mmenanno opened a new pull request, #76326:
URL: https://github.com/apache/airflow/pull/76326
When `KubernetesPodOperator` follows a container's logs and the read ends
while the container still reports as running, `fetch_container_logs` reads
again from the last logged timestamp. The log API only accepts `sinceSeconds`,
so the new read starts up to a second earlier, and those lines were logged a
second time. #39861 added a warning for this ("Logs generated in the last one
second might get duplicated") but left the duplication in place.
This change skips the repeated lines in the loop that already parses each
line's timestamp, so no extra parsing is added. `PodManager` now records the
newest timestamp read from each container and how many lines carried it. A
resumed read drops lines before that timestamp, plus their untimestamped
continuation lines. At that exact timestamp it drops only as many lines as were
already read. Timestamps are parsed to the microsecond, so a burst of lines
(such as a traceback) can share one, and a read can stop partway through it.
The warning no longer says logs might be duplicated.
We hit this in production on 10.21.0, where some task logs repeated their
last lines and one repeated every line. We've been running the same logic as a
`PodManager` subclass.
Tests: a parametrized test covers a replayed last second, replayed
continuation lines, and an unread line that shares the last logged timestamp.
All three fail without the change.
A possible follow-up, not changed here: on a read timeout, `consume_logs`
resumes from the last timestamp plus 2 seconds to avoid duplicates. That can
drop lines written in those 2 seconds, and with this change the offset is no
longer needed.
related: #39236, #33498
---
##### Was generative AI tooling used to co-author this PR?
- [X] Yes (please specify the tool below)
Generated-by: Claude Code 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]
