Nik-Tenzo opened a new pull request, #76305:
URL: https://github.com/apache/airflow/pull/76305
CloudWatch remote logging always checks that the log group exists before
writing: watchtower's
`CloudWatchLogHandler` runs a paginated `DescribeLogGroups`, then
`CreateLogGroup` if the group is missing, each time a
handler is built. This adds an `[aws]
cloudwatch_task_handler_create_log_group` option (default `True`, so nothing
changes by default), passed through to watchtower, so a deployment that
manages the log group itself can skip both
calls:
```ini
[aws]
cloudwatch_task_handler_create_log_group = False
```
As with `[aws] s3_task_handler_acl_policy` (#70394), the same value can also
be set through
`[logging] remote_task_handler_kwargs`, e.g. `{"create_log_group": false}`,
since it's a field of
`CloudWatchRemoteLogIO`.
Why: we create the task log group with infrastructure as code, so the check
is redundant, and on Airflow 2 it got us
throttled. Each task try ran in a fresh process with a new handler, so every
try made a `DescribeLogGroups` call. That
went well over the account's `DescribeLogGroups` quota, and because the
quota is account-wide, other callers of the
API (our Terraform plans) slowed down too, until AWS raised it. Since then
we've carried subclasses of the handler and
IO that pass `create_log_group=False`, re-copied whenever the handler
lifecycle changed. At today's ~50,000 task tries
an hour, the per-try check would be about 14 calls a second.
On Airflow 3 the exposure is much smaller: the IO keeps one handler for the
life of the process, so the check runs once
per process that writes task logs, not once per task. It still runs per task
try on Airflow 2.11, which this provider
supports, and in either case the option lets a deployment drop the call and
the `logs:DescribeLogGroups` /
`logs:CreateLogGroup` permissions it needs.
#73004 also edits `_build_handler` (per-stream handlers that skip the check
after the first one). The two changes are
independent; if #73004 merges first I'll rebase this onto it, and the option
then applies to its first handler.
related: #35164, #70394, #73004
---
##### Was generative AI tooling used to co-author this PR?
- [X] Yes — Claude Code (Claude Opus 5.5)
Generated-by: Claude Code (Claude Opus 5.5) following [the
guidelines](https://github.com/apache/airflow/blob/main/contributing-docs/05_pull_requests.rst#gen-ai-assisted-contributions)
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
