>
> Another note here is problems come and go away in unpredictable ways.
Before SOLR-14665 I never thought about doing a performance test of
creating thousands of collections. The problem here is the same with our
tests, even Solr has a huge number of tests, bugs still happen here and
there, sometimes they are serious bugs. So no matter how good a performance
tool we have (not mentioning that we do not have a consistent, unified way
to do that) degrade performance can still happen in unpredictable ways.

Moreover, if a commit changes a particular codepath, like PeerSync class,
we do not have an available tool for that. Requiring people to write a tool
just to measure their changes to do a simple commit (and likely throw it
away) seems a big -1 to me, not to mention that bugs can arise in their
performance tools and numbers are hardly to be trusted.

Reply via email to