[
https://issues.apache.org/jira/browse/CAMEL-25003?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=18118894#comment-18118894
]
shashank commented on CAMEL-25003:
----------------------------------
PR: https://github.com/apache/camel/pull/26861
_Claude Code on behalf of allthingssecurity_
> camel-core - ServicePool hands out a stopped non-singleton producer after LRU
> eviction, and stops producers in use
> ------------------------------------------------------------------------------------------------------------------
>
> Key: CAMEL-25003
> URL: https://issues.apache.org/jira/browse/CAMEL-25003
> Project: Camel
> Issue Type: Bug
> Components: camel-core
> Reporter: shashank
> Priority: Minor
>
> For endpoints whose producer is not a singleton ({{isSingletonProducer() ==
> false}}: camel-ftp/sftp/ftps, camel-smb, camel-ssh, and the polling consumers
> used by pollEnrich), {{ServicePool}} keeps one {{MultiplePool}} per endpoint
> (a queue of idle producers). It also keeps an LRU cache, of size
> {{cacheSize}}, of every producer it handed out.
> When the LRU evicts such a producer, {{onEvict}} only adds it to the pool's
> {{evicts}} list. The next {{acquire}} or {{release}} on that pool runs
> {{cleanupEvicts}}, which stops the producer. But cleanupEvicts does not take
> the producer out of the pool queue:
> # If the producer was idle, the {{acquire}} that polls the queue right after
> {{cleanupEvicts}} hands it out, stopped.
> # If the producer was in use by another exchange, it is stopped under that
> exchange. When that exchange releases it, {{release}} offers it back to the
> queue, and it is handed out again, stopped.
> Any workload that uses more producer instances than {{cacheSize}} evicts
> producers that are idle in a queue or in use.
> Reproduced against the real classes, using a test component whose producers
> are non-singleton and which records use after stop. With
> {{from("direct:in").toD("pooled:${header.to}", 2)}}, 4 threads and 2000
> exchanges to two endpoints:
> {noformat}
> created=222 stopped=222 usedAfterStop=1664 stopWhileInUse=1
> {noformat}
> 1664 of 2000 exchanges were sent with a stopped producer.
> Handing out a stopped producer is a regression from CAMEL-20829 (4.7), which
> removed the {{queue.remove(evict)}} from {{cleanupEvicts}}.
> Found with a TLA+ model of ServicePool (pooled producers, LRU eviction,
> concurrent acquire/release). The invariants "never used after stop", "never
> stopped in use" and "queue holds only started producers" are violated.
> Proposed fix, in {{ServicePool.MultiplePool}}:
> * {{evict}} takes an idle producer out of the queue before it is stopped, and
> leaves a producer in use alone.
> * {{release}} stops a producer that was evicted while in use, instead of
> offering it back to the queue.
> * {{onEvict}} no longer stops a producer whose pool is gone. At that point
> the producer can only be in use, and {{release}} stops it, as it does since
> CAMEL-24248.
> _Filed with Claude Code on behalf of allthingssecurity._
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
