[prev in list] [next in list] [prev in thread] [next in thread] 

List:       solr-user
Subject:    Re: Timeout on LTR query causing out of memory
From:       rajani m <rajinimaski () gmail ! com>
Date:       2023-09-27 15:38:28
Message-ID: CAE0ry3gPz+1Xk3=zrgns0uu_JBVVY-tC0cOOk7NtUfFmYJ1XGw () mail ! gmail ! com
[Download RAW message or body]


Any alternative to timeout LTR re-ranker similar to the "timeAllowed" that
limits the main ranker? or is there a way to tell if a query is expensive
for ltr given a number of terms or has N hits?

The LTR queries are taking forever and timing out at some point with
"ExitableDirectoryReader
'' exception, and at the same time this is causing the heap to completely
fill up and trigger an oom. An oom that invokes oom killer script that
brings solr down for sometime.  Errors stack trace here on the gist
<https://gist.github.com/rajanim/1ee04de066aa4da6f0916f0c38274489>


   1. too long to iterate over terms. Timeout: timeoutAt: 946228141325435
   (System.nanoTime(): 946228142713792),
   TermsEnum=org.apache.lucene.codecs.lucene90.blocktree.SegmentTermsEnum@5f420d22
   2. at
   org.apache.solr.search.ReRankCollector.topDocs(ReRankCollector.java:163)
   ~[?:?]



   1. at
   org.apache.lucene.search.TermQuery$TermWeight.getTermsEnum(TermQuery.java:144)
   ~[?:?]
   2. at
   org.apache.lucene.search.TermQuery$TermWeight.scorer(TermQuery.java:107)
   ~[?:?]
   3. at
   org.apache.solr.ltr.feature.SolrFeature$SolrFeatureWeight.scorer(SolrFeature.java:229)
   ~[?:?]
   4. at
   org.apache.solr.ltr.LTRScoringQuery$ModelWeight.scorer(LTRScoringQuery.java:500)
   ~[?:?]


On Tue, Sep 26, 2023 at 10:20 PM rajani m <rajinimaski@gmail.com> wrote:

> Hi Solr Users,
>
>    Is there an alternative way to time limit an LTR query and request main
> ranker results if a reranker such as an LTR query takes more time than
> allowed?  I see there is a SOLR-14607
> <https://issues.apache.org/jira/browse/SOLR-14607>  regarding LTR time
> allowed issue, but it is open. Wondering if there are any other
> alternatives to handle this?
>
> Some ltr based rerank queries are returning the following exception and
> causing solr heap to fill up before it is garbage collected  by g1gc hence
> reporting a full heap usage and triggering oom.I am looking for solutions
> to prevent it from  happening. Any suggestions appreciated.
>
> error -
> Caused by:
> org.apache.lucene.index.ExitableDirectoryReader$ExitingReaderException: The
> request took too long to iterate over point values.
>


[prev in list] [next in list] [prev in thread] [next in thread]