Soham Gandhi created SOLR-17539:
-----------------------------------
Summary: Increased memory usage after upgrading to Solr 9.6.0 -
DocValuesIteratorCache
Key: SOLR-17539
URL: https://issues.apache.org/jira/browse/SOLR-17539
Project: Solr
Issue Type: Bug
Security Level: Public (Default Security Level. Issues are Public)
Components: search
Affects Versions: 9.7, 9.6, 9.4
Reporter: Soham Gandhi
We recently upgraded from Solr 8.11.1 to Solr 9.6.0 and noticed an increase of
in memory usage in our testing.
Here's a typical test environment setup -
* 2 mil documents in the index taking up ~3 GB disk size
* Test suite executes 100 unique queries every other minute for an hour
* Metrics like query response times, CPU and memory usage are captured and
compared against a baseline benchmark.
The upgrade did not include any change to schema or solrconfig as compared to
Solr 8.11.1, The test results reported an increase in memory usage from 4 GB to
6 GB.
Here are the memory usage snapshots from our test runs for Solr 8.11.1 (left)
and Solr 9.6.0
!https://mail.google.com/mail/u/0?ui=2&ik=8e78b77afe&attid=0.1&permmsgid=msg-a:r-8567447807355761378&th=1924eb8757fe54af&view=fimg&fur=ip&sz=s0-l75-ft&attbid=ANGjdJ8l_61BAciOjcTBSQgWAezcaZAc6wDNI7FzU8EXloaLbQbQUrqFhl8s0BKIUSwdacQJSO44H39Mu6u4fQfTH7o11M5pY_9XDvzvn1zOeMrZu2kqkho29mcoZO8&disp=emb&realattid=ii_m1s92j1g5|width=916,height=261!
Here's a snippet of memory profiling during test runs for Solr 8.11.1 (left)
and Solr 9.6.0
!https://mail.google.com/mail/u/0?ui=2&ik=8e78b77afe&attid=0.2&permmsgid=msg-a:r-8567447807355761378&th=1924eb8757fe54af&view=fimg&fur=ip&sz=s0-l75-ft&attbid=ANGjdJ_snNNurQP7b4OrCQ53e_VylKmRw-pmgot0S9rCoCmLTOhAPdtVzkFEjOevL-Cm2pyHc6kz9RYJAQL-c3IhX2YZyLqN8E5UwhUX1K8Y7d-TClyWMt8Pw5WimgE&disp=emb&realattid=ii_m1s966xw6!
>From our investigation we found this behaviour to be linked to the
>{{DocValuesIteratorCache}} introduced in Solr 9.4.0 as part of SOLR-16989
>Optimize and consolidate reuse of DocValues iterators for value retrieval.
The improvement in value retrieval is not seen in our runs as the queries were
already fast (~100ms). Query response times improve if the page size is
increased by a factor of 10 (50 -> 500); each query retrieves the first 50
results.
The schema in our test environment contains ~800 fields, with {{docValues}}
enabled on 94 (44 predefined fields and 50 dynamic fields) where
{{useDocValuesAsStored=true}} is set.
Each query retrieves the first 50 documents, with 5 to 7 fields populated in
each.
The workaround suggested on the forum where we do not retrieve any
{{docValues}} (with {{{}useDocValuesAsStored=true){}}}may not be trivial as our
queries use a glob pattern ({{{}fl=id, field_prefix_*{}}}), where the
{{field_prefix}} matches most of the fields.
We would like to have a configuration to disable the feature introduced in
SOLR-16989.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
