[
https://issues.apache.org/jira/browse/SOLR-14457?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17100277#comment-17100277
]
Samuel García Martínez commented on SOLR-14457:
-----------------------------------------------
Just for me to understand, I have several questions:
* In your scenario, are you using the HTTP1 or HTTP2 client?
* When do you observe the leaks?
* How a connection is leaked with proper HTTP responses? Do you know where it
happens?
In this bug, the exception happens like the following:
* HTTPSolrClient receives an error code or a malformed response
* The client throws an error
* The client, on a finally block, ensures the connection is closed with
Utils.consumeFully(HttpEntity)
* Utils.consumeFully uses HttpEntity#getContent
* When compression is enabled and response is compressed, the interceptor
returns a new GzipInputStream
* The GzipInputStream expects the header and it's not there because it's not
really compressed.
Does your scenario look like the one I explained above?
> SolrClient leaks connections on compressed responses if the response is
> malformed
> ---------------------------------------------------------------------------------
>
> Key: SOLR-14457
> URL: https://issues.apache.org/jira/browse/SOLR-14457
> Project: Solr
> Issue Type: Bug
> Security Level: Public(Default Security Level. Issues are Public)
> Components: SolrJ
> Affects Versions: 7.7.2
> Environment: Solr version: 7.7.2
> Solr cloud enabled
> Cluster topology: 6 nodes, 1 single collection, 10 shards and 3 replicas. 1
> HTTP LB using
> Round Robin over all nodes
> All cluster nodes have gzip enabled for all paths, all HTTP verbs and all
> MIME types.
> Solr client: HttpSolrClient targeting the HTTP LB
> Reporter: Samuel García Martínez
> Priority: Major
>
> h3. Summary
> When the SolrJ receives a malformed response Entity, for example like the one
> described in SOLR-14456, the client leaks the connection forever as it's
> never released back to the pool.
> h3. Problem description
> HttpSolrClient should have compression enabled, so it uses the compression
> interceptors.
> When the request is marked with "Content-Encoding: gzip" but for whatever
> reason the response is not in GZIP format, when HttpSolrClient tries to
> close the connection using Utils.consumeFully(), it tries to create the
> GzipInputStream failing and throwing an Exception. The exception thrown makes
> it impossible to access the underlying InputStream to be closed, therefore
> the connection is leaked.
> Despite that the content in the response should honour the headers specified
> for it, SolrJ should be reliable enough to prevent the connection leak when
> this scenario happens. On top of the bug itself, not being able to set a
> timeout while waiting for a connection to be available, makes any application
> unresponsive as it will run out of threads eventually.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]