twitter/20190118
ttim
tagged this
18 Jan 18:49
We found that when a lot of sink writes happen concurrently through `Execution`s using `JobTest` you can observe data race for `Buffer`s of tuples created. I've created a test to show this - if you do writes to 100 sinks in parallel it breaks internals of mutable map where we store tuple buffers for each mocked source and make test infinite. I've added a lock to fix this race condition. This also includes fix for Ruby's version on CI.