[
https://issues.apache.org/jira/browse/SOLR-17158?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17891152#comment-17891152
]
Gus Heck commented on SOLR-17158:
---------------------------------
I think that test you suggest defeats the code's efforts by putting all shard
queries into a 5 second sleep? All the Limit code relies on the thread
executing the query to check itself. There's no supervisor thread, so if the
thread is forced into a sleep, it can't check itself until it wakes up. It's
not a very realistic test because typically there will be variation among
shards, and in conjunction with the searcher level checks it takes a pretty
interesting scenario for all shards to get stuck simultaneously within the
Lucene searcher code, or a loop that has no check. This ticket mostly ensures
that once one thread times out, the effort/time spent waiting on the remaining
threads is minimized. One could imagine designs for timeAllowed where the
coordinator is completely divorced from the shard processing has a
countdownLatch that waits with a timeout and simply returns whatever happens to
be available when that expires, but that's not how things currently work and
relative to CPU and Memory limits that's a special case.
As for exceptions I've tried to avoid special casing like you suggest, and
stick to a simple clear rule: One should only get an exception if the server
*refused (4xx)* or *failed(5xx)* the task you gave it. When you add a limit you
give it a task that would be phrased as: "Spend X effort searching this query
and if it takes more effort than that (do|don't) tell me the documents you
managed to find" In the case you describe where the request limit is
unreasonably short and nothing is found, the server has certainly not refused,
nor has it failed. The server will set the partial results attribute to
indicate that there might be more results. If there's any issue, it might be
the naming of "partial results" since that name sort-of sounds like there must
be some results. Perhaps a more communicative response would say
{code:java}
"resultMode":"bestEffort",
"resultStatus":"resultsIncomplete" {code}
or
{code:java}
"resultMode":"allOrNothing",
"resultStatus":"incompleteResultsOmitted"{code}
or
{code:java}
"resultMode":"bestEffort",
"resultStatus":"resultsComplete" {code}
or
{code:java}
"resultMode":"allOrNothing",
"resultStatus":"resultsComplete"{code}
However that's a much more backwards incompatible change.
I have thought it would be much nicer if there were an HTTP status code that
could be set to indicate partial results without having to parse the response,
and the closest is
[206|https://developer.mozilla.org/en-US/docs/Web/HTTP/Headers/Range], but
that's really meant for [Range
Requests|https://developer.mozilla.org/en-US/docs/Web/HTTP/Headers/Range] and
since we neither use the associated headers nor have a way to express, or even
satisfy "give me the second 30 seconds of results" it's not a good fit. We
could just co-opt our own 2xx number and document it, perhaps even propose it
(since results limited by processing constraints probably could be considered a
generic situation... maybe {{225 LIMITED RESULTS}} ?), but that may cause
problems for current users and is backwards incompatible for folks who check
for {{200 OK}} specifically.
> Terminate distributed processing quickly when query limit is reached
> --------------------------------------------------------------------
>
> Key: SOLR-17158
> URL: https://issues.apache.org/jira/browse/SOLR-17158
> Project: Solr
> Issue Type: Sub-task
> Components: Query Limits
> Reporter: Andrzej Bialecki
> Assignee: Gus Heck
> Priority: Major
> Labels: pull-request-available
> Fix For: main (10.0), 9.8
>
> Time Spent: 8h 50m
> Remaining Estimate: 0h
>
> Solr should make sure that when query limits are reached and partial results
> are not needed (and not wanted) then both the processing in shards and in the
> query coordinator should be terminated as quickly as possible, and Solr
> should minimize wasted resources spent on eg. returning data from the
> remaining shards, merging responses in the coordinator, or returning any data
> back to the user.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscr...@solr.apache.org
For additional commands, e-mail: issues-h...@solr.apache.org
