Use statsd client instead of Promotheus (#160)

* Use statsd client instead of Prometheus

* Remove mentions of Prometheus (unused in production)

* Add statsd to mypy ignores

* Set container entrypoint back to /bin/sh (default)

Author: leplatrem

Dockerfile

@@ -22,8 +22,7 @@ RUN poetry install --no-dev --no-root
 
 # `production` stage uses the dependencies downloaded in the `base` stage
 FROM python:3.10.5-slim as production
-ENV PROMETHEUS_MULTIPROC=1 \
-    PORT=8000 \
+ENV PORT=8000 \
     PYTHONUNBUFFERED=1 \
     PYTHONDONTWRITEBYTECODE=1 \
     VIRTUAL_ENV=/opt/pysetup/.venv
@@ -42,5 +41,4 @@ WORKDIR /app
 COPY . .
 
 EXPOSE $PORT
-ENTRYPOINT ["bin/docker-entrypoint.sh"]
 CMD ["gunicorn", "-k", "uvicorn.workers.UvicornWorker", "-c", "config/gunicorn_conf.py", "src.app.api:app"]

README.md

@@ -221,12 +221,12 @@ GET /whiteboard_tags/
 
 ## Metrics
 
-The following metrics are sent via Prometheus:
-
-- `jbi_bugzilla_ignored_total`
-- `jbi_bugzilla_processed_total`
-- `jbi_action_execution_milliseconds`
-- `jbi_jira_methods_milliseconds`
-- `jbi_jira_methods_total`
-- `jbi_bugzilla_methods_total`
-- `jbi_bugzilla_methods_milliseconds`
+The following metrics are sent via StatsD:
+
+- `jbi.bugzilla.ignored.count`
+- `jbi.bugzilla.processed.count`
+- `jbi.action.execution.timer`
+- `jbi.jira.methods.*.count`
+- `jbi.jira.methods.*.timer`
+- `jbi.bugzilla.methods.*.count`
+- `jbi.bugzilla.methods.*.timer`

bin/docker-entrypoint.sh

@@ -6,7 +6,6 @@
 import multiprocessing
 from typing import Optional
 
-from prometheus_client import multiprocess
 from pydantic import BaseSettings, conint, root_validator, validator
 
 
@@ -77,7 +76,3 @@ def set_workers(cls, values):
     "host": gunicorn_settings.host,
     "port": gunicorn_settings.port,
 }
-
-
-def child_exit(server, worker):  # pylint: disable=missing-function-docstring
-    multiprocess.mark_process_dead(worker.pid)

config/gunicorn_conf.py

@@ -11,14 +11,14 @@ fastapi = "^0.73.0"
 pydantic = {version = "^1.9.1", extras = ["dotenv", "email"]}
 uvicorn = {extras = ["standard"], version = "^0.18.2"}
 gunicorn = "^20.1.0"
-prometheus-client = "^0.14.1"
 python-bugzilla = "^3.2.0"
 atlassian-python-api = "^3.25.0"
 dockerflow = "2022.7.0"
 Jinja2 = "^3.1.2"
 pydantic-yaml = {extras = ["pyyaml","ruamel"], version = "^0.6.1"}
 sentry-sdk = "^1.8.0"
 backoff = "^2.1.2"
+statsd = "^3.3.0"
 
 
 [tool.poetry.dev-dependencies]
@@ -74,7 +74,7 @@ python_version = "3.10"
 warn_return_any = true
 
 [[tool.mypy.overrides]]
-module = ["ruamel", "bugzilla", "atlassian"]
+module = ["ruamel", "bugzilla", "atlassian", "statsd.defaults.env"]
 ignore_missing_imports = true
 
 [tool.coverage]

poetry.lock

@@ -3,7 +3,7 @@
 """
 import logging
 
-from prometheus_client import Counter, Summary
+from statsd.defaults.env import statsd
 
 from src.app.environment import Settings
 from src.jbi.bugzilla import BugzillaBug, BugzillaWebhookRequest
@@ -12,14 +12,6 @@
 
 logger = logging.getLogger(__name__)
 
-counter_ignored = Counter("jbi_bugzilla_ignored_total", "Bugzilla WebHooks ignored")
-counter_processed = Counter(
-    "jbi_bugzilla_processed_total", "Bugzilla WebHooks processed"
-)
-action_execution_timer = Summary(
-    "jbi_action_execution_milliseconds", "Action execution duration"
-)
-
 
 class Operations:
     """Track status of incoming requests in log entries."""
@@ -30,7 +22,7 @@ class Operations:
     SUCCESS = "success"
 
 
-@action_execution_timer.time()
+@statsd.timer("jbi.action.execution.timer")
 def execute_action(
     request: BugzillaWebhookRequest,
     actions: Actions,
@@ -96,13 +88,13 @@ def execute_action(
             bug_obj.id,
             extra={"operation": Operations.SUCCESS, **log_context},
         )
-        counter_processed.inc()
+        statsd.incr("jbi.bugzilla.processed.count")
         return content
     except IgnoreInvalidRequestError as exception:
         logger.debug(
             "Ignore incoming request: %s",
             exception,
             extra={"operation": Operations.IGNORE, **log_context},
         )
-        counter_ignored.inc()
+        statsd.incr("jbi.bugzilla.ignored.count")
         raise

pyproject.toml

@@ -5,7 +5,7 @@
 import backoff
 import bugzilla as rh_bugzilla
 from atlassian import Jira, errors
-from prometheus_client import Counter, Summary
+from statsd.defaults.env import statsd
 
 from src.app import environment
 from src.jbi.models import Actions
@@ -24,29 +24,12 @@ class InstrumentedClient:
     It retries the methods if the specified exceptions are raised.
     """
 
-    counters: Dict[str, Counter] = {}
-    timers: Dict[str, Counter] = {}
-
     def __init__(self, wrapped, prefix, methods, exceptions):
         self.wrapped = wrapped
+        self.prefix = prefix
         self.methods = methods
         self.exceptions = exceptions
 
-        # We have a single instrument per prefix. Methods are reported as labels.
-        counter_name = prefix + "_methods_total"
-        if counter_name not in self.counters:
-            self.counters[counter_name] = Counter(
-                counter_name, f"{prefix} method calls", labelnames=["method"]
-            )
-        self.counter = self.counters[counter_name]
-
-        timer_name = prefix + "_methods_milliseconds"
-        if timer_name not in self.timers:
-            self.timers[timer_name] = Summary(
-                timer_name, f"{prefix} method timing", labelnames=["method"]
-            )
-        self.timer = self.timers[timer_name]
-
     def __getattr__(self, attr):
         if attr not in self.methods:
             return getattr(self.wrapped, attr)
@@ -58,9 +41,9 @@ def __getattr__(self, attr):
         )
         def wrapped_func(*args, **kwargs):
             # Increment the call counter.
-            self.counter.labels(method=attr).inc()
+            statsd.incr(f"jbi.{self.prefix}.methods.{attr}.count")
             # Time its execution.
-            with self.timer.labels(method=attr).time():
+            with statsd.timer(f"jbi.{self.prefix}.methods.{attr}.timer"):
                 return getattr(self.wrapped, attr)(*args, **kwargs)
 
         # The method was not called yet.
@@ -83,7 +66,7 @@ def get_jira():
     )
     return InstrumentedClient(
         wrapped=jira_client,
-        prefix="jbi_jira",
+        prefix="jira",
         methods=instrumented_methods,
         exceptions=(errors.ApiError,),
     )
@@ -108,7 +91,7 @@ def get_bugzilla():
     )
     return InstrumentedClient(
         wrapped=bugzilla_client,
-        prefix="jbi_bugzilla",
+        prefix="bugzilla",
         methods=instrumented_methods,
         exceptions=(rh_bugzilla.BugzillaError,),
     )

src/jbi/runner.py

@@ -140,25 +140,25 @@ def test_counter_is_incremented_on_ignored_requests(
     assert webhook_create_example.bug
     webhook_create_example.bug.whiteboard = "foo"
 
-    with mock.patch("src.jbi.runner.counter_ignored") as mocked:
+    with mock.patch("src.jbi.runner.statsd") as mocked:
         with pytest.raises(IgnoreInvalidRequestError):
             execute_action(
                 request=webhook_create_example,
                 actions=actions_example,
                 settings=settings,
             )
-    assert mocked.inc.called, "ignored counter was called"
+    mocked.incr.assert_called_with("jbi.bugzilla.ignored.count")
 
 
 def test_counter_is_incremented_on_processed_requests(
     webhook_create_example: BugzillaWebhookRequest,
     actions_example: Actions,
     settings: Settings,
 ):
-    with mock.patch("src.jbi.runner.counter_processed") as mocked:
+    with mock.patch("src.jbi.runner.statsd") as mocked:
         execute_action(
             request=webhook_create_example,
             actions=actions_example,
             settings=settings,
         )
-    assert mocked.inc.called, "processed counter was called"
+    mocked.incr.assert_called_with("jbi.bugzilla.processed.count")

src/jbi/services.py

@@ -10,42 +10,30 @@
 
 
 def test_counter_is_incremented_on_jira_create_issue():
-    with mock.patch("src.jbi.services.Jira.create_issue"):
-        jira_client = get_jira()
+    jira_client = get_jira()
 
-        labelled_counter = jira_client.counters["jbi_jira_methods_total"].labels(
-            method="create_issue"
-        )
-        with mock.patch.object(labelled_counter, "inc") as mocked:
-            jira_client.create_issue({})
+    with mock.patch("src.jbi.services.statsd") as mocked:
+        jira_client.create_issue({})
 
-    assert mocked.called, "Counter was incremented on create_issue()"
+    mocked.incr.assert_called_with("jbi.jira.methods.create_issue.count")
 
 
 def test_timer_is_used_on_jira_create_issue():
-    with mock.patch("src.jbi.services.Jira.create_issue"):
-        jira_client = get_jira()
+    jira_client = get_jira()
 
-        labelled_timer = jira_client.timers["jbi_jira_methods_milliseconds"].labels(
-            method="create_issue"
-        )
-        with mock.patch.object(labelled_timer, "time") as mocked:
-            jira_client.create_issue({})
+    with mock.patch("src.jbi.services.statsd") as mocked:
+        jira_client.create_issue({})
 
-    assert mocked.called, "Timer was used on create_issue()"
+    mocked.timer.assert_called_with("jbi.jira.methods.create_issue.timer")
 
 
 def test_timer_is_used_on_bugzilla_getcomments():
-    with mock.patch("src.jbi.services.rh_bugzilla.Bugzilla"):
-        bugzilla_client = get_bugzilla()
+    bugzilla_client = get_bugzilla()
 
-        labelled_timer = bugzilla_client.timers[
-            "jbi_bugzilla_methods_milliseconds"
-        ].labels(method="get_comments")
-        with mock.patch.object(labelled_timer, "time") as mocked:
-            bugzilla_client.get_comments([])
+    with mock.patch("src.jbi.services.statsd") as mocked:
+        bugzilla_client.get_comments([])
 
-    assert mocked.called, "Timer was used on get_comments()"
+    mocked.timer.assert_called_with("jbi.bugzilla.methods.get_comments.timer")
 
 
 def test_bugzilla_methods_are_retried_if_raising():