Details
-
Type:
Improvement
-
Status: Closed
-
Priority:
Minor
-
Resolution: Fixed
-
Affects Version/s: None
-
Fix Version/s: 6.3
-
Component/s: None
-
Labels:None
Description
We output a lot of logging data from various bits of SolrCloud, a lot of which is basically implementation detail that isn't particularly helpful. Here's a patch to move a bunch of stuff to #debug level, rather than #info.
Issue Links
- is related to
-
SOLR-6677
Reduce logging during startup and shutdown
-
- Closed
-
Activity
- All
- Comments
- Work Log
- History
- Activity
- Transitions
I've started to look at reducing log noise in general, this looks like a good bunch of improvements for Cloud. Alan Woodward, are you able to update this for "master" and commit?
I agree that perhaps the "A cluster state change" message could remain INFO, if it does not occur too often?
Nice work Jan. Most of this patch is now completely out of date, but I'll work up a new one. Are you planning on looking at noise on shutdown as well? There's always a vast amount of bumpf there that makes debugging test failures a right pain.
This moves a bunch of logging from ZkController, SolrZkClient, ZkStateReader and the various Overseer classes from info to debug.
This looks very good.
I think we just need to fix the logging one bite at a time and commit stuff as we go. This is a huge step wrt readability of Cloud logging! Then we can tackle shutdown cases next perhaps, could be a bit more tricky to detect normal cases from the irregular ones perhaps?
Commit 082f8e3f92de67ba694c414e526e622a51d2c056 in lucene-solr's branch refs/heads/branch_6x from Alan Woodward
[ https://git-wip-us.apache.org/repos/asf?p=lucene-solr.git;h=082f8e3 ]
SOLR-5563: Quieten down SolrCloud logging
Commit a002aa52487fcf9f848794d3e11e5fad568c6222 in lucene-solr's branch refs/heads/master from Alan Woodward
[ https://git-wip-us.apache.org/repos/asf?p=lucene-solr.git;h=a002aa5 ]
SOLR-5563: Quieten down SolrCloud logging
Thanks for prodding me into action Jan!
Closing after 6.3.0 release.
Just a note, we are still in the process of trying to debug some cloud-related issues that we only see in production, and messages like the Watcher (from ConnectionManager.java) and "A cluster state change" (from ZKStateReader.java) are really useful to us, without those we would never have been able to debug our latest issue (which we should be logging later today).
I know we could configure logging to enable some debug trace in specific components, but we then need to go through all the components and work out which ones we need to enable. but I would just urge some caution on this before committing IMHO. SolrCloud issues are still being reported issues on the list, so there still seem to be some issues related to the Overseer that haven't been caught yet.
Just my tuppence worth