Gus Heck created SOLR-17298:
-------------------------------
Summary: Multithreaded search breaks limits, and possibly other
things
Key: SOLR-17298
URL: https://issues.apache.org/jira/browse/SOLR-17298
Project: Solr
Issue Type: Bug
Security Level: Public (Default Security Level. Issues are Public)
Components: search
Affects Versions: 9x, main (10.0), 9.7
Reporter: Gus Heck
https://issues.apache.org/jira/browse/SOLR-13350 was merged to main somewhat
unexpectedly, and then back-ported to 9x without any response to feedback from
multiple committers, including feedback that
* By turning it on by default, it breaks the recently released CPU limits (as
shown by changes to unit tests).
* Incompatibility with timeAllowed, cpuTimeAllowed, segmentTerminateEarly,
GraphQuery, RankQuery and JoinQuery and perhaps was not clearly documented
* The code presents a possibility for users to receive a non-numeric max score
("NaN").
I have not verified it yet, but I would also be worried about the health of CPU
time logging to be added in https://issues.apache.org/jira/browse/SOLR-16986
after this change.
Given that:
* Some of the above issues represent back compatibility breaks or potential
back compatibility breaks for released features
* The decision to break compatibility within the 9x release series deserves a
formal vote.
* There has been no communication/response from the committer who merged these
changes since May 6 (aside from the backport to 9x on May 13) it seems that
this state may persist for some time.
Therefore it appears necessary to file this issue to ensure anything but a
9.6.1 is blocked until the above issues are sorted out. This ticket can serve
as a parent ticket to whatever various solutions are agreed upon.
Multi-threaded search is an awesome feature that has taken a very long time to
be realized and is obviously desirable, but we have now placed ourselves in an
awkward position by not resolving these last few issues before back porting.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]