[Bug]: There will be more than one task that state is started when bulk load when there is only one datanode

[zhuwenxing]

Is there an existing issue for this?

• I have searched the existing issues

Environment

- Milvus version: master-20220928-adeaac4f
- Deployment mode(standalone or cluster): cluster
- SDK version(e.g. pymilvus v2.0.0rc2): pymilvus==2.2.0.dev36
- OS(Ubuntu or CentOS): 
- CPU/Memory: 
- GPU: 
- Others:

Current Behavior

Expected Behavior

No response

Steps To Reproduce

No response

Milvus Log

test-milvus-bulk-load-09-29-17-45 .zip

Anything else?

No response

[zhuwenxing]

/assign @soothing-rain
/unassign @yanliang567

[soothing-rain]

Should be fixed by #19561

[soothing-rain]

/unassign
/assign @zhuwenxing

[zhuwenxing]

pymilvus==2.2.0.dev38
Milvus version: master-20221008-5b4038e9

failed job: https://qa-jenkins.milvus.io/blue/organizations/jenkins/bulk_load_test/detail/bulk_load_test/19/pipeline

log:

artifacts-bulk-load-test-19-pytest-logs (1).tar.gz

artifacts-bulk-load-test-19-server-logs (2).tar.gz

[zhuwenxing]

/assign @soothing-rain

[soothing-rain]

pymilvus==2.2.0.dev38 Milvus version: master-20221008-5b4038e9

failed job: https://qa-jenkins.milvus.io/blue/organizations/jenkins/bulk_load_test/detail/bulk_load_test/19/pipeline

log:

artifacts-bulk-load-test-19-pytest-logs (1).tar.gz

artifacts-bulk-load-test-19-server-logs (2).tar.gz

But there are two DataNode services right?

[zhuwenxing]

But there are two DataNode services right?

Yes, so it is as expected!

[zhuwenxing]

/unassign @soothing-rain

[zhuwenxing]

Milvus version: master-20221008-7e620a7a

It still happened in the datanode pod killchaos test
log:
artifacts-bulk-load-test-6-server-logs.tar.gz
artifacts-bulk-load-test-6-pytest-logs.tar.gz

before chaos, it works well, there is only one working task when datanode number is 1.

during chaos, it becomes abnormal.

after chaos, the working tasks number is still wrong

[zhuwenxing]

/assign @soothing-rain

[soothing-rain]

OK this might make sense in chaos testing.

If a DataNode gets killed while working on a bulk load task, the state of that task will stay in STARTED and will eventually be removed when expired. In other words, these are "dead" bulk load tasks in STARTED states.

Also these "dead" tasks will not impact future bulk load tasks, as new tasks will be assigned to new DataNodes.

[zhuwenxing]

Though they are "dead" STARTED tasks, it would be better to clean them, otherwise, it will make the user confused.

So I think it is still a bug that needs to fix, but the priority is not that high.

[zhuwenxing]

/unassign

[soothing-rain]

Though they are "dead" STARTED tasks, it would be better to clean them, otherwise, it will make the user confused.

So I think it is still a bug that needs to fix, but the priority is not that high.

Sure I created a new issue for this one:
#19777

We can close 19563 for now.

/assign @zhuwenxing

[stale]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.
Rotten issues close after 30d of inactivity. Reopen the issue with /reopen.