async def _get(
cls,
redis: redis_utils.RedisCache,
api_access_key: str,
api_secret_key: str,
account_scope: typing.Optional[github_types.GitHubLogin],
) -> "ApplicationSaas":
cached_application = await cls._retrieve_from_cache(
redis, api_access_key, api_secret_key, account_scope)
if cached_application is None or cached_application._has_expired():
try:
db_application = await cls._retrieve_from_db(
redis, api_access_key, api_secret_key, account_scope)
except http.HTTPForbidden:
# api key is valid, but not the scope
raise ApplicationUserNotFound()
except http.HTTPNotFound:
raise ApplicationUserNotFound()
except Exception as exc:
if cached_application is not None and (
exceptions.should_be_ignored(exc)
or exceptions.need_retry(exc)):
# NOTE(sileht): return the cached application, instead of
# retrying the stream, just because the dashboard has a
# connectivity issue.
return cached_application
raise
await db_application.save_to_cache()
return db_application
return cached_application
async def exec_action(
method_name: typing.Literal["run", "cancel"],
rule: rules.EvaluatedRule,
action: str,
ctxt: context.Context,
) -> check_api.Result:
try:
if method_name == "run":
method = rule.actions[action].run
elif method_name == "cancel":
method = rule.actions[action].cancel
else:
raise RuntimeError("wrong method_name")
return await method(ctxt, rule)
except Exception as e: # pragma: no cover
# Forward those to worker
if exceptions.should_be_ignored(e) or exceptions.need_retry(e):
raise
ctxt.log.error("action failed",
action=action,
rule=rule,
exc_info=True)
# TODO(sileht): extract sentry event id and post it, so
# we can track it easly
return check_api.Result(check_api.Conclusion.FAILURE,
f"action '{action}' failed", "")
async def exec_action(
method_name: typing.Literal["run", "cancel"],
rule: rules.EvaluatedRule,
action: str,
ctxt: context.Context,
) -> check_api.Result:
try:
if method_name == "run":
method = rule.actions[action].run
elif method_name == "cancel":
method = rule.actions[action].cancel
else:
raise RuntimeError("wrong method_name")
return await method(ctxt, rule)
except Exception as e: # pragma: no cover
# Forward those to worker
if (
exceptions.should_be_ignored(e)
or exceptions.need_retry(e)
or isinstance(e, exceptions.UnprocessablePullRequest)
):
raise
# NOTE(sileht): the action fails, this is a bug!!!, so just set the
# result as pending and retry in 5 minutes...
ctxt.log.error("action failed", action=action, rule=rule, exc_info=True)
await delayed_refresh.plan_refresh_at_least_at(
ctxt.repository,
ctxt.pull["number"],
date.utcnow() + datetime.timedelta(minutes=5),
)
return check_api.Result(
check_api.Conclusion.PENDING,
f"Action '{action}' has unexpectedly failed, Mergify team is working on it, the state will be refreshed automatically.",
"",
)
def process(self):
pull_numbers = self.get_pulls()
self.log.info("%d pulls queued", len(pull_numbers), queue=list(pull_numbers))
if not pull_numbers:
return
pull_number = pull_numbers[0]
with github.get_client(self.owner, self.repo) as client:
ctxt = None
try:
subscription = asyncio.run(
sub_utils.get_subscription(client.auth.owner_id)
)
data = client.item(f"pulls/{pull_number}")
ctxt = context.Context(client, data, subscription)
if ctxt.pull["base"]["ref"] != self.ref:
ctxt.log.info(
"pull request base branch have changed",
old_branch=self.ref,
new_branch=ctxt.pull["base"]["ref"],
)
self._move_pull_to_new_base_branch(ctxt.pull["number"], self.ref)
elif ctxt.pull["state"] == "closed" or ctxt.is_behind:
# NOTE(sileht): Pick up this pull request and rebase it again
# or update its status and remove it from the queue
ctxt.log.info(
"pull request needs to be updated again or has been closed",
)
self.handle_first_pull_in_queue(ctxt)
else:
# NOTE(sileht): Pull request has not been merged or cancelled
# yet wait next loop
ctxt.log.info("pull request checks are still in progress")
except Exception as exc: # pragma: no cover
log = self.log if ctxt is None else ctxt.log
if exceptions.should_be_ignored(exc):
log.info(
"Fail to process merge queue, remove the pull request from the queue",
exc_info=True,
)
self.remove_pull(ctxt.pull["number"])
elif exceptions.need_retry(exc):
log.info("Fail to process merge queue, need retry", exc_info=True)
if isinstance(exc, exceptions.MergeableStateUnknown):
# NOTE(sileht): We need GitHub to recompute the state here (by
# merging something else for example), so move it to the end
self._move_pull_at_end(pull_number)
else:
log.error("Fail to process merge queue", exc_info=True)
self._move_pull_at_end(pull_number)
async def job_filter_and_dispatch(redis, event_type, event_id, data):
# TODO(sileht): is statsd async ?
meter_event(event_type, data)
if "repository" in data:
owner = data["repository"]["owner"]["login"]
repo = data["repository"]["name"]
else:
owner = "<unknown>"
repo = "<unknown>"
reason = get_ignore_reason(event_type, data)
if reason:
msg_action = f"ignored: {reason}"
else:
msg_action = "pushed to worker"
source_data = _extract_source_data(event_type, data)
if "pull_request" in data:
pull_number = data["pull_request"]["number"]
elif event_type == "issue_comment":
pull_number = data["issue"]["number"]
else:
pull_number = None
await worker.push(
redis,
owner,
repo,
pull_number,
event_type,
source_data,
)
# NOTE(sileht): nothing important should happen in this hook as we don't retry it
try:
await commands_runner.on_each_event(owner, repo, event_type, data)
except Exception as e:
if exceptions.should_be_ignored(e) or exceptions.need_retry(e):
LOG.debug("commands_runner.on_each_event failed",
exc_info=True)
else:
raise
LOG.info(
"GithubApp event %s",
msg_action,
event_type=event_type,
event_id=event_id,
sender=data["sender"]["login"],
gh_owner=owner,
gh_repo=repo,
)
if reason:
raise IgnoredEvent(event_type, event_id, reason)
def retry_task_on_exception(sender, task_id, exception, args, kwargs,
traceback, einfo, **other): # pragma: no cover
backoff = exceptions.need_retry(exception)
if backoff is None:
return
LOG.warning("job %s: failed %d times - retrying", task_id,
sender.request.retries)
# Exponential ^3 backoff
retry_in = 3**sender.request.retries * backoff
sender.retry(countdown=retry_in)
async def _translate_exception_to_retries(
self,
e,
stream_name,
attempts_key=None,
):
if isinstance(e, exceptions.MergifyNotInstalled):
if attempts_key:
await self.redis.hdel("attempts", attempts_key)
await self.redis.hdel("attempts", stream_name)
raise StreamUnused(stream_name) from e
if isinstance(e, github.TooManyPages):
# TODO(sileht): Ideally this should be catcher earlier to post an
# appropriate check-runs to inform user the PR is too big to be handled
# by Mergify, but this need a bit of refactory to do it, so in the
# meantimes...
if attempts_key:
await self.redis.hdel("attempts", attempts_key)
await self.redis.hdel("attempts", stream_name)
raise IgnoredException() from e
if exceptions.should_be_ignored(e):
if attempts_key:
await self.redis.hdel("attempts", attempts_key)
await self.redis.hdel("attempts", stream_name)
raise IgnoredException() from e
if isinstance(e, exceptions.RateLimited):
retry_at = utils.utcnow() + e.countdown
score = retry_at.timestamp()
if attempts_key:
await self.redis.hdel("attempts", attempts_key)
await self.redis.hdel("attempts", stream_name)
await self.redis.zaddoption("streams", "XX",
**{stream_name: score})
raise StreamRetry(stream_name, 0, retry_at) from e
backoff = exceptions.need_retry(e)
if backoff is None:
# NOTE(sileht): This is our fault, so retry until we fix the bug but
# without increasing the attempts
raise
attempts = await self.redis.hincrby("attempts", stream_name)
retry_in = 3**min(attempts, 3) * backoff
retry_at = utils.utcnow() + datetime.timedelta(seconds=retry_in)
score = retry_at.timestamp()
await self.redis.zaddoption("streams", "XX", **{stream_name: score})
raise StreamRetry(stream_name, attempts, retry_at) from e
def exec_action(method_name, rule, action, ctxt):
try:
method = getattr(rule.actions[action], method_name)
return method(ctxt, rule)
except Exception as e: # pragma: no cover
# Forward those to worker
if exceptions.should_be_ignored(e) or exceptions.need_retry(e):
raise
ctxt.log.error("action failed",
action=action,
rule=rule,
exc_info=True)
# TODO(sileht): extract sentry event id and post it, so
# we can track it easly
return check_api.Result(check_api.Conclusion.FAILURE,
"action '%s' failed" % action, "")
async def _get(cls, redis: utils.RedisCache,
owner_id: int) -> "UserTokensSaas":
cached_tokens = await cls._retrieve_from_cache(redis, owner_id)
if cached_tokens is None or await cached_tokens._has_expired():
try:
db_tokens = await cls._retrieve_from_db(redis, owner_id)
except Exception as exc:
if cached_tokens is not None and (
exceptions.should_be_ignored(exc)
or exceptions.need_retry(exc)):
# NOTE(sileht): return the cached tokens, instead of retring the
# stream, just because the dashboard has a connectivity issue.
return cached_tokens
raise
await db_tokens.save_to_cache()
return db_tokens
return cached_tokens
async def get_subscription(
cls, redis: utils.RedisCache, owner_id: int
) -> "Subscription":
"""Get a subscription."""
cached_sub = await cls._retrieve_subscription_from_cache(redis, owner_id)
if cached_sub is None or await cached_sub._has_expired():
try:
db_sub = await cls._retrieve_subscription_from_db(redis, owner_id)
except Exception as exc:
if cached_sub is not None and (
exceptions.should_be_ignored(exc) or exceptions.need_retry(exc)
):
# NOTE(sileht): return the cached sub, instead of retry the stream,
# just because the dashboard have connectivity issue.
return cached_sub
raise
await db_sub.save_subscription_to_cache()
return db_sub
return cached_sub
def main(): # pragma: no cover
utils.setup_logging()
LOG.info("Starting")
prometheus_client.start_http_server(8889)
LOG.info("Started")
while True:
try:
collect_metrics()
except Exception as e: # pragma: no cover
if exceptions.need_retry(e):
LOG.warning("fail to gather metrics: %s", str(e))
time.sleep(10 * 60)
continue
else:
LOG.error("Unexpected error during metrics gathering",
exc_info=True)
# Only generate metrics once per hour
time.sleep(60 * 60)
async def send_seats(seats: SeatsCountResultT) -> None:
async with http.AsyncClient() as client:
try:
await client.post(
f"{config.SUBSCRIPTION_BASE_URL}/on-premise/report",
headers={
"Authorization": f"token {config.SUBSCRIPTION_TOKEN}"
},
json={
"write_users": seats.write_users,
"active_users": seats.active_users,
"engine_version": config.VERSION,
# Deprecated version
"seats": seats.write_users,
},
)
except Exception as exc:
if exceptions.should_be_ignored(exc):
return
elif exceptions.need_retry(exc):
raise tenacity.TryAgain
else:
raise
def _log_on_exception(exc: Exception, msg: str) -> None:
if exceptions.should_be_ignored(exc) or exceptions.need_retry(exc):
log = LOG.debug
else:
log = LOG.error
log(msg, exc_info=exc)
async def run_actions(
ctxt: context.Context,
match: rules.RulesEvaluator,
checks: typing.Dict[str, github_types.GitHubCheckRun],
previous_conclusions: typing.Dict[str, check_api.Conclusion],
) -> typing.Dict[str, check_api.Conclusion]:
"""
What action.run() and action.cancel() return should be reworked a bit. Currently the
meaning is not really clear, it could be:
- None - (succeed but no dedicated report is posted with check api
- (None, "<title>", "<summary>") - (action is pending, for merge/backport/...)
- ("success", "<title>", "<summary>")
- ("failure", "<title>", "<summary>")
- ("neutral", "<title>", "<summary>")
- ("cancelled", "<title>", "<summary>")
"""
user_refresh_requested = ctxt.user_refresh_requested()
admin_refresh_requested = ctxt.admin_refresh_requested()
actions_ran = set()
conclusions = {}
# NOTE(sileht): We put first rules with missing conditions to do cancellation first.
# In case of a canceled merge action and another that need to be run. We want first
# to remove the PR from the queue and then add it back with the new config and not the
# reverse
matching_rules = sorted(match.matching_rules,
key=lambda rule: len(rule.missing_conditions) == 0)
method_name: typing.Literal["run", "cancel"]
for rule in matching_rules:
for action, action_obj in rule.actions.items():
check_name = f"Rule: {rule.name} ({action})"
done_by_another_action = action_obj.only_once and action in actions_ran
action_rule = await action_obj.get_rule(ctxt)
if rule.missing_conditions or action_rule.missing_conditions:
method_name = "cancel"
expected_conclusions = [
check_api.Conclusion.NEUTRAL,
check_api.Conclusion.CANCELLED,
]
else:
method_name = "run"
expected_conclusions = [
check_api.Conclusion.SUCCESS,
check_api.Conclusion.FAILURE,
]
actions_ran.add(action)
previous_conclusion = get_previous_conclusion(
previous_conclusions, check_name, checks)
need_to_be_run = (
action_obj.always_run or admin_refresh_requested
or (user_refresh_requested
and previous_conclusion == check_api.Conclusion.FAILURE)
or previous_conclusion not in expected_conclusions)
# TODO(sileht): refactor it to store the whole report in the check summary,
# not just the conclusions
if not need_to_be_run:
report = check_api.Result(previous_conclusion,
"Already in expected state", "")
message = "ignored, already in expected state"
elif done_by_another_action:
# NOTE(sileht) We can't run two action merge for example,
# This assumes the action produce a report
report = check_api.Result(
check_api.Conclusion.SUCCESS,
f"Another {action} action already ran",
"",
)
message = "ignored, another has already been run"
else:
# NOTE(sileht): check state change so we have to run "run" or "cancel"
report = await exec_action(
method_name,
rule,
action,
ctxt,
)
message = "executed"
if (report
and report.conclusion is not check_api.Conclusion.PENDING
and method_name == "run"):
statsd.increment("engine.actions.count",
tags=[f"name:{action}"])
if report:
if need_to_be_run and (not action_obj.silent_report
or report.conclusion not in (
check_api.Conclusion.SUCCESS,
check_api.Conclusion.CANCELLED,
check_api.Conclusion.PENDING,
)):
external_id = (check_api.USER_CREATED_CHECKS
if action_obj.allow_retrigger_mergify else
None)
try:
await check_api.set_check_run(
ctxt,
check_name,
report,
external_id=external_id,
)
except Exception as e:
if exceptions.should_be_ignored(e):
ctxt.log.info("Fail to post check `%s`",
check_name,
exc_info=True)
elif exceptions.need_retry(e):
raise
else:
ctxt.log.error("Fail to post check `%s`",
check_name,
exc_info=True)
conclusions[check_name] = report.conclusion
else:
# NOTE(sileht): action doesn't have report (eg:
# comment/request_reviews/..) So just assume it succeed
ctxt.log.error("action must return a conclusion",
action=action)
conclusions[check_name] = expected_conclusions[0]
ctxt.log.info(
"action evaluation: `%s` %s: %s/%s -> %s",
action,
message,
method_name,
previous_conclusion.value,
conclusions[check_name].value,
report=report,
previous_conclusion=previous_conclusion.value,
conclusion=conclusions[check_name].value,
action=action,
check_name=check_name,
event_types=[se["event_type"] for se in ctxt.sources],
)
return conclusions
def _wait_time_for_exception(retry_state):
return exceptions.need_retry(retry_state.outcome.exception())
def _exception_need_retry(retry_state):
return (
retry_state.outcome.failed and
exceptions.need_retry(retry_state.outcome.exception()) is not None
)
async def _translate_exception_to_retries(
self,
stream_name: StreamNameType,
attempts_key: typing.Optional[str] = None,
) -> typing.AsyncIterator[None]:
try:
yield
except Exception as e:
if isinstance(e, exceptions.MergeableStateUnknown) and attempts_key:
attempts = await self.redis_stream.hincrby("attempts", attempts_key)
if attempts < MAX_RETRIES:
raise PullRetry(attempts) from e
else:
await self.redis_stream.hdel("attempts", attempts_key)
raise MaxPullRetry(attempts) from e
if isinstance(e, exceptions.MergifyNotInstalled):
if attempts_key:
await self.redis_stream.hdel("attempts", attempts_key)
await self.redis_stream.hdel("attempts", stream_name)
raise StreamUnused(stream_name)
if isinstance(e, github.TooManyPages):
# TODO(sileht): Ideally this should be catcher earlier to post an
# appropriate check-runs to inform user the PR is too big to be handled
# by Mergify, but this need a bit of refactory to do it, so in the
# meantimes...
if attempts_key:
await self.redis_stream.hdel("attempts", attempts_key)
await self.redis_stream.hdel("attempts", stream_name)
raise IgnoredException()
if exceptions.should_be_ignored(e):
if attempts_key:
await self.redis_stream.hdel("attempts", attempts_key)
await self.redis_stream.hdel("attempts", stream_name)
raise IgnoredException()
if isinstance(e, exceptions.RateLimited):
retry_at = utils.utcnow() + e.countdown
score = retry_at.timestamp()
if attempts_key:
await self.redis_stream.hdel("attempts", attempts_key)
await self.redis_stream.hdel("attempts", stream_name)
await self.redis_stream.zaddoption(
"streams", "XX", **{stream_name: score}
)
raise StreamRetry(stream_name, 0, retry_at)
backoff = exceptions.need_retry(e)
if backoff is None:
# NOTE(sileht): This is our fault, so retry until we fix the bug but
# without increasing the attempts
raise
attempts = await self.redis_stream.hincrby("attempts", stream_name)
retry_in = 3 ** min(attempts, 3) * backoff
retry_at = utils.utcnow() + retry_in
score = retry_at.timestamp()
await self.redis_stream.zaddoption("streams", "XX", **{stream_name: score})
raise StreamRetry(stream_name, attempts, retry_at)
async def run_actions(
ctxt: context.Context,
match: rules.RulesEvaluator,
checks: typing.Dict[str, github_types.CachedGitHubCheckRun],
previous_conclusions: typing.Dict[str, check_api.Conclusion],
) -> typing.Dict[str, check_api.Conclusion]:
"""
What action.run() and action.cancel() return should be reworked a bit. Currently the
meaning is not really clear, it could be:
- None - (succeed but no dedicated report is posted with check api
- (None, "<title>", "<summary>") - (action is pending, for merge/backport/...)
- ("success", "<title>", "<summary>")
- ("failure", "<title>", "<summary>")
- ("neutral", "<title>", "<summary>")
- ("cancelled", "<title>", "<summary>")
"""
user_refresh_requested = ctxt.user_refresh_requested()
admin_refresh_requested = ctxt.admin_refresh_requested()
actions_ran = set()
conclusions = {}
# NOTE(sileht): We put first rules with missing conditions to do cancellation first.
# In case of a canceled merge action and another that need to be run. We want first
# to remove the PR from the queue and then add it back with the new config and not the
# reverse
matching_rules = sorted(
match.matching_rules, key=lambda rule: rule.conditions.match
)
method_name: typing.Literal["run", "cancel"]
for rule in matching_rules:
for action, action_obj in rule.actions.items():
check_name = rule.get_check_name(action)
done_by_another_action = (
actions.ActionFlag.DISALLOW_RERUN_ON_OTHER_RULES in action_obj.flags
and action in actions_ran
)
if (
not rule.conditions.match
or rule.disabled is not None
or (
ctxt.configuration_changed
and actions.ActionFlag.ALLOW_ON_CONFIGURATION_CHANGED
not in action_obj.flags
)
):
method_name = "cancel"
expected_conclusions = [
check_api.Conclusion.NEUTRAL,
check_api.Conclusion.CANCELLED,
]
else:
method_name = "run"
expected_conclusions = [
check_api.Conclusion.SUCCESS,
check_api.Conclusion.FAILURE,
]
actions_ran.add(action)
previous_conclusion = get_previous_conclusion(
previous_conclusions, check_name, checks
)
need_to_be_run = (
actions.ActionFlag.ALWAYS_RUN in action_obj.flags
or (
actions.ActionFlag.SUCCESS_IS_FINAL_STATE in action_obj.flags
and previous_conclusion == check_api.Conclusion.SUCCESS
)
or admin_refresh_requested
or (
user_refresh_requested
and previous_conclusion == check_api.Conclusion.FAILURE
)
or previous_conclusion not in expected_conclusions
)
# TODO(sileht): refactor it to store the whole report in the check summary,
# not just the conclusions
if not need_to_be_run:
report = check_api.Result(
previous_conclusion, "Already in expected state", ""
)
message = "ignored, already in expected state"
elif done_by_another_action:
# NOTE(sileht) We can't run two action merge for example,
# This assumes the action produce a report
report = check_api.Result(
check_api.Conclusion.SUCCESS,
f"Another {action} action already ran",
"",
)
message = "ignored, another has already been run"
else:
with ddtrace.tracer.trace(
f"action.{action}",
span_type="worker",
resource=str(ctxt),
) as span:
# NOTE(sileht): check state change so we have to run "run" or "cancel"
report = await exec_action(
method_name,
rule,
action,
ctxt,
)
span.set_tags({"conclusion": str(report.conclusion)})
message = "executed"
conclusions[check_name] = report.conclusion
if (
report.conclusion is not check_api.Conclusion.PENDING
and method_name == "run"
):
statsd.increment("engine.actions.count", tags=[f"name:{action}"])
if need_to_be_run and (
actions.ActionFlag.ALWAYS_SEND_REPORT in action_obj.flags
or report.conclusion
not in (
check_api.Conclusion.SUCCESS,
check_api.Conclusion.CANCELLED,
check_api.Conclusion.PENDING,
)
):
external_id = (
check_api.USER_CREATED_CHECKS
if actions.ActionFlag.ALLOW_RETRIGGER_MERGIFY in action_obj.flags
else None
)
try:
await check_api.set_check_run(
ctxt,
check_name,
report,
external_id=external_id,
)
except Exception as e:
if exceptions.should_be_ignored(e):
ctxt.log.info(
"Fail to post check `%s`", check_name, exc_info=True
)
elif exceptions.need_retry(e):
raise
else:
ctxt.log.error(
"Fail to post check `%s`", check_name, exc_info=True
)
ctxt.log.info(
"action evaluation: `%s` %s: %s/%s -> %s",
action,
message,
method_name,
previous_conclusion.value,
conclusions[check_name].value,
report=report,
previous_conclusion=previous_conclusion.value,
conclusion=conclusions[check_name].value,
action=action,
check_name=check_name,
event_types=[se["event_type"] for se in ctxt.sources],
)
return conclusions
