Jobs were completed actually but still are 'running' in show jobs

One node in cluster died several times that made something strange happened.

And there are some other pending jobs

I know it is impossible to manually cancel jobs now but there should be some automatic mechanism to cancel or kill these holding jobs.

Hi @sydnever,

That does seem like strange behavior from the screenshots.

Can you provide more detail about the sequence of events that occurred leading up to this? What commands did you run to start the schema changes , and when did the node die in relation to them?

Also, can you describe the cluster you’re running this on? How many nodes are in the cluster, and what kind of hardware are they on?


Hi @couchand ,

We used two databases to test crdb.

I use “tpch10g” to do tpch test by your providing tool called “loadgen/tpch” which does insert data and add indexes(schema changes on tpch10g) automatically.

My partner use “sbtest100” to do sysbench test. When he were inserting data and adding indexes by sql shell, the node which was connected to him died. After the node rejoined, he dropped the table then reloaded data and added index. The old jobs which did schema changes on the dropped table are still pending or running.

I guess that the action to drop tables can’t cancel schema change jobs but these jobs can’t continue anymore, so they are still in job list and are running or pending.

Sorry for the slow reply here and thanks for the report.

This sounds like a bug. Was the end state of the database after your schema changes correct?