Cockroach: partitionccl: partitioning hangs for 30s and has inconsistent UX relative to other ALTERs

[nstewart]

Ran the command alter table movr.vehicles partition by list (city) ( partition north_america values in ('new york', 'boston', 'san francisco'), partition europe values in ('paris')) and saw the following behavior:

This command hung for 30 seconds before completing. There are 50k rows in this table. The schema is:

I see an event that a schema change started but there's not an end event, even though I see other schema changes ending in the event log.

If I have to wait for 30 seconds I would expect this to show up in the jobs table, especially when according to docs "Schema changes through ALTER TABLE" are supposed to be there.

[nstewart]

cc @awoods187 @dianasaur323

[benesch]

I'm curious if ALTER TABLE ... RENAME TO ... exhibits the same problem. If it does, it's definitely not partitioning's fault. @nstewart, next time you're testing the demo, could you see if renaming the table is as slow?

[nstewart]

[email protected]:26257/> alter table movr.vehicles rename to movr.vehicle;
RENAME TABLE

Time: 3.172001189s

[email protected]:26257/> alter table movr.vehicle rename to movr.vehicles;
RENAME TABLE

Time: 6.13844549s

[email protected]:26257/> alter table movr.vehicles partition by list (city) ( partition north_america values in ('new york', 'boston', 'san francisco'), partition europe values in ('paris'));
ALTER TABLE

Time: 1.988987371s

This time it was much faster. Neither of the rename to actions showed up in the event log though. We can track that separately.

I'll continue to monitor. Not sure when it hangs and when it proceeds normally.

[dianasaur323]

The event log issue is a known issue, and on @piyush-singh's plate for the next release.

[rolandcrosby]

@jordanlewis can you take a look at this? (feel free to send back to bulk I/O if you think this is specific to schema changes)

[jordanlewis]

Hmm, I'm not entirely sure what the ask here is. Is it about the event log in particular, or investigating some slowness? If the latter, I'm inclined to close as this was over a year ago. If it's about the event log, should we rename/make a new issue?

[rolandcrosby]

@piyush-singh do you have another issue we can rely on for general improvements to partitioning visibility and status? If so I think we can close this.

[piyush-singh]

Yes indeed, Andy is tracking enhancements on #31551. We can probably close out this issue in favor of that one.