirely. * Fix double spaces in changelog (#217) * Fix unsafe access to flag's trackEvents field (#218) When using the test data source, the trackEvents may not be required. Additionally, when receiving this information from LD, there is no guarantee that field will be provided. Instead of assuming that field always exists on the flag model, we should instead retrieve it with a default value to prevent index errors. --------- Co-authored-by: Eli Bishop <[email protected]> Co-authored-by: LaunchDarklyCI <[email protected]> Co-authored-by: Ben Woskow <[email protected]> Co-authored-by: Gavin Whelan <[email protected]> Co-authored-by: Elliot <[email protected]> Co-authored-by: Gabor Angeli <[email protected]> Co-authored-by: Elliot <[email protected]> Co-authored-by: Ben Woskow <[email protected]> Co-authored-by: LaunchDarklyCI <[email protected]> Co-authored-by: hroederld <[email protected]> Co-authored-by: Robert J. Neal <[email protected]> Co-authored-by: Robert J. Neal <[email protected]> Co-authored-by: Ember Stevens <[email protected]> Co-authored-by: ember-stevens <[email protected]> Co-authored-by: Matthew M. Keeler <[email protected]> Co-authored-by: charukiewicz <[email protected]> Co-authored-by: LaunchDarklyReleaseBot <[email protected]> Co-authored-by: Christian Charukiewicz <[email protected]> Co-authored-by: Matthew M. Keeler <[email protected]> * Releasing version 7.6.1 --------- Co-authored-by: Eli Bishop <[email protected]> Co-authored-by: LaunchDarklyCI <[email protected]> Co-authored-by: Elliot <[email protected]> Co-authored-by: Elliot <[email protected]> Co-authored-by: Ben Woskow <[email protected]> Co-authored-by: LaunchDarklyCI <[email protected]> Co-authored-by: hroederld <[email protected]> Co-authored-by: Robert J. Neal <[email protected]> Co-authored-by: Robert J. Neal <[email protected]> Co-authored-by: Ember Stevens <[email protected]> Co-authored-by: ember-stevens <[email protected]> Co-authored-by: Matthew M. Keeler <[email protected]> Co-authored-by: charukiewicz <[email protected]> Co-authored-by: LaunchDarklyReleaseBot <[email protected]> Co-authored-by: Christian Charukiewicz <[email protected]> Co-authored-by: Matthew M. Keeler <[email protected]> Co-authored-by: Ben Woskow <[email protected]> Co-authored-by: Gavin Whelan <[email protected]> Co-authored-by: Gabor Angeli <[email protected]>
* add type hints to some of the public facing api. update some docs * Revert "add type hints to some of the public facing api." This reverts commit c35fa61. * Ehaisley/ch86857/type hints (#138) * add typehints to the public API * validate typehints in the public api and tests with mypy * remove all current deprecations (#139) * remove all currently deprecated classes, methods, arguments, and tests * also update semver usage to remove calls to deprecated functions and classes * remove global set_sdk_key, make SDK key required in Config (#140) * Removed the guides link * Pinning mypy and running it against different python versions (#141) * fix time zone mishandling that could make event debugging not work (#142) * fix 6.x build (#143) * fix time zone mishandling that could make event debugging not work (6.x) (#144) * prepare 6.13.3 release (#154) * Releasing version 6.13.3 * [ch99756] Add alias events (#145) * add support for experiment rollouts * fix unit test * address PR comments * use Releaser v2 config * Use newer docker images (#147) * Updates docs URLs * Add support for 3.10 (#150) * started work on FlagBuilder in as part of test data source implementation * finished FlagBuilder implementation and added FlagRuleBuilder implementation * added initial TestData interface and updated tests to not rely on test data internals * started data source implementation * changed FlagBuilder to public class; changed FlagBuilder attributes to be initialized in __init__ and eliminated use of try ... except: pass for handling empty attributes * (big segments 1) add public config/interface types * added implementation of test data source * docstring * formatting * ensure property doesn't return None * (big segments 2) implement evaluation, refactor eval logic & modules * linting * (big segments 3) implement big segment status tracking, wire up components * typing fixes * typing fixes * implement SSE contract tests * fix CI * fix CI again * fix CI * disable SSE tests in Python 3.5 * make test service port configurable * better SSE implementation that fixes linefeed and multi-byte char issues * fix constructor parameters in test service * comment * test improvements * rm obsolete default config logic * (big segments 4) implement big segment stores in Redis+DynamoDB, refactor db tests (#158) * converted ldclient.integrations module from file to directory; started moving public classes out of ldclient.impl.integrations.test_data* and instead into ldclient.integrations.test_data*; started adding TestData documentation * removed setup/teardown functions leftover from test scaffold * added TestData, FlagBuilder, and FlagRuleBuilder documentation; minor adjustments to implementation details * removed warning supression from TestData tests * fix big segments user hash algorithm to use SHA256 * update mypy version * updates to tests and related bug fixes * always cache Big Segment query result even if it's None * fix test assertion * lint * fix big segment ref format * fix big segments cache TTL being set to wrong value * fixed structure of fallthrough variation in result of FlagBuilder.build() * moved __test__ attribute into TestData class definition to prevent mypy from complaining about a missing class attribute * minor doc comment fix * Apply suggestions related to Sphinx docstring formatting from code review Co-authored-by: Eli Bishop <[email protected]> * fixed errors in the implementation of FlagBuilder's fallthrough_variation and off_variation when passing boolean variation values; updated tests to assert the expected behavior * added missing value_for_all_users() method to FlagBuilder class * Fix operator parsing errors (#169) * identify should not emit event if user key is empty (#164) * secondary should be treated as built-in attribute (#168) * URIs should have trailing slashes trimmed (#165) * all_flags_state should always include flag version (#166) * output event should not include a null prereqOf key (#167) * Account for traffic allocation on all flags (#171) * Add SDK contract tests (#170) * misc fixes to test data docs + add type hints * more type hints * remove some methods from the public test_data API * can't use "x|y" shortcut in typehints in older Pythons; use Union * fix misc type mistakes because I forgot to run the linter * update CONTRIBUTING.md and provide make targets * fixed a bug with flag rule clause builder internals; added unit test to verify rule evaluation * added ready argument to _TestDataSource class and indicated ready upon start to avoid delays in TestData initialization * Update contract tests to latest flask version (#176) Our contract tests depend on flask v1, which in turn depends on Jinja 2. Both of these are terribly dated and no longer supported. Jinja depends on markupsafe. markupsafe recently updated its code to no longer provide soft_unicode which in turn broke Jinja. Updating to the latest flask keeps all transitive dependencies better aligned and addresses this mismatch. * Adds link to Relay Proxy docs * Handle explicit None values in test payload (#179) The test harness may send explicit None values which should be treated the same as if the value was omitted entirely. * Fix "unhandled response" error in test harness (#180) When we return a `('', 204)` response from the flask handler, [Werkzeug intentionally removes the 'Content-Type' header][1], which causes the response to be created as a chunked response. The test harness is likely seeing a 204 response and isn't trying to read anything more from the stream. But since we are re-using connections, the next time it reads from the stream, it sees the `0\r\n\r\n` chunk and outputs an error: > 2022/04/20 14:23:39 Unsolicited response received on idle HTTP channel starting with "0\r\n\r\n"; err=<nil> Changing this response to 202 causes Werkzeug to return an empty response and silences the error. [1]: https://github.com/pallets/werkzeug/blob/560dd5f320bff318175f209595d42f5a80045417/src/werkzeug/wrappers/response.py#L540 * Exclude booleans when getting bucketable value (#181) When calculating a bucket, we get the bucketable value from the specified bucket by attribute. If this value is a string or an int, we can use it. Otherwise, we return None. Python considers a bool an instance of an int, which isn't what we want. So we need to add an explicit exclusion for this. * master -> main (#182) * Loosen restriction on expiringdict (#183) Originally this was pinned to a max version to deal with the incompatibility of Python 3.3 and the `typing` package. See [this PR][1]. Now that we now only support >=3.5, we can safely relax this restriction again. [1]: launchdarkly/python-server-sdk-private#120 * Fix mypy type checking (#184) A [customer requested][original-pr] that we start including a py.typed file in our repository. This would enable mypy to take advantage of our typehints. Unfortunately, this didn't completely solve the customers issue. A [second pr][second-pr] was opened to address the missing step of including the py.typed file in the `Manifest.in` file. However, this change alone is not sufficient. According to the [documentation][include_package_data], you must also include the `include_package_data=True` directive so that files specified in the `Manifest.in` file are included in distribution. [original-pr]: #166 [second-pr]: #172 [include_package_data]: https://setuptools.pypa.io/en/latest/userguide/datafiles.html#include-package-data * Add support for extra Redis connection parameters (#185) * Include wheel artifact when publishing package (#186) * remove warn-level logging done for every Big Segments query * skip tests that use a self-signed TLS cert in Python 3.7 * (U2C 1) drop EOL Python versions (#189) * drop EOL Python versions * misc cleanup, show Python version in CI * add Python 3.11 CI job * add Python 3.11 to package metadata * (U2C 2) remove alias event functionality (#187) * (U2C 3) remove inline users in events (#188) * (U2C 4) remove deprecated things (#192) * remove warn-level logging done for every Big Segments query (#190) * remove warn-level logging done for every Big Segments query * skip tests that use a self-signed TLS cert in Python 3.7 * implement context model * fix exports * specify exports * add copy constructor * minimal changes for SDK methods & evaluator to accept Context * update tests, add subscript method * lint * in type hints, must use Dict[A, B] rather than dict[A, B] for Python <3.9 * support context kind in clauses + enable v2 contract tests * misc fixes * misc fixes * support contextTargets * support contextKind in rollouts/experiments * support includedContexts/excludedContexts in segment * comment copyedit Co-authored-by: Matthew M. Keeler <[email protected]> * comment fixes * rm unused Co-authored-by: Matthew M. Keeler <[email protected]> * fix create_multi to support flattening * lint * use custom classes for flag/segment data model * use store adapter for safety * misc cleanup * misc fixes for persistent stores * more database store fixes * support attribute reference lookups in evaluations * pass logger from client * context kind logic for big segments + enable big segment contract tests * formatting fixes + test cleanup * prerequisite cycle detection * segment recursion * define custom classes for event data * add module init file * linting * fix prereq stack logic * (U2C 17) U2C changes for events, not including private attributes (#205) * private attribute redaction * move a lot of code out of top-level modules * TestData changes for contexts * general doc comment fixes for 8.0 * U2C configuration updates * update release metadata * store flag/segment target lists as sets * fix type hint * preprocess clause values for time/regex/semver operators * fix type checking for matches operator * Add application info support (#214) * Add application info support (#214) (#215) * Upgrade pip to fix failing CI build (#216) The CI build was failing because pip had an outdated list of available wheels for installation. Since it couldn't find a match, it was trying to build a package from source, which requires the rust compiler, which in turn isn't present on some of the docker images. By updating pip we get the updated list of available wheels, thereby allowing us to bypass source building and the need for the rust compiler entirely. * prepare 7.6.0 release (#192) * comment * add end-to-end unit tests for proxy config * indents * add 3.8 build * image name * fail on SyntaxWarning * typo * command syntax * pin expiringdict dependency for Python 3.3 compatibility * add Windows CircleCI job * periods are no longer valid in CircleCI job names * syntax fix * install Python in Windows * set path * move command * turn off debug logging * Py3 in Windows * config param * rm redundant step * choco switch * refactor Linux jobs using CircleCI 2.1 features * set log level before anything else * rm Azure config * use yaml.safe_load() to avoid code execution vulnerability in file data source * Initial work on wrapper_name, wrapper_version, diagnostic config options and start of diagnostic config event creation. * Python 2 compat changes. * More event generation code and starting to integrate tracking diagnostic values. * Add minimum diagnostic recording interval. Fix diagnostic.py to be importable. Add more diagnostic event fields. * don't let user fall outside of last bucket in rollout * fixing conditional logic * Add docstrings for diagnostic configuration options. * fix off-by-1 error * avoid redundant dict lookups * add unit tests for basic bucketing logic and edge case * Stream init tracking. Feeding of accumulator object through SDK. Various fixes. * Track events in last batch. * Fix sdk version field, some stylistic improvements. * Last of diagnostic configuration object fields. * Fill out rest of platform fields. * Cleanup and failed stream initialization tracking. * Add diagnostic config option test. * Add tests for diagnostics.py * Testing rest of diagnostic fields. * Test that streaming update processor records successful and unsuccessful connection attempts in the diagnostic accumulator when available. * Improvements to testability of event processor. * Rest of event processor tests. * Remove janky reflection. * Test change to filesource optional test requirements. * [ch61092] Add event payload ID on event requests. * normalize data store type and OS name in diagnostic events * gitignore * copyedit to diagnostic event config property comment * fix spurious error after sending diagnostic event * make verify_ssl=False turn off certificate verification too (#129) * add more TLS config options and collect HTTP/HTTPS config options in a class (#130) * make stream retry/backoff/jitter behavior consistent with other SDKs + improve testing (#131) * streams shouldn't use the same read timeout as the rest of the SDK (#132) * implement our own retry logic & logging for event posts, don't use urllib3.Retry (#133) * remove support for indirect/patch and indirect/put * remove unused logic for individual flag/segment poll for indirect/patch * Ehaisley/84082/remove python2 (#136) * remove all references to six and remove queue fallback imports * remove NullHandler logger backwards compat * update circleci config to remove python 2.7 tests * remove ordereddict backwards compat * update setup.py to no longer list python 2.7 as compatible * no longer inherit from object for python 2 backwards compat * update readme and manifest to reflect python 2.7 removal * remove unicode type compatibility * remove 2.7 support from circleci * Allow authenticating with proxy This commit allows for authenticating with a proxy configured with the `http_proxy` environment variable. Authentication requires passing a header, and is not parsed by urllib3 from the proxy_url. * reimplement proxy tests for DRY and add test of proxy auth params * doc comment on auth params in proxy URL * add type hints to some of the public facing api. update some docs * Revert "add type hints to some of the public facing api." This reverts commit c35fa61. * Ehaisley/ch86857/type hints (#138) * add typehints to the public API * validate typehints in the public api and tests with mypy * remove all current deprecations (#139) * remove all currently deprecated classes, methods, arguments, and tests * also update semver usage to remove calls to deprecated functions and classes * remove global set_sdk_key, make SDK key required in Config (#140) * Removed the guides link * Pinning mypy and running it against different python versions (#141) * fix time zone mishandling that could make event debugging not work (#142) * fix 6.x build (#143) * fix time zone mishandling that could make event debugging not work (6.x) (#144) * prepare 6.13.3 release (#154) * Releasing version 6.13.3 * [ch99756] Add alias events (#145) * add support for experiment rollouts * fix unit test * address PR comments * use Releaser v2 config * Use newer docker images (#147) * Updates docs URLs * Add support for 3.10 (#150) * started work on FlagBuilder in as part of test data source implementation * finished FlagBuilder implementation and added FlagRuleBuilder implementation * added initial TestData interface and updated tests to not rely on test data internals * started data source implementation * changed FlagBuilder to public class; changed FlagBuilder attributes to be initialized in __init__ and eliminated use of try ... except: pass for handling empty attributes * (big segments 1) add public config/interface types * added implementation of test data source * docstring * formatting * ensure property doesn't return None * (big segments 2) implement evaluation, refactor eval logic & modules * linting * (big segments 3) implement big segment status tracking, wire up components * typing fixes * typing fixes * implement SSE contract tests * fix CI * fix CI again * fix CI * disable SSE tests in Python 3.5 * make test service port configurable * better SSE implementation that fixes linefeed and multi-byte char issues * fix constructor parameters in test service * comment * test improvements * rm obsolete default config logic * (big segments 4) implement big segment stores in Redis+DynamoDB, refactor db tests (#158) * converted ldclient.integrations module from file to directory; started moving public classes out of ldclient.impl.integrations.test_data* and instead into ldclient.integrations.test_data*; started adding TestData documentation * removed setup/teardown functions leftover from test scaffold * added TestData, FlagBuilder, and FlagRuleBuilder documentation; minor adjustments to implementation details * removed warning supression from TestData tests * fix big segments user hash algorithm to use SHA256 * update mypy version * updates to tests and related bug fixes * always cache Big Segment query result even if it's None * fix test assertion * lint * fix big segment ref format * fix big segments cache TTL being set to wrong value * fixed structure of fallthrough variation in result of FlagBuilder.build() * moved __test__ attribute into TestData class definition to prevent mypy from complaining about a missing class attribute * minor doc comment fix * Apply suggestions related to Sphinx docstring formatting from code review Co-authored-by: Eli Bishop <[email protected]> * fixed errors in the implementation of FlagBuilder's fallthrough_variation and off_variation when passing boolean variation values; updated tests to assert the expected behavior * added missing value_for_all_users() method to FlagBuilder class * Fix operator parsing errors (#169) * identify should not emit event if user key is empty (#164) * secondary should be treated as built-in attribute (#168) * URIs should have trailing slashes trimmed (#165) * all_flags_state should always include flag version (#166) * output event should not include a null prereqOf key (#167) * Account for traffic allocation on all flags (#171) * Add SDK contract tests (#170) * misc fixes to test data docs + add type hints * more type hints * remove some methods from the public test_data API * can't use "x|y" shortcut in typehints in older Pythons; use Union * fix misc type mistakes because I forgot to run the linter * update CONTRIBUTING.md and provide make targets * fixed a bug with flag rule clause builder internals; added unit test to verify rule evaluation * added ready argument to _TestDataSource class and indicated ready upon start to avoid delays in TestData initialization * Update contract tests to latest flask version (#176) Our contract tests depend on flask v1, which in turn depends on Jinja 2. Both of these are terribly dated and no longer supported. Jinja depends on markupsafe. markupsafe recently updated its code to no longer provide soft_unicode which in turn broke Jinja. Updating to the latest flask keeps all transitive dependencies better aligned and addresses this mismatch. * Adds link to Relay Proxy docs * Handle explicit None values in test payload (#179) The test harness may send explicit None values which should be treated the same as if the value was omitted entirely. * Fix "unhandled response" error in test harness (#180) When we return a `('', 204)` response from the flask handler, [Werkzeug intentionally removes the 'Content-Type' header][1], which causes the response to be created as a chunked response. The test harness is likely seeing a 204 response and isn't trying to read anything more from the stream. But since we are re-using connections, the next time it reads from the stream, it sees the `0\r\n\r\n` chunk and outputs an error: > 2022/04/20 14:23:39 Unsolicited response received on idle HTTP channel starting with "0\r\n\r\n"; err=<nil> Changing this response to 202 causes Werkzeug to return an empty response and silences the error. [1]: https://github.com/pallets/werkzeug/blob/560dd5f320bff318175f209595d42f5a80045417/src/werkzeug/wrappers/response.py#L540 * Exclude booleans when getting bucketable value (#181) When calculating a bucket, we get the bucketable value from the specified bucket by attribute. If this value is a string or an int, we can use it. Otherwise, we return None. Python considers a bool an instance of an int, which isn't what we want. So we need to add an explicit exclusion for this. * master -> main (#182) * Loosen restriction on expiringdict (#183) Originally this was pinned to a max version to deal with the incompatibility of Python 3.3 and the `typing` package. See [this PR][1]. Now that we now only support >=3.5, we can safely relax this restriction again. [1]: launchdarkly/python-server-sdk-private#120 * Fix mypy type checking (#184) A [customer requested][original-pr] that we start including a py.typed file in our repository. This would enable mypy to take advantage of our typehints. Unfortunately, this didn't completely solve the customers issue. A [second pr][second-pr] was opened to address the missing step of including the py.typed file in the `Manifest.in` file. However, this change alone is not sufficient. According to the [documentation][include_package_data], you must also include the `include_package_data=True` directive so that files specified in the `Manifest.in` file are included in distribution. [original-pr]: #166 [second-pr]: #172 [include_package_data]: https://setuptools.pypa.io/en/latest/userguide/datafiles.html#include-package-data * Add support for extra Redis connection parameters (#185) * Include wheel artifact when publishing package (#186) * skip tests that use a self-signed TLS cert in Python 3.7 * remove warn-level logging done for every Big Segments query (#190) * remove warn-level logging done for every Big Segments query * skip tests that use a self-signed TLS cert in Python 3.7 * update release metadata * Add application info support (#214) * Upgrade pip to fix failing CI build (#216) The CI build was failing because pip had an outdated list of available wheels for installation. Since it couldn't find a match, it was trying to build a package from source, which requires the rust compiler, which in turn isn't present on some of the docker images. By updating pip we get the updated list of available wheels, thereby allowing us to bypass source building and the need for the rust compiler entirely. --------- Co-authored-by: Eli Bishop <[email protected]> Co-authored-by: LaunchDarklyCI <[email protected]> Co-authored-by: Ben Woskow <[email protected]> Co-authored-by: Gavin Whelan <[email protected]> Co-authored-by: Elliot <[email protected]> Co-authored-by: Gabor Angeli <[email protected]> Co-authored-by: Elliot <[email protected]> Co-authored-by: Ben Woskow <[email protected]> Co-authored-by: LaunchDarklyCI <[email protected]> Co-authored-by: hroederld <[email protected]> Co-authored-by: Robert J. Neal <[email protected]> Co-authored-by: Robert J. Neal <[email protected]> Co-authored-by: Ember Stevens <[email protected]> Co-authored-by: ember-stevens <[email protected]> Co-authored-by: Matthew M. Keeler <[email protected]> Co-authored-by: charukiewicz <[email protected]> Co-authored-by: LaunchDarklyReleaseBot <[email protected]> Co-authored-by: Christian Charukiewicz <[email protected]> Co-authored-by: Matthew M. Keeler <[email protected]> * Releasing version 7.6.0 * Fix double spaces in changelog (#217) * Fix unsafe access to flag's trackEvents field (#218) When using the test data source, the trackEvents may not be required. Additionally, when receiving this information from LD, there is no guarantee that field will be provided. Instead of assuming that field always exists on the flag model, we should instead retrieve it with a default value to prevent index errors. * prepare 7.6.1 release (#195) * add 3.8 build * image name * fail on SyntaxWarning * typo * command syntax * pin expiringdict dependency for Python 3.3 compatibility * add Windows CircleCI job * periods are no longer valid in CircleCI job names * syntax fix * install Python in Windows * set path * move command * turn off debug logging * Py3 in Windows * config param * rm redundant step * choco switch * refactor Linux jobs using CircleCI 2.1 features * set log level before anything else * rm Azure config * use yaml.safe_load() to avoid code execution vulnerability in file data source * Initial work on wrapper_name, wrapper_version, diagnostic config options and start of diagnostic config event creation. * Python 2 compat changes. * More event generation code and starting to integrate tracking diagnostic values. * Add minimum diagnostic recording interval. Fix diagnostic.py to be importable. Add more diagnostic event fields. * don't let user fall outside of last bucket in rollout * fixing conditional logic * Add docstrings for diagnostic configuration options. * fix off-by-1 error * avoid redundant dict lookups * add unit tests for basic bucketing logic and edge case * Stream init tracking. Feeding of accumulator object through SDK. Various fixes. * Track events in last batch. * Fix sdk version field, some stylistic improvements. * Last of diagnostic configuration object fields. * Fill out rest of platform fields. * Cleanup and failed stream initialization tracking. * Add diagnostic config option test. * Add tests for diagnostics.py * Testing rest of diagnostic fields. * Test that streaming update processor records successful and unsuccessful connection attempts in the diagnostic accumulator when available. * Improvements to testability of event processor. * Rest of event processor tests. * Remove janky reflection. * Test change to filesource optional test requirements. * [ch61092] Add event payload ID on event requests. * normalize data store type and OS name in diagnostic events * gitignore * copyedit to diagnostic event config property comment * fix spurious error after sending diagnostic event * make verify_ssl=False turn off certificate verification too (#129) * add more TLS config options and collect HTTP/HTTPS config options in a class (#130) * make stream retry/backoff/jitter behavior consistent with other SDKs + improve testing (#131) * streams shouldn't use the same read timeout as the rest of the SDK (#132) * implement our own retry logic & logging for event posts, don't use urllib3.Retry (#133) * remove support for indirect/patch and indirect/put * remove unused logic for individual flag/segment poll for indirect/patch * Ehaisley/84082/remove python2 (#136) * remove all references to six and remove queue fallback imports * remove NullHandler logger backwards compat * update circleci config to remove python 2.7 tests * remove ordereddict backwards compat * update setup.py to no longer list python 2.7 as compatible * no longer inherit from object for python 2 backwards compat * update readme and manifest to reflect python 2.7 removal * remove unicode type compatibility * remove 2.7 support from circleci * Allow authenticating with proxy This commit allows for authenticating with a proxy configured with the `http_proxy` environment variable. Authentication requires passing a header, and is not parsed by urllib3 from the proxy_url. * reimplement proxy tests for DRY and add test of proxy auth params * doc comment on auth params in proxy URL * add type hints to some of the public facing api. update some docs * Revert "add type hints to some of the public facing api." This reverts commit c35fa61. * Ehaisley/ch86857/type hints (#138) * add typehints to the public API * validate typehints in the public api and tests with mypy * remove all current deprecations (#139) * remove all currently deprecated classes, methods, arguments, and tests * also update semver usage to remove calls to deprecated functions and classes * remove global set_sdk_key, make SDK key required in Config (#140) * Removed the guides link * Pinning mypy and running it against different python versions (#141) * fix time zone mishandling that could make event debugging not work (#142) * fix 6.x build (#143) * fix time zone mishandling that could make event debugging not work (6.x) (#144) * prepare 6.13.3 release (#154) * Releasing version 6.13.3 * [ch99756] Add alias events (#145) * add support for experiment rollouts * fix unit test * address PR comments * use Releaser v2 config * Use newer docker images (#147) * Updates docs URLs * Add support for 3.10 (#150) * started work on FlagBuilder in as part of test data source implementation * finished FlagBuilder implementation and added FlagRuleBuilder implementation * added initial TestData interface and updated tests to not rely on test data internals * started data source implementation * changed FlagBuilder to public class; changed FlagBuilder attributes to be initialized in __init__ and eliminated use of try ... except: pass for handling empty attributes * (big segments 1) add public config/interface types * added implementation of test data source * docstring * formatting * ensure property doesn't return None * (big segments 2) implement evaluation, refactor eval logic & modules * linting * (big segments 3) implement big segment status tracking, wire up components * typing fixes * typing fixes * implement SSE contract tests * fix CI * fix CI again * fix CI * disable SSE tests in Python 3.5 * make test service port configurable * better SSE implementation that fixes linefeed and multi-byte char issues * fix constructor parameters in test service * comment * test improvements * rm obsolete default config logic * (big segments 4) implement big segment stores in Redis+DynamoDB, refactor db tests (#158) * converted ldclient.integrations module from file to directory; started moving public classes out of ldclient.impl.integrations.test_data* and instead into ldclient.integrations.test_data*; started adding TestData documentation * removed setup/teardown functions leftover from test scaffold * added TestData, FlagBuilder, and FlagRuleBuilder documentation; minor adjustments to implementation details * removed warning supression from TestData tests * fix big segments user hash algorithm to use SHA256 * update mypy version * updates to tests and related bug fixes * always cache Big Segment query result even if it's None * fix test assertion * lint * fix big segment ref format * fix big segments cache TTL being set to wrong value * fixed structure of fallthrough variation in result of FlagBuilder.build() * moved __test__ attribute into TestData class definition to prevent mypy from complaining about a missing class attribute * minor doc comment fix * Apply suggestions related to Sphinx docstring formatting from code review Co-authored-by: Eli Bishop <[email protected]> * fixed errors in the implementation of FlagBuilder's fallthrough_variation and off_variation when passing boolean variation values; updated tests to assert the expected behavior * added missing value_for_all_users() method to FlagBuilder class * Fix operator parsing errors (#169) * identify should not emit event if user key is empty (#164) * secondary should be treated as built-in attribute (#168) * URIs should have trailing slashes trimmed (#165) * all_flags_state should always include flag version (#166) * output event should not include a null prereqOf key (#167) * Account for traffic allocation on all flags (#171) * Add SDK contract tests (#170) * misc fixes to test data docs + add type hints * more type hints * remove some methods from the public test_data API * can't use "x|y" shortcut in typehints in older Pythons; use Union * fix misc type mistakes because I forgot to run the linter * update CONTRIBUTING.md and provide make targets * fixed a bug with flag rule clause builder internals; added unit test to verify rule evaluation * added ready argument to _TestDataSource class and indicated ready upon start to avoid delays in TestData initialization * Update contract tests to latest flask version (#176) Our contract tests depend on flask v1, which in turn depends on Jinja 2. Both of these are terribly dated and no longer supported. Jinja depends on markupsafe. markupsafe recently updated its code to no longer provide soft_unicode which in turn broke Jinja. Updating to the latest flask keeps all transitive dependencies better aligned and addresses this mismatch. * Adds link to Relay Proxy docs * Handle explicit None values in test payload (#179) The test harness may send explicit None values which should be treated the same as if the value was omitted entirely. * Fix "unhandled response" error in test harness (#180) When we return a `('', 204)` response from the flask handler, [Werkzeug intentionally removes the 'Content-Type' header][1], which causes the response to be created as a chunked response. The test harness is likely seeing a 204 response and isn't trying to read anything more from the stream. But since we are re-using connections, the next time it reads from the stream, it sees the `0\r\n\r\n` chunk and outputs an error: > 2022/04/20 14:23:39 Unsolicited response received on idle HTTP channel starting with "0\r\n\r\n"; err=<nil> Changing this response to 202 causes Werkzeug to return an empty response and silences the error. [1]: https://github.com/pallets/werkzeug/blob/560dd5f320bff318175f209595d42f5a80045417/src/werkzeug/wrappers/response.py#L540 * Exclude booleans when getting bucketable value (#181) When calculating a bucket, we get the bucketable value from the specified bucket by attribute. If this value is a string or an int, we can use it. Otherwise, we return None. Python considers a bool an instance of an int, which isn't what we want. So we need to add an explicit exclusion for this. * master -> main (#182) * Loosen restriction on expiringdict (#183) Originally this was pinned to a max version to deal with the incompatibility of Python 3.3 and the `typing` package. See [this PR][1]. Now that we now only support >=3.5, we can safely relax this restriction again. [1]: launchdarkly/python-server-sdk-private#120 * Fix mypy type checking (#184) A [customer requested][original-pr] that we start including a py.typed file in our repository. This would enable mypy to take advantage of our typehints. Unfortunately, this didn't completely solve the customers issue. A [second pr][second-pr] was opened to address the missing step of including the py.typed file in the `Manifest.in` file. However, this change alone is not sufficient. According to the [documentation][include_package_data], you must also include the `include_package_data=True` directive so that files specified in the `Manifest.in` file are included in distribution. [original-pr]: #166 [second-pr]: #172 [include_package_data]: https://setuptools.pypa.io/en/latest/userguide/datafiles.html#include-package-data * Add support for extra Redis connection parameters (#185) * Include wheel artifact when publishing package (#186) * skip tests that use a self-signed TLS cert in Python 3.7 * remove warn-level logging done for every Big Segments query (#190) * remove warn-level logging done for every Big Segments query * skip tests that use a self-signed TLS cert in Python 3.7 * update release metadata * Add application info support (#214) * Upgrade pip to fix failing CI build (#216) The CI build was failing because pip had an outdated list of available wheels for installation. Since it couldn't find a match, it was trying to build a package from source, which requires the rust compiler, which in turn isn't present on some of the docker images. By updating pip we get the updated list of available wheels, thereby allowing us to bypass source building and the need for the rust compiler entirely. * Fix double spaces in changelog (#217) * Fix unsafe access to flag's trackEvents field (#218) When using the test data source, the trackEvents may not be required. Additionally, when receiving this information from LD, there is no guarantee that field will be provided. Instead of assuming that field always exists on the flag model, we should instead retrieve it with a default value to prevent index errors. --------- Co-authored-by: Eli Bishop <[email protected]> Co-authored-by: LaunchDarklyCI <[email protected]> Co-authored-by: Ben Woskow <[email protected]> Co-authored-by: Gavin Whelan <[email protected]> Co-authored-by: Elliot <[email protected]> Co-authored-by: Gabor Angeli <[email protected]> Co-authored-by: Elliot <[email protected]> Co-authored-by: Ben Woskow <[email protected]> Co-authored-by: LaunchDarklyCI <[email protected]> Co-authored-by: hroederld <[email protected]> Co-authored-by: Robert J. Neal <[email protected]> Co-authored-by: Robert J. Neal <[email protected]> Co-authored-by: Ember Stevens <[email protected]> Co-authored-by: ember-stevens <[email protected]> Co-authored-by: Matthew M. Keeler <[email protected]> Co-authored-by: charukiewicz <[email protected]> Co-authored-by: LaunchDarklyReleaseBot <[email protected]> Co-authored-by: Christian Charukiewicz <[email protected]> Co-authored-by: Matthew M. Keeler <[email protected]> * Releasing version 7.6.1 * docs: Fix sphinx documentation failures (#219) Our documentation file had a configuration issue, which emits a warning when building with sphinx. This caused the build process on readthedocs.org to fail because they treat warnings as errors. We didn't catch this because: 1. Warnings are treated like warnings when generating documentation locally. 2. We don't try to generate the documentation as part of our CI process. This commit should resolve both the underlying issue and the deficiencies that prevented us from catching this initially. * Fix documentation generation (#197) (#220) * fix: Restrict urllib to <2.0.0 (#202) --------- Co-authored-by: LaunchDarklyCI <[email protected]> Co-authored-by: Eli Bishop <[email protected]> Co-authored-by: Elliot <[email protected]> Co-authored-by: Elliot <[email protected]> Co-authored-by: Ben Woskow <[email protected]> Co-authored-by: LaunchDarklyCI <[email protected]> Co-authored-by: hroederld <[email protected]> Co-authored-by: Robert J. Neal <[email protected]> Co-authored-by: Robert J. Neal <[email protected]> Co-authored-by: Ember Stevens <[email protected]> Co-authored-by: ember-stevens <[email protected]> Co-authored-by: Matthew M. Keeler <[email protected]> Co-authored-by: charukiewicz <[email protected]> Co-authored-by: LaunchDarklyReleaseBot <[email protected]> Co-authored-by: Christian Charukiewicz <[email protected]> Co-authored-by: Matthew M. Keeler <[email protected]> Co-authored-by: Ben Woskow <[email protected]> Co-authored-by: Gavin Whelan <[email protected]> Co-authored-by: Gabor Angeli <[email protected]> Co-authored-by: Louis Chan <[email protected]> Co-authored-by: prpnmac <[email protected]>
* Revert "add type hints to some of the public facing api." This reverts commit c35fa61. * Ehaisley/ch86857/type hints (#138) * add typehints to the public API * validate typehints in the public api and tests with mypy * remove all current deprecations (#139) * remove all currently deprecated classes, methods, arguments, and tests * also update semver usage to remove calls to deprecated functions and classes * remove global set_sdk_key, make SDK key required in Config (#140) * Removed the guides link * Pinning mypy and running it against different python versions (#141) * fix time zone mishandling that could make event debugging not work (#142) * fix 6.x build (#143) * fix time zone mishandling that could make event debugging not work (6.x) (#144) * prepare 6.13.3 release (#154) * Releasing version 6.13.3 * [ch99756] Add alias events (#145) * add support for experiment rollouts * fix unit test * address PR comments * use Releaser v2 config * Use newer docker images (#147) * Updates docs URLs * Add support for 3.10 (#150) * started work on FlagBuilder in as part of test data source implementation * finished FlagBuilder implementation and added FlagRuleBuilder implementation * added initial TestData interface and updated tests to not rely on test data internals * started data source implementation * changed FlagBuilder to public class; changed FlagBuilder attributes to be initialized in __init__ and eliminated use of try ... except: pass for handling empty attributes * (big segments 1) add public config/interface types * added implementation of test data source * docstring * formatting * ensure property doesn't return None * (big segments 2) implement evaluation, refactor eval logic & modules * linting * (big segments 3) implement big segment status tracking, wire up components * typing fixes * typing fixes * implement SSE contract tests * fix CI * fix CI again * fix CI * disable SSE tests in Python 3.5 * make test service port configurable * better SSE implementation that fixes linefeed and multi-byte char issues * fix constructor parameters in test service * comment * test improvements * rm obsolete default config logic * (big segments 4) implement big segment stores in Redis+DynamoDB, refactor db tests (#158) * converted ldclient.integrations module from file to directory; started moving public classes out of ldclient.impl.integrations.test_data* and instead into ldclient.integrations.test_data*; started adding TestData documentation * removed setup/teardown functions leftover from test scaffold * added TestData, FlagBuilder, and FlagRuleBuilder documentation; minor adjustments to implementation details * removed warning supression from TestData tests * fix big segments user hash algorithm to use SHA256 * update mypy version * updates to tests and related bug fixes * always cache Big Segment query result even if it's None * fix test assertion * lint * fix big segment ref format * fix big segments cache TTL being set to wrong value * fixed structure of fallthrough variation in result of FlagBuilder.build() * moved __test__ attribute into TestData class definition to prevent mypy from complaining about a missing class attribute * minor doc comment fix * Apply suggestions related to Sphinx docstring formatting from code review Co-authored-by: Eli Bishop <[email protected]> * fixed errors in the implementation of FlagBuilder's fallthrough_variation and off_variation when passing boolean variation values; updated tests to assert the expected behavior * added missing value_for_all_users() method to FlagBuilder class * Fix operator parsing errors (#169) * identify should not emit event if user key is empty (#164) * secondary should be treated as built-in attribute (#168) * URIs should have trailing slashes trimmed (#165) * all_flags_state should always include flag version (#166) * output event should not include a null prereqOf key (#167) * Account for traffic allocation on all flags (#171) * Add SDK contract tests (#170) * misc fixes to test data docs + add type hints * more type hints * remove some methods from the public test_data API * can't use "x|y" shortcut in typehints in older Pythons; use Union * fix misc type mistakes because I forgot to run the linter * update CONTRIBUTING.md and provide make targets * fixed a bug with flag rule clause builder internals; added unit test to verify rule evaluation * added ready argument to _TestDataSource class and indicated ready upon start to avoid delays in TestData initialization * Update contract tests to latest flask version (#176) Our contract tests depend on flask v1, which in turn depends on Jinja 2. Both of these are terribly dated and no longer supported. Jinja depends on markupsafe. markupsafe recently updated its code to no longer provide soft_unicode which in turn broke Jinja. Updating to the latest flask keeps all transitive dependencies better aligned and addresses this mismatch. * Adds link to Relay Proxy docs * Handle explicit None values in test payload (#179) The test harness may send explicit None values which should be treated the same as if the value was omitted entirely. * Fix "unhandled response" error in test harness (#180) When we return a `('', 204)` response from the flask handler, [Werkzeug intentionally removes the 'Content-Type' header][1], which causes the response to be created as a chunked response. The test harness is likely seeing a 204 response and isn't trying to read anything more from the stream. But since we are re-using connections, the next time it reads from the stream, it sees the `0\r\n\r\n` chunk and outputs an error: > 2022/04/20 14:23:39 Unsolicited response received on idle HTTP channel starting with "0\r\n\r\n"; err=<nil> Changing this response to 202 causes Werkzeug to return an empty response and silences the error. [1]: https://github.com/pallets/werkzeug/blob/560dd5f320bff318175f209595d42f5a80045417/src/werkzeug/wrappers/response.py#L540 * Exclude booleans when getting bucketable value (#181) When calculating a bucket, we get the bucketable value from the specified bucket by attribute. If this value is a string or an int, we can use it. Otherwise, we return None. Python considers a bool an instance of an int, which isn't what we want. So we need to add an explicit exclusion for this. * master -> main (#182) * Loosen restriction on expiringdict (#183) Originally this was pinned to a max version to deal with the incompatibility of Python 3.3 and the `typing` package. See [this PR][1]. Now that we now only support >=3.5, we can safely relax this restriction again. [1]: launchdarkly/python-server-sdk-private#120 * Fix mypy type checking (#184) A [customer requested][original-pr] that we start including a py.typed file in our repository. This would enable mypy to take advantage of our typehints. Unfortunately, this didn't completely solve the customers issue. A [second pr][second-pr] was opened to address the missing step of including the py.typed file in the `Manifest.in` file. However, this change alone is not sufficient. According to the [documentation][include_package_data], you must also include the `include_package_data=True` directive so that files specified in the `Manifest.in` file are included in distribution. [original-pr]: #166 [second-pr]: #172 [include_package_data]: https://setuptools.pypa.io/en/latest/userguide/datafiles.html#include-package-data * Add support for extra Redis connection parameters (#185) * Include wheel artifact when publishing package (#186) * remove warn-level logging done for every Big Segments query * skip tests that use a self-signed TLS cert in Python 3.7 * (U2C 1) drop EOL Python versions (#189) * drop EOL Python versions * misc cleanup, show Python version in CI * add Python 3.11 CI job * add Python 3.11 to package metadata * (U2C 2) remove alias event functionality (#187) * (U2C 3) remove inline users in events (#188) * (U2C 4) remove deprecated things (#192) * remove warn-level logging done for every Big Segments query (#190) * remove warn-level logging done for every Big Segments query * skip tests that use a self-signed TLS cert in Python 3.7 * implement context model * fix exports * specify exports * add copy constructor * minimal changes for SDK methods & evaluator to accept Context * update tests, add subscript method * lint * in type hints, must use Dict[A, B] rather than dict[A, B] for Python <3.9 * support context kind in clauses + enable v2 contract tests * misc fixes * misc fixes * support contextTargets * support contextKind in rollouts/experiments * support includedContexts/excludedContexts in segment * comment copyedit Co-authored-by: Matthew M. Keeler <[email protected]> * comment fixes * rm unused Co-authored-by: Matthew M. Keeler <[email protected]> * fix create_multi to support flattening * lint * use custom classes for flag/segment data model * use store adapter for safety * misc cleanup * misc fixes for persistent stores * more database store fixes * support attribute reference lookups in evaluations * pass logger from client * context kind logic for big segments + enable big segment contract tests * formatting fixes + test cleanup * prerequisite cycle detection * segment recursion * define custom classes for event data * add module init file * linting * fix prereq stack logic * (U2C 17) U2C changes for events, not including private attributes (#205) * private attribute redaction * move a lot of code out of top-level modules * TestData changes for contexts * general doc comment fixes for 8.0 * U2C configuration updates * update release metadata * store flag/segment target lists as sets * fix type hint * preprocess clause values for time/regex/semver operators * fix type checking for matches operator * Add application info support (#214) * Add application info support (#214) (#215) * Upgrade pip to fix failing CI build (#216) The CI build was failing because pip had an outdated list of available wheels for installation. Since it couldn't find a match, it was trying to build a package from source, which requires the rust compiler, which in turn isn't present on some of the docker images. By updating pip we get the updated list of available wheels, thereby allowing us to bypass source building and the need for the rust compiler entirely. * prepare 7.6.0 release (#192) * comment * add end-to-end unit tests for proxy config * indents * add 3.8 build * image name * fail on SyntaxWarning * typo * command syntax * pin expiringdict dependency for Python 3.3 compatibility * add Windows CircleCI job * periods are no longer valid in CircleCI job names * syntax fix * install Python in Windows * set path * move command * turn off debug logging * Py3 in Windows * config param * rm redundant step * choco switch * refactor Linux jobs using CircleCI 2.1 features * set log level before anything else * rm Azure config * use yaml.safe_load() to avoid code execution vulnerability in file data source * Initial work on wrapper_name, wrapper_version, diagnostic config options and start of diagnostic config event creation. * Python 2 compat changes. * More event generation code and starting to integrate tracking diagnostic values. * Add minimum diagnostic recording interval. Fix diagnostic.py to be importable. Add more diagnostic event fields. * don't let user fall outside of last bucket in rollout * fixing conditional logic * Add docstrings for diagnostic configuration options. * fix off-by-1 error * avoid redundant dict lookups * add unit tests for basic bucketing logic and edge case * Stream init tracking. Feeding of accumulator object through SDK. Various fixes. * Track events in last batch. * Fix sdk version field, some stylistic improvements. * Last of diagnostic configuration object fields. * Fill out rest of platform fields. * Cleanup and failed stream initialization tracking. * Add diagnostic config option test. * Add tests for diagnostics.py * Testing rest of diagnostic fields. * Test that streaming update processor records successful and unsuccessful connection attempts in the diagnostic accumulator when available. * Improvements to testability of event processor. * Rest of event processor tests. * Remove janky reflection. * Test change to filesource optional test requirements. * [ch61092] Add event payload ID on event requests. * normalize data store type and OS name in diagnostic events * gitignore * copyedit to diagnostic event config property comment * fix spurious error after sending diagnostic event * make verify_ssl=False turn off certificate verification too (#129) * add more TLS config options and collect HTTP/HTTPS config options in a class (#130) * make stream retry/backoff/jitter behavior consistent with other SDKs + improve testing (#131) * streams shouldn't use the same read timeout as the rest of the SDK (#132) * implement our own retry logic & logging for event posts, don't use urllib3.Retry (#133) * remove support for indirect/patch and indirect/put * remove unused logic for individual flag/segment poll for indirect/patch * Ehaisley/84082/remove python2 (#136) * remove all references to six and remove queue fallback imports * remove NullHandler logger backwards compat * update circleci config to remove python 2.7 tests * remove ordereddict backwards compat * update setup.py to no longer list python 2.7 as compatible * no longer inherit from object for python 2 backwards compat * update readme and manifest to reflect python 2.7 removal * remove unicode type compatibility * remove 2.7 support from circleci * Allow authenticating with proxy This commit allows for authenticating with a proxy configured with the `http_proxy` environment variable. Authentication requires passing a header, and is not parsed by urllib3 from the proxy_url. * reimplement proxy tests for DRY and add test of proxy auth params * doc comment on auth params in proxy URL * add type hints to some of the public facing api. update some docs * Revert "add type hints to some of the public facing api." This reverts commit c35fa61. * Ehaisley/ch86857/type hints (#138) * add typehints to the public API * validate typehints in the public api and tests with mypy * remove all current deprecations (#139) * remove all currently deprecated classes, methods, arguments, and tests * also update semver usage to remove calls to deprecated functions and classes * remove global set_sdk_key, make SDK key required in Config (#140) * Removed the guides link * Pinning mypy and running it against different python versions (#141) * fix time zone mishandling that could make event debugging not work (#142) * fix 6.x build (#143) * fix time zone mishandling that could make event debugging not work (6.x) (#144) * prepare 6.13.3 release (#154) * Releasing version 6.13.3 * [ch99756] Add alias events (#145) * add support for experiment rollouts * fix unit test * address PR comments * use Releaser v2 config * Use newer docker images (#147) * Updates docs URLs * Add support for 3.10 (#150) * started work on FlagBuilder in as part of test data source implementation * finished FlagBuilder implementation and added FlagRuleBuilder implementation * added initial TestData interface and updated tests to not rely on test data internals * started data source implementation * changed FlagBuilder to public class; changed FlagBuilder attributes to be initialized in __init__ and eliminated use of try ... except: pass for handling empty attributes * (big segments 1) add public config/interface types * added implementation of test data source * docstring * formatting * ensure property doesn't return None * (big segments 2) implement evaluation, refactor eval logic & modules * linting * (big segments 3) implement big segment status tracking, wire up components * typing fixes * typing fixes * implement SSE contract tests * fix CI * fix CI again * fix CI * disable SSE tests in Python 3.5 * make test service port configurable * better SSE implementation that fixes linefeed and multi-byte char issues * fix constructor parameters in test service * comment * test improvements * rm obsolete default config logic * (big segments 4) implement big segment stores in Redis+DynamoDB, refactor db tests (#158) * converted ldclient.integrations module from file to directory; started moving public classes out of ldclient.impl.integrations.test_data* and instead into ldclient.integrations.test_data*; started adding TestData documentation * removed setup/teardown functions leftover from test scaffold * added TestData, FlagBuilder, and FlagRuleBuilder documentation; minor adjustments to implementation details * removed warning supression from TestData tests * fix big segments user hash algorithm to use SHA256 * update mypy version * updates to tests and related bug fixes * always cache Big Segment query result even if it's None * fix test assertion * lint * fix big segment ref format * fix big segments cache TTL being set to wrong value * fixed structure of fallthrough variation in result of FlagBuilder.build() * moved __test__ attribute into TestData class definition to prevent mypy from complaining about a missing class attribute * minor doc comment fix * Apply suggestions related to Sphinx docstring formatting from code review Co-authored-by: Eli Bishop <[email protected]> * fixed errors in the implementation of FlagBuilder's fallthrough_variation and off_variation when passing boolean variation values; updated tests to assert the expected behavior * added missing value_for_all_users() method to FlagBuilder class * Fix operator parsing errors (#169) * identify should not emit event if user key is empty (#164) * secondary should be treated as built-in attribute (#168) * URIs should have trailing slashes trimmed (#165) * all_flags_state should always include flag version (#166) * output event should not include a null prereqOf key (#167) * Account for traffic allocation on all flags (#171) * Add SDK contract tests (#170) * misc fixes to test data docs + add type hints * more type hints * remove some methods from the public test_data API * can't use "x|y" shortcut in typehints in older Pythons; use Union * fix misc type mistakes because I forgot to run the linter * update CONTRIBUTING.md and provide make targets * fixed a bug with flag rule clause builder internals; added unit test to verify rule evaluation * added ready argument to _TestDataSource class and indicated ready upon start to avoid delays in TestData initialization * Update contract tests to latest flask version (#176) Our contract tests depend on flask v1, which in turn depends on Jinja 2. Both of these are terribly dated and no longer supported. Jinja depends on markupsafe. markupsafe recently updated its code to no longer provide soft_unicode which in turn broke Jinja. Updating to the latest flask keeps all transitive dependencies better aligned and addresses this mismatch. * Adds link to Relay Proxy docs * Handle explicit None values in test payload (#179) The test harness may send explicit None values which should be treated the same as if the value was omitted entirely. * Fix "unhandled response" error in test harness (#180) When we return a `('', 204)` response from the flask handler, [Werkzeug intentionally removes the 'Content-Type' header][1], which causes the response to be created as a chunked response. The test harness is likely seeing a 204 response and isn't trying to read anything more from the stream. But since we are re-using connections, the next time it reads from the stream, it sees the `0\r\n\r\n` chunk and outputs an error: > 2022/04/20 14:23:39 Unsolicited response received on idle HTTP channel starting with "0\r\n\r\n"; err=<nil> Changing this response to 202 causes Werkzeug to return an empty response and silences the error. [1]: https://github.com/pallets/werkzeug/blob/560dd5f320bff318175f209595d42f5a80045417/src/werkzeug/wrappers/response.py#L540 * Exclude booleans when getting bucketable value (#181) When calculating a bucket, we get the bucketable value from the specified bucket by attribute. If this value is a string or an int, we can use it. Otherwise, we return None. Python considers a bool an instance of an int, which isn't what we want. So we need to add an explicit exclusion for this. * master -> main (#182) * Loosen restriction on expiringdict (#183) Originally this was pinned to a max version to deal with the incompatibility of Python 3.3 and the `typing` package. See [this PR][1]. Now that we now only support >=3.5, we can safely relax this restriction again. [1]: launchdarkly/python-server-sdk-private#120 * Fix mypy type checking (#184) A [customer requested][original-pr] that we start including a py.typed file in our repository. This would enable mypy to take advantage of our typehints. Unfortunately, this didn't completely solve the customers issue. A [second pr][second-pr] was opened to address the missing step of including the py.typed file in the `Manifest.in` file. However, this change alone is not sufficient. According to the [documentation][include_package_data], you must also include the `include_package_data=True` directive so that files specified in the `Manifest.in` file are included in distribution. [original-pr]: #166 [second-pr]: #172 [include_package_data]: https://setuptools.pypa.io/en/latest/userguide/datafiles.html#include-package-data * Add support for extra Redis connection parameters (#185) * Include wheel artifact when publishing package (#186) * skip tests that use a self-signed TLS cert in Python 3.7 * remove warn-level logging done for every Big Segments query (#190) * remove warn-level logging done for every Big Segments query * skip tests that use a self-signed TLS cert in Python 3.7 * update release metadata * Add application info support (#214) * Upgrade pip to fix failing CI build (#216) The CI build was failing because pip had an outdated list of available wheels for installation. Since it couldn't find a match, it was trying to build a package from source, which requires the rust compiler, which in turn isn't present on some of the docker images. By updating pip we get the updated list of available wheels, thereby allowing us to bypass source building and the need for the rust compiler entirely. --------- Co-authored-by: Eli Bishop <[email protected]> Co-authored-by: LaunchDarklyCI <[email protected]> Co-authored-by: Ben Woskow <[email protected]> Co-authored-by: Gavin Whelan <[email protected]> Co-authored-by: Elliot <[email protected]> Co-authored-by: Gabor Angeli <[email protected]> Co-authored-by: Elliot <[email protected]> Co-authored-by: Ben Woskow <[email protected]> Co-authored-by: LaunchDarklyCI <[email protected]> Co-authored-by: hroederld <[email protected]> Co-authored-by: Robert J. Neal <[email protected]> Co-authored-by: Robert J. Neal <[email protected]> Co-authored-by: Ember Stevens <[email protected]> Co-authored-by: ember-stevens <[email protected]> Co-authored-by: Matthew M. Keeler <[email protected]> Co-authored-by: charukiewicz <[email protected]> Co-authored-by: LaunchDarklyReleaseBot <[email protected]> Co-authored-by: Christian Charukiewicz <[email protected]> Co-authored-by: Matthew M. Keeler <[email protected]> * Releasing version 7.6.0 * Fix double spaces in changelog (#217) * Fix unsafe access to flag's trackEvents field (#218) When using the test data source, the trackEvents may not be required. Additionally, when receiving this information from LD, there is no guarantee that field will be provided. Instead of assuming that field always exists on the flag model, we should instead retrieve it with a default value to prevent index errors. * prepare 7.6.1 release (#195) * add 3.8 build * image name * fail on SyntaxWarning * typo * command syntax * pin expiringdict dependency for Python 3.3 compatibility * add Windows CircleCI job * periods are no longer valid in CircleCI job names * syntax fix * install Python in Windows * set path * move command * turn off debug logging * Py3 in Windows * config param * rm redundant step * choco switch * refactor Linux jobs using CircleCI 2.1 features * set log level before anything else * rm Azure config * use yaml.safe_load() to avoid code execution vulnerability in file data source * Initial work on wrapper_name, wrapper_version, diagnostic config options and start of diagnostic config event creation. * Python 2 compat changes. * More event generation code and starting to integrate tracking diagnostic values. * Add minimum diagnostic recording interval. Fix diagnostic.py to be importable. Add more diagnostic event fields. * don't let user fall outside of last bucket in rollout * fixing conditional logic * Add docstrings for diagnostic configuration options. * fix off-by-1 error * avoid redundant dict lookups * add unit tests for basic bucketing logic and edge case * Stream init tracking. Feeding of accumulator object through SDK. Various fixes. * Track events in last batch. * Fix sdk version field, some stylistic improvements. * Last of diagnostic configuration object fields. * Fill out rest of platform fields. * Cleanup and failed stream initialization tracking. * Add diagnostic config option test. * Add tests for diagnostics.py * Testing rest of diagnostic fields. * Test that streaming update processor records successful and unsuccessful connection attempts in the diagnostic accumulator when available. * Improvements to testability of event processor. * Rest of event processor tests. * Remove janky reflection. * Test change to filesource optional test requirements. * [ch61092] Add event payload ID on event requests. * normalize data store type and OS name in diagnostic events * gitignore * copyedit to diagnostic event config property comment * fix spurious error after sending diagnostic event * make verify_ssl=False turn off certificate verification too (#129) * add more TLS config options and collect HTTP/HTTPS config options in a class (#130) * make stream retry/backoff/jitter behavior consistent with other SDKs + improve testing (#131) * streams shouldn't use the same read timeout as the rest of the SDK (#132) * implement our own retry logic & logging for event posts, don't use urllib3.Retry (#133) * remove support for indirect/patch and indirect/put * remove unused logic for individual flag/segment poll for indirect/patch * Ehaisley/84082/remove python2 (#136) * remove all references to six and remove queue fallback imports * remove NullHandler logger backwards compat * update circleci config to remove python 2.7 tests * remove ordereddict backwards compat * update setup.py to no longer list python 2.7 as compatible * no longer inherit from object for python 2 backwards compat * update readme and manifest to reflect python 2.7 removal * remove unicode type compatibility * remove 2.7 support from circleci * Allow authenticating with proxy This commit allows for authenticating with a proxy configured with the `http_proxy` environment variable. Authentication requires passing a header, and is not parsed by urllib3 from the proxy_url. * reimplement proxy tests for DRY and add test of proxy auth params * doc comment on auth params in proxy URL * add type hints to some of the public facing api. update some docs * Revert "add type hints to some of the public facing api." This reverts commit c35fa61. * Ehaisley/ch86857/type hints (#138) * add typehints to the public API * validate typehints in the public api and tests with mypy * remove all current deprecations (#139) * remove all currently deprecated classes, methods, arguments, and tests * also update semver usage to remove calls to deprecated functions and classes * remove global set_sdk_key, make SDK key required in Config (#140) * Removed the guides link * Pinning mypy and running it against different python versions (#141) * fix time zone mishandling that could make event debugging not work (#142) * fix 6.x build (#143) * fix time zone mishandling that could make event debugging not work (6.x) (#144) * prepare 6.13.3 release (#154) * Releasing version 6.13.3 * [ch99756] Add alias events (#145) * add support for experiment rollouts * fix unit test * address PR comments * use Releaser v2 config * Use newer docker images (#147) * Updates docs URLs * Add support for 3.10 (#150) * started work on FlagBuilder in as part of test data source implementation * finished FlagBuilder implementation and added FlagRuleBuilder implementation * added initial TestData interface and updated tests to not rely on test data internals * started data source implementation * changed FlagBuilder to public class; changed FlagBuilder attributes to be initialized in __init__ and eliminated use of try ... except: pass for handling empty attributes * (big segments 1) add public config/interface types * added implementation of test data source * docstring * formatting * ensure property doesn't return None * (big segments 2) implement evaluation, refactor eval logic & modules * linting * (big segments 3) implement big segment status tracking, wire up components * typing fixes * typing fixes * implement SSE contract tests * fix CI * fix CI again * fix CI * disable SSE tests in Python 3.5 * make test service port configurable * better SSE implementation that fixes linefeed and multi-byte char issues * fix constructor parameters in test service * comment * test improvements * rm obsolete default config logic * (big segments 4) implement big segment stores in Redis+DynamoDB, refactor db tests (#158) * converted ldclient.integrations module from file to directory; started moving public classes out of ldclient.impl.integrations.test_data* and instead into ldclient.integrations.test_data*; started adding TestData documentation * removed setup/teardown functions leftover from test scaffold * added TestData, FlagBuilder, and FlagRuleBuilder documentation; minor adjustments to implementation details * removed warning supression from TestData tests * fix big segments user hash algorithm to use SHA256 * update mypy version * updates to tests and related bug fixes * always cache Big Segment query result even if it's None * fix test assertion * lint * fix big segment ref format * fix big segments cache TTL being set to wrong value * fixed structure of fallthrough variation in result of FlagBuilder.build() * moved __test__ attribute into TestData class definition to prevent mypy from complaining about a missing class attribute * minor doc comment fix * Apply suggestions related to Sphinx docstring formatting from code review Co-authored-by: Eli Bishop <[email protected]> * fixed errors in the implementation of FlagBuilder's fallthrough_variation and off_variation when passing boolean variation values; updated tests to assert the expected behavior * added missing value_for_all_users() method to FlagBuilder class * Fix operator parsing errors (#169) * identify should not emit event if user key is empty (#164) * secondary should be treated as built-in attribute (#168) * URIs should have trailing slashes trimmed (#165) * all_flags_state should always include flag version (#166) * output event should not include a null prereqOf key (#167) * Account for traffic allocation on all flags (#171) * Add SDK contract tests (#170) * misc fixes to test data docs + add type hints * more type hints * remove some methods from the public test_data API * can't use "x|y" shortcut in typehints in older Pythons; use Union * fix misc type mistakes because I forgot to run the linter * update CONTRIBUTING.md and provide make targets * fixed a bug with flag rule clause builder internals; added unit test to verify rule evaluation * added ready argument to _TestDataSource class and indicated ready upon start to avoid delays in TestData initialization * Update contract tests to latest flask version (#176) Our contract tests depend on flask v1, which in turn depends on Jinja 2. Both of these are terribly dated and no longer supported. Jinja depends on markupsafe. markupsafe recently updated its code to no longer provide soft_unicode which in turn broke Jinja. Updating to the latest flask keeps all transitive dependencies better aligned and addresses this mismatch. * Adds link to Relay Proxy docs * Handle explicit None values in test payload (#179) The test harness may send explicit None values which should be treated the same as if the value was omitted entirely. * Fix "unhandled response" error in test harness (#180) When we return a `('', 204)` response from the flask handler, [Werkzeug intentionally removes the 'Content-Type' header][1], which causes the response to be created as a chunked response. The test harness is likely seeing a 204 response and isn't trying to read anything more from the stream. But since we are re-using connections, the next time it reads from the stream, it sees the `0\r\n\r\n` chunk and outputs an error: > 2022/04/20 14:23:39 Unsolicited response received on idle HTTP channel starting with "0\r\n\r\n"; err=<nil> Changing this response to 202 causes Werkzeug to return an empty response and silences the error. [1]: https://github.com/pallets/werkzeug/blob/560dd5f320bff318175f209595d42f5a80045417/src/werkzeug/wrappers/response.py#L540 * Exclude booleans when getting bucketable value (#181) When calculating a bucket, we get the bucketable value from the specified bucket by attribute. If this value is a string or an int, we can use it. Otherwise, we return None. Python considers a bool an instance of an int, which isn't what we want. So we need to add an explicit exclusion for this. * master -> main (#182) * Loosen restriction on expiringdict (#183) Originally this was pinned to a max version to deal with the incompatibility of Python 3.3 and the `typing` package. See [this PR][1]. Now that we now only support >=3.5, we can safely relax this restriction again. [1]: launchdarkly/python-server-sdk-private#120 * Fix mypy type checking (#184) A [customer requested][original-pr] that we start including a py.typed file in our repository. This would enable mypy to take advantage of our typehints. Unfortunately, this didn't completely solve the customers issue. A [second pr][second-pr] was opened to address the missing step of including the py.typed file in the `Manifest.in` file. However, this change alone is not sufficient. According to the [documentation][include_package_data], you must also include the `include_package_data=True` directive so that files specified in the `Manifest.in` file are included in distribution. [original-pr]: #166 [second-pr]: #172 [include_package_data]: https://setuptools.pypa.io/en/latest/userguide/datafiles.html#include-package-data * Add support for extra Redis connection parameters (#185) * Include wheel artifact when publishing package (#186) * skip tests that use a self-signed TLS cert in Python 3.7 * remove warn-level logging done for every Big Segments query (#190) * remove warn-level logging done for every Big Segments query * skip tests that use a self-signed TLS cert in Python 3.7 * update release metadata * Add application info support (#214) * Upgrade pip to fix failing CI build (#216) The CI build was failing because pip had an outdated list of available wheels for installation. Since it couldn't find a match, it was trying to build a package from source, which requires the rust compiler, which in turn isn't present on some of the docker images. By updating pip we get the updated list of available wheels, thereby allowing us to bypass source building and the need for the rust compiler entirely. * Fix double spaces in changelog (#217) * Fix unsafe access to flag's trackEvents field (#218) When using the test data source, the trackEvents may not be required. Additionally, when receiving this information from LD, there is no guarantee that field will be provided. Instead of assuming that field always exists on the flag model, we should instead retrieve it with a default value to prevent index errors. --------- Co-authored-by: Eli Bishop <[email protected]> Co-authored-by: LaunchDarklyCI <[email protected]> Co-authored-by: Ben Woskow <[email protected]> Co-authored-by: Gavin Whelan <[email protected]> Co-authored-by: Elliot <[email protected]> Co-authored-by: Gabor Angeli <[email protected]> Co-authored-by: Elliot <[email protected]> Co-authored-by: Ben Woskow <[email protected]> Co-authored-by: LaunchDarklyCI <[email protected]> Co-authored-by: hroederld <[email protected]> Co-authored-by: Robert J. Neal <[email protected]> Co-authored-by: Robert J. Neal <[email protected]> Co-authored-by: Ember Stevens <[email protected]> Co-authored-by: ember-stevens <[email protected]> Co-authored-by: Matthew M. Keeler <[email protected]> Co-authored-by: charukiewicz <[email protected]> Co-authored-by: LaunchDarklyReleaseBot <[email protected]> Co-authored-by: Christian Charukiewicz <[email protected]> Co-authored-by: Matthew M. Keeler <[email protected]> * Releasing version 7.6.1 * docs: Fix sphinx documentation failures (#219) Our documentation file had a configuration issue, which emits a warning when building with sphinx. This caused the build process on readthedocs.org to fail because they treat warnings as errors. We didn't catch this because: 1. Warnings are treated like warnings when generating documentation locally. 2. We don't try to generate the documentation as part of our CI process. This commit should resolve both the underlying issue and the deficiencies that prevented us from catching this initially. * Fix documentation generation (#197) (#220) * fix: Restrict urllib to <2.0.0 (#202) * ci: Generate coverage as part of standard CI (#222) --------- Co-authored-by: Elliot <[email protected]> Co-authored-by: Elliot <[email protected]> Co-authored-by: Eli Bishop <[email protected]> Co-authored-by: LaunchDarklyCI <[email protected]> Co-authored-by: Ben Woskow <[email protected]> Co-authored-by: LaunchDarklyCI <[email protected]> Co-authored-by: hroederld <[email protected]> Co-authored-by: Robert J. Neal <[email protected]> Co-authored-by: Robert J. Neal <[email protected]> Co-authored-by: Ember Stevens <[email protected]> Co-authored-by: ember-stevens <[email protected]> Co-authored-by: Matthew M. Keeler <[email protected]> Co-authored-by: charukiewicz <[email protected]> Co-authored-by: LaunchDarklyReleaseBot <[email protected]> Co-authored-by: Christian Charukiewicz <[email protected]> Co-authored-by: Matthew M. Keeler <[email protected]> Co-authored-by: Ben Woskow <[email protected]> Co-authored-by: Gavin Whelan <[email protected]> Co-authored-by: Gabor Angeli <[email protected]> Co-authored-by: Louis Chan <[email protected]> Co-authored-by: prpnmac <[email protected]>
Is your feature request related to a problem? Please describe.
The SDK is currently launching 11 threads by default when doing
ldclient.set_config(Config(sdk_key)); ldclient.get()
. This is resource intensive in python and costing me more than 1Gb of memory per instance of my service that leverages the SDK and is expensive when running a large number of services.Describe the solution you'd like
The SDK should use a single thread by default or at least make it easy to configure it to only use a single thread.
Describe alternatives you've considered
I've monkey patched the SDK to reduce the number of threads consumed, but will eventually have to abandon the SDK in favor of directly accessing the APIs.
The text was updated successfully, but these errors were encountered: