[v3-0-test] Resolve OOM When Reading Large Logs in Webserver

[jason810496]

related: #49470

What

Backport Resolve OOM When Reading Large Logs in Webserver #49470 to v3-0-test branch.

The only conflict when cherry-pick:

Merge Changes
└── airflow-core/src/airflow
├── api_fastapi/core_api/routes/public
│ └── log.py [9+, !]
├── utils/log
│ ├── file_task_handler.py [4, !]
│ └── log_reader.py [!]
└── providers
├── amazon/tests/unit/amazon/aws/log
│ └── test_cloudwatch_task_handler.py [!]
├── elasticsearch/src/airflow/providers/elasticsearch/log
│ └── es_task_handler.py [!]
└── opensearch/src/airflow/providers/opensearch/log
└── os_task_handler.py [!]

Maybe we should wait for 3.0.3 release before merging it into the test branch. (Just in case there is an RC6)

[gopidesupavan]

@jason810496 is backport something went wrong? could see many files

[kaxil]

@jason810496 is backport something went wrong? could see many files

Fixed it

[jason810496]

@jason810496 is backport something went wrong? could see many files

Fixed it

Thanks @kaxil!
I just check this backport PR modify the same files by the following script.

REPO=apache/airflow
PR1=53167
PR2=49470
# Get the list of changed files from each PR
FILES_PR1=$(gh pr view "$PR1" --repo "$REPO" --json files --jq '.files[].path'| sort)
FILES_PR2=$(gh pr view "$PR2" --repo "$REPO" --json files --jq '.files[].path'| sort)
TMP1=$(mktemp)
TMP2=$(mktemp)echo"$FILES_PR1">"$TMP1"echo"$FILES_PR2">"$TMP2"# Compare the two sorted listsif cmp -s "$TMP1""$TMP2";thenecho"✅ PR #$PR1 and PR #$PR2 modify the exact same set of files."elseecho"❌ PR #$PR1 and PR #$PR2 modify different sets of files."echo -e "\n📄 Files only in PR #$PR1:"
comm -23 "$TMP1""$TMP2"echo -e "\n📄 Files only in PR #$PR2:"
comm -13 "$TMP1""$TMP2"fi
rm -f "$TMP1""$TMP2"

I have also test the behavior using breeze k8s with the following setup, and all work well!

1. kubernetes executor
2. celery executor
3. kubernetes executor + remote logging with Google Cloud Storage
4. celery executor + remote logging with Google Cloud Storage

Update

but the setup with remote logging with Google Cloud Storage result in errors.
I just found the root cause and will raise another PR to fix it, mark this PR to draft to void mislead merging.

This is because the providers and helm chart are outdate when I run k8s build-k8s-image and k8s deploy-airflow on and switch to v3-0-test branch.
So I encounter this error for GCSRemoteLogIO which fixed in #50590 already and the extraVolumeMounts not work properly.

[jason810496]

After configuring the remote logging setup correctly, all the behavior of getting task instance logs page work well.
Ready to merge if this CI run is green.

[potiuk]

cool!