bot: Replace retry function with Tenacity (#366)

Fixes #110

Author: seb-sojka

.isort.cfg

@@ -1,6 +1,6 @@
 [settings]
 known_first_party = code_coverage_backend,code_coverage_bot,code_coverage_events,code_coverage_tools,conftest,firefox_code_coverage
-known_third_party = connexion,datadog,dateutil,fakeredis,flask,flask_cors,flask_talisman,google,hglib,jsone,jsonschema,libmozdata,libmozevent,logbook,pytest,pytz,raven,redis,requests,responses,setuptools,structlog,taskcluster,werkzeug,zstandard
+known_third_party = connexion,datadog,dateutil,fakeredis,flask,flask_cors,flask_talisman,google,hglib,jsone,jsonschema,libmozdata,libmozevent,logbook,pytest,pytz,raven,redis,requests,responses,setuptools,structlog,taskcluster,tenacity,werkzeug,zstandard
 force_single_line = True
 default_section=FIRSTPARTY
 line_length=159

bot/code_coverage_bot/taskcluster.py

@@ -7,10 +7,9 @@
 import requests
 import structlog
 import taskcluster
+import tenacity
 from taskcluster.helper import TaskclusterConfig
 
-from code_coverage_bot.utils import retry
-
 logger = structlog.getLogger(__name__)
 taskcluster_config = TaskclusterConfig("https://firefox-ci-tc.services.mozilla.com")
 
@@ -72,6 +71,9 @@ def download_artifact(artifact_path, task_id, artifact_name):
     url = queue.buildUrl("getLatestArtifact", task_id, artifact_name)
     logger.debug("Downloading artifact", url=url)
 
+    @tenacity.retry(
+        reraise=True, wait=tenacity.wait_fixed(30), stop=tenacity.stop_after_attempt(5)
+    )
     def perform_download():
         r = requests.get(url, stream=True)
         r.raise_for_status()
@@ -83,7 +85,7 @@ def perform_download():
         if artifact_path.endswith(".zip") and not is_zipfile(artifact_path):
             raise BadZipFile("File is not a zip file")
 
-    retry(perform_download)
+    perform_download()
 
 
 def is_coverage_task(task):

bot/code_coverage_bot/uploader.py

@@ -4,10 +4,10 @@
 
 import requests
 import structlog
+import tenacity
 import zstandard as zstd
 
 from code_coverage_bot.secrets import secrets
-from code_coverage_bot.utils import retry
 from code_coverage_tools.gcp import get_bucket
 
 logger = structlog.get_logger(__name__)
@@ -45,11 +45,7 @@ def gcp(repository, revision, report, platform, suite):
     logger.info("Uploaded {} on {}".format(path, bucket))
 
     # Trigger ingestion on backend
-    retry(
-        lambda: gcp_ingest(repository, revision, platform, suite),
-        retries=10,
-        wait_between_retries=60,
-    )
+    gcp_ingest(repository, revision, platform, suite)
 
     return blob
 
@@ -66,6 +62,9 @@ def gcp_covdir_exists(repository, revision, platform, suite):
     return blob.exists()
 
 
+@tenacity.retry(
+    stop=tenacity.stop_after_attempt(10), wait=tenacity.wait_fixed(60), reraise=True
+)
 def gcp_ingest(repository, revision, platform, suite):
     """
     The GCP report ingestion is triggered remotely on a backend

bot/code_coverage_bot/utils.py

@@ -4,34 +4,12 @@
 # file, You can obtain one at http://mozilla.org/MPL/2.0/.
 import concurrent.futures
 import subprocess
-import time
 
 import structlog
 
 log = structlog.get_logger(__name__)
 
 
-class RunException(Exception):
-    """
-    Exception used to stop retrying
-    """
-
-
-def retry(
-    operation, retries=5, wait_between_retries=30, exception_to_break=RunException
-):
-    while True:
-        try:
-            return operation()
-        except Exception as e:
-            if isinstance(e, exception_to_break):
-                raise
-            retries -= 1
-            if retries == 0:
-                raise
-            time.sleep(wait_between_retries)
-
-
 def hide_secrets(text, secrets):
     if type(text) is bytes:
         encode_secret, xxx = lambda x: bytes(x, encoding="utf-8"), b"XXX"
@@ -54,7 +32,7 @@ def run_check(command, **kwargs):
     assert isinstance(command, list)
 
     if len(command) == 0:
-        raise RunException("Can't run an empty command.")
+        raise Exception("Can't run an empty command.")
 
     _kwargs = dict(
         stdin=subprocess.DEVNULL,  # no interactions
@@ -81,7 +59,7 @@ def run_check(command, **kwargs):
             error=error,
         )
 
-        raise RunException(f"`{command[0]}` failed with code: {proc.returncode}.")
+        raise Exception(f"`{command[0]}` failed with code: {proc.returncode}.")
 
     return output
 

bot/requirements.txt

@@ -3,4 +3,5 @@ google-cloud-storage==1.25.0
 libmozdata==0.1.64
 pytoml==0.1.21
 pytz==2019.3
+tenacity==6.0.0
 zstandard==0.13.0

bot/tests/test_taskcluster.py

@@ -275,7 +275,7 @@ def test_download_artifact_forbidden(mocked_sleep, mock_taskcluster, tmpdir):
             "public/test_info/code-coverage-grcov.zip",
         )
 
-    assert mocked_sleep.call_count == 4
+    assert len(responses.calls) == 5
 
 
 @mock.patch("time.sleep")
@@ -295,4 +295,4 @@ def test_download_artifact_badzip(mocked_sleep, mock_taskcluster, tmpdir):
             "public/test_info/code-coverage-grcov.zip",
         )
 
-    assert mocked_sleep.call_count == 4
+    assert len(responses.calls) == 5