Update kubernetes probes (#50)

* Update kubernetes probes and create version endpoint.

* Remove hot version test.

* Fix version endpoint.

biothings_annotator/application/views/__init__.py

@@ -1,4 +1,4 @@
-from .annotator import StatusView, BatchCurieView, CurieView, TrapiView, CurieLegacyView, TrapiLegacyView
+from .annotator import StatusView, VersionView, BatchCurieView, CurieView, TrapiView, CurieLegacyView, TrapiLegacyView
 
 
 def build_routes() -> list[dict]:
@@ -20,6 +20,13 @@ def build_routes() -> list[dict]:
         "name": "status_endpoint",
     }
 
+    # --- VERSION ROUTE ---
+    version_route_main = {
+        "handler": VersionView.as_view(),
+        "uri": r"/version",
+        "name": "version_endpoint",
+    }
+
     # --- CURIE ROUTES ---
     curie_route_main = {
         "handler": CurieView.as_view(),
@@ -45,6 +52,7 @@ def build_routes() -> list[dict]:
 
     route_collection = [
         status_route_main,
+        version_route_main,
         curie_route_main,
         curie_route_mirror,
         batch_curie_route,

biothings_annotator/application/views/annotator.py

@@ -4,10 +4,11 @@
 
 import logging
 import json
+import pathlib
+import git
 
 import sanic
 from sanic.views import HTTPMethodView
-from sanic.exceptions import SanicException
 from sanic.request import Request
 from sanic import response
 
@@ -43,6 +44,36 @@ async def get(self, _: Request):
             return sanic.json(result)
 
 
+class VersionView(HTTPMethodView):
+    def __init__(self):
+        super().__init__()
+        application = sanic.Sanic.get_app()
+        cache = application.config.CACHE_MAX_AGE
+        self.default_headers = {"Cache-Control": f"max-age={cache}, public"}
+
+    async def get(self, _: Request):
+        try:
+            # Resolve the absolute path to the current file
+            file_path = pathlib.Path(__file__).resolve()
+
+            # Use git.Repo to find the root of the repository
+            repo = git.Repo(file_path, search_parent_directories=True)
+
+            if repo.bare:
+                repo_dir = repo.working_tree_dir
+                logger.error(f"Git repository not found in directory: {repo_dir}")
+                result = {"version": "Unknown"}
+                return sanic.json(result)
+
+            commit_hash = repo.head.commit.hexsha  # Get the latest commit hash
+            result = {"version": commit_hash}
+            return sanic.json(result, headers=self.default_headers)
+        except Exception as exc:
+            logger.error(f"Error getting GitHub commit hash: {exc}")
+            result = {"version": "Unknown"}
+            return sanic.json(result)
+
+
 class CurieView(HTTPMethodView):
     def __init__(self):
         super().__init__()

deploy/templates/deployment.yaml

@@ -6,7 +6,12 @@ metadata:
     {{- include "biothings-annotator.labels" . | nindent 4 }}
 spec:
   strategy:
-        type: {{ .Values.deployment.strategyType }}
+    type: {{ .Values.deployment.strategy.type }}
+    {{- if eq .Values.deployment.strategy.type "RollingUpdate" }}
+    rollingUpdate:
+      maxUnavailable: {{ .Values.deployment.strategy.rollingUpdate.maxUnavailable | default 1 }}
+      maxSurge: {{ .Values.deployment.strategy.rollingUpdate.maxSurge | default 1 }}
+    {{- end }}
   replicas: {{ .Values.deployment.replicaCount }}
   selector:
     matchLabels:
@@ -34,14 +39,33 @@ spec:
             - name: http
               containerPort: {{ .Values.containers.port }}
               protocol: TCP
-          livenessProbe:
+          startupProbe:  # To determine if a container application has started successfully.
             httpGet:
-              path: /
-              port: http
-          readinessProbe:
+              path: /status
+              port: {{ .Values.containers.port }}
+            initialDelaySeconds: 10  # The number of seconds to wait after the container has started before performing the first startup probe.
+            periodSeconds: 15  # How often (in seconds) to perform the startup probe.
+            timeoutSeconds: 10  # The number of seconds after which the probe times out.
+            successThreshold: 1  # The number of consecutive successes required to consider the container started successfully.
+            failureThreshold: 5  # The number of consecutive failures required to consider the container startup to have failed.
+          readinessProbe:  # To determine when the container is ready to start accepting traffic
             httpGet:
-              path: /
-              port: http
+              path: /status
+              port: {{ .Values.containers.port }}
+            initialDelaySeconds: 20  #  The number of seconds to wait after the container has started before performing the first readiness probe.
+            periodSeconds: 20  # How often (in seconds) to perform the readiness probe.
+            timeoutSeconds: 10  # The number of seconds after which the probe times out.
+            successThreshold: 1  # The number of consecutive successes required to consider the container ready after it has been failing.
+            failureThreshold: 3  # The number of consecutive failures required to consider the container not ready.
+          livenessProbe:  # To determine if a container is still running
+            httpGet:
+              path: /status
+              port: {{ .Values.containers.port }}
+            initialDelaySeconds: 60  # The number of seconds to wait after the container has started before performing the first liveness probe.
+            periodSeconds: 60  # How often (in seconds) to perform the liveness probe.
+            timeoutSeconds: 30  # The number of seconds after which the probe times out.
+            successThreshold: 1  # The number of consecutive successes required to consider the container healthy after it has been failing.
+            failureThreshold: 3  # The number of consecutive failures required to consider the container unhealthy and restart it.
       {{- with .Values.affinity }}
       affinity:
         {{- toYaml . | nindent 8 }}

deploy/values.yaml

@@ -23,33 +23,6 @@ containers:
   name: biothingsannotator
   es_host: ES_HOST_VALUE
   port: 9000
-  startupProbe:  # To determine if a container application has started successfully.
-    httpGet:
-      path: /status
-      port: 9000
-    initialDelaySeconds: 10  # The number of seconds to wait after the container has started before performing the first startup probe.
-    periodSeconds: 15  # How often (in seconds) to perform the startup probe.
-    timeoutSeconds: 10  # The number of seconds after which the probe times out.
-    successThreshold: 1  # The number of consecutive successes required to consider the container started successfully.
-    failureThreshold: 5  # The number of consecutive failures required to consider the container startup to have failed.
-  readinessProbe:  # To determine when the container is ready to start accepting traffic
-    httpGet:
-      path: /status
-      port: 9000
-    initialDelaySeconds: 20  #  The number of seconds to wait after the container has started before performing the first readiness probe.
-    periodSeconds: 20  # How often (in seconds) to perform the readiness probe.
-    timeoutSeconds: 10  # The number of seconds after which the probe times out.
-    successThreshold: 1  # The number of consecutive successes required to consider the container ready after it has been failing.
-    failureThreshold: 3  # The number of consecutive failures required to consider the container not ready.
-  livenessProbe:  # To determine if a container is still running
-    httpGet:
-      path: /status
-      port: 9000
-    initialDelaySeconds: 60  # The number of seconds to wait after the container has started before performing the first liveness probe.
-    periodSeconds: 60  # How often (in seconds) to perform the liveness probe.
-    timeoutSeconds: 30  # The number of seconds after which the probe times out.
-    successThreshold: 1  # The number of consecutive successes required to consider the container healthy after it has been failing.
-    failureThreshold: 3  # The number of consecutive failures required to consider the container unhealthy and restart it.
 
 env:
   OPENTELEMETRY_ENABLED_VALUE: True

pyproject.toml

@@ -42,6 +42,7 @@ dependencies = [
     "brotli >= 1.1.0",
     "biothings_client >= 0.3.1",
     "sanic[ext] >= 23.12.1",
+    "GitPython >= 3.1.43",
 ]
 
 [project.optional-dependencies]

tests/test_application_endpoints.py

@@ -1,12 +1,12 @@
-import json
-
 import pytest
+import git
+import re
 import sanic
 from typing import Union
 
 from biothings_annotator import utils
 from biothings_annotator.annotator import Annotator
-from unittest.mock import patch
+from unittest.mock import patch, PropertyMock
 
 
 @pytest.mark.parametrize("endpoint", ["/status/"])
@@ -84,6 +84,88 @@ def test_status_get_failed_data_check(test_annotator: sanic.Sanic, endpoint: str
         assert response.json == expected_response_body
 
 
+@pytest.mark.parametrize("endpoint", ["/version/"])
+def test_version_get_success(test_annotator: sanic.Sanic, endpoint: str):
+    """
+    Tests the Version endpoint GET for a successful case
+    Mocking git.Repo to return a valid commit hash
+    """
+    with patch.object(git.Repo, "head", create=True) as mock_repo_head, \
+         patch.object(git.Repo, "bare", new_callable=PropertyMock, return_value=False), \
+         patch.object(git.Repo, "__init__", lambda *args, **kwargs: None):
+
+        mock_repo_head.commit.hexsha = "abc123"
+
+        request, response = test_annotator.test_client.request(endpoint, http_method="get")
+
+        assert request.method == "GET"
+        assert request.query_string == ""
+        assert request.scheme == "http"
+        assert request.server_path == endpoint
+
+        expected_response_body = {"version": "abc123"}
+        assert response.http_version == "HTTP/1.1"
+        assert response.content_type == "application/json"
+        assert response.is_success
+        assert not response.is_error
+        assert response.is_closed
+        assert response.status_code == 200
+        assert response.encoding == "utf-8"
+        assert response.json == expected_response_body
+
+
+@pytest.mark.parametrize("endpoint", ["/version/"])
+def test_version_get_bare_repo(test_annotator: sanic.Sanic, endpoint: str):
+    """
+    Tests the Version endpoint GET when the repository is bare (missing repository)
+    Mocking git.Repo to simulate a bare repository
+    """
+    with patch.object(git.Repo, "bare", new_callable=PropertyMock, return_value=True), \
+         patch.object(git.Repo, "__init__", lambda *args, **kwargs: None):
+
+        request, response = test_annotator.test_client.request(endpoint, http_method="get")
+
+        assert request.method == "GET"
+        assert request.query_string == ""
+        assert request.scheme == "http"
+        assert request.server_path == endpoint
+
+        expected_response_body = {"version": "Unknown"}
+        assert response.http_version == "HTTP/1.1"
+        assert response.content_type == "application/json"
+        assert response.is_success
+        assert not response.is_error
+        assert response.is_closed
+        assert response.status_code == 200
+        assert response.encoding == "utf-8"
+        assert response.json == expected_response_body
+
+
+@pytest.mark.parametrize("endpoint", ["/version/"])
+def test_version_get_exception(test_annotator: sanic.Sanic, endpoint: str):
+    """
+    Tests the Version endpoint GET when an Exception is raised
+    Mocking git.Repo to raise an exception
+    """
+    with patch.object(git.Repo, "__init__", side_effect=Exception("Simulated error")):
+        request, response = test_annotator.test_client.request(endpoint, http_method="get")
+
+        assert request.method == "GET"
+        assert request.query_string == ""
+        assert request.scheme == "http"
+        assert request.server_path == endpoint
+
+        expected_response_body = {"version": "Unknown"}
+        assert response.http_version == "HTTP/1.1"
+        assert response.content_type == "application/json"
+        assert response.is_success
+        assert not response.is_error
+        assert response.is_closed
+        assert response.status_code == 200
+        assert response.encoding == "utf-8"
+        assert response.json == expected_response_body
+
+
 def test_curie_get(test_annotator: sanic.Sanic):
     """
     Tests the CURIE endpoint GET