[
https://issues.apache.org/jira/browse/CASSANDRA-7579?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Joshua McKenzie updated CASSANDRA-7579:
---------------------------------------
Attachment: 7579_v1.txt
v1 attached, based off of branch for CASSANDRA-7927 ([branch
here|https://github.com/josh-mckenzie/cassandra/tree/7927]). Took a similar
approach to CASSANDRA-7507.
I've confirmed that both FD exhaustion from ulimit and SocketException failure
are caught by the JVMStabilityInspector's logic. The current unit test
explicitly creates the exception types w/messages we're looking for rather than
attempting to reproduce the actual exhaustion as ulimits on machines may very
and we can't rely on them being limited to trigger a failure in a unit test.
I've also gone through and patched up any places where we're catching a blanket
"Exception" and handed off to the StabilityInspector.
> File descriptor exhaustion can lead to unreliable state in exception condition
> ------------------------------------------------------------------------------
>
> Key: CASSANDRA-7579
> URL: https://issues.apache.org/jira/browse/CASSANDRA-7579
> Project: Cassandra
> Issue Type: New Feature
> Reporter: Joshua McKenzie
> Assignee: Joshua McKenzie
> Priority: Minor
> Fix For: 2.1.2
>
> Attachments: 7579_v1.txt
>
>
> If the JVM runs out of file descriptors we can get into an unreliable state
> (similar to CASSANDRA-7507 on OOM) where we cannot trust our shutdown hook to
> run successfully to completion. We need to check IOExceptions and
> appropriate Throwable to see if we have a FileNotFoundException w/message
> "Too many files open" and forcefully shutdown the Daemon in these cases.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
