[
https://issues.apache.org/jira/browse/SOLR-797?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13882944#comment-13882944
]
Gregg Donovan commented on SOLR-797:
------------------------------------
I'm interested in this as well. We had an custom API that was similar to the
attached patch. When we switched to EmbeddedSolrServer we noticed an increase
in time spent deserializing the Solr response, memory allocated, and GC
spikiness.
One issue with the current EmbeddedSolrServer code is that it starts with
ByteArrayOutputStream of 32 bytes and resizes repeatedly it to fit the results.
We have large responses and we notice the GC hit. We experimented with a
ThreadLocal<ByteBuffer>, but avoiding serializing and parsing altogether for
EmbeddedSolrServer seems like an even better idea.
If there's interest, we'd be happy to revive/update/test this patch.
> Construct EmbeddedSolrServer response without serializing/parsing
> -----------------------------------------------------------------
>
> Key: SOLR-797
> URL: https://issues.apache.org/jira/browse/SOLR-797
> Project: Solr
> Issue Type: Improvement
> Components: clients - java
> Affects Versions: 1.3
> Reporter: Jonathan Lee
> Priority: Minor
> Fix For: 4.7
>
> Attachments: SOLR-797.patch, SOLR-797.patch
>
>
> Currently, the EmbeddedSolrServer serializes the response and reparses in
> order to create the final NamedList response. From the comment in
> EmbeddedSolrServer.java, the goal is to:
> * convert the response directly into a named list
--
This message was sent by Atlassian JIRA
(v6.1.5#6160)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
