Replies: 5 comments 6 replies
-
Is it possible that somehow the purging is causing the index to have to be completely rebuilt instead of just updated incrementally? |
Beta Was this translation helpful? Give feedback.
-
Purge info limit can updated with https://docs.couchdb.org/en/3.2.0/api/database/misc.html#put--db-_purged_infos_limit. It's per database. I think the As for the views being rebuilt I could not see any place where the view could be reset due to a stale purge sequence. See if there are any crashes or stack traces in the logs. Otherwise make sure the design document is not updated, otherwise every update will generate a new view hash and trigger a full view rebuild process. |
Beta Was this translation helpful? Give feedback.
-
Thanks for the response @nickva! For better or worse, we cannot really replicate the errors above (so I cannot be sure what effect removing There is a lack of any interesting errors in the logs. I have spent days sifting through them and almost all the errors are just timeout errors like this:
I do not see any updates to the design docs (no PUTs to |
Beta Was this translation helpful? Give feedback.
-
@jkuester would there be a way to reproduce the issue like a script or series of step like: create db, populate with data, build view, purge, etc? |
Beta Was this translation helpful? Give feedback.
-
As an additional note, I did find this discussion with the same fabric timeout in
This seems interesting and perhaps related since I guess we are seeing some funky interaction between these timeouts and indexing... |
Beta Was this translation helpful? Give feedback.
-
I am trying to understand the causes behind a recent outage. Basically Couch was swamped with traffic and most calls were returning timeout errors (even though the host server had plenty of free resources). This went on for days and the timeouts continued even when almost all traffic stopped besides a nightly purge script. Then, one night, things just magically started working again and the timeouts were gone. Now, I am trying to piece together what happened...
Specifically, though, I am trying to understand if/how the calls to
_purge
might have effected the overall performance of the instance. Here is an example from the Couch logs of the indexing for a specific shard over a 24hr period. These are all of the log entries for that specific shard for the complete 24hr period.The original message at
2022-08-01T21:53
is right in the middle of the massive batch of_purge
calls that were run that night (almost all of which just ended up timing out...). You can see from the log dates that the indexing for that shard takes ~12hrs to complete. Then the indexing immediately restarts again and runs for another 11hrs. Is this excessive time spent building the index for the shard caused by theInvalid purge doc
errors? What does theThe purge sequence for ... exceeds configured threshold
mean? Is there some way we can increase that configuration to avoid this issue in the future?I found some information in the docs about the purge sequence, but did not find any way to configure the limit...
Details:
2.3.1
222G
2141
databases5,000,000+
docs in the largest dbThank you in advance!
Beta Was this translation helpful? Give feedback.
All reactions