It seems there was some push back due to the complexity and the impact on the overseer code. So this PR is now probably stale.
I opened a simpler version that introduces a dedicated thread pool for "expensive" operations. End behavior is the same: we don't execute more than 5 concurrent expensive operations per Solr node. But the change in the code base is more scoped, mostly there is no change in task tracking. Any feedback is appreciated Thanks [1] https://github.com/apache/solr/pull/1864 Le lun. 7 août 2023 à 19:28, David Smiley <david.w.smi...@gmail.com> a écrit : > In https://issues.apache.org/jira/browse/SOLR-16879, we have a proposal > for > certain core level admin operations to be deemed expensive and throttled > (e.g. only 5 at a time). It only works when invoked with the async style > (with the "async" param). > > At present, only BACKUPCORE & RESTORECORE are flagged as expensive in the > PR. I'm thinking other operations should consider such designation too: > > MERGEINDEXES, SPLIT. > > That's probably it. Opinions & reviews are welcome as usual! > > ~ David >
