cassandra-commits mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Benjamin Lerer (JIRA)" <j...@apache.org>
Subject [jira] [Updated] (CASSANDRA-9861) When forcibly exiting due to OOM, we should produce a heap dump
Date Fri, 01 Apr 2016 14:17:25 GMT

     [ https://issues.apache.org/jira/browse/CASSANDRA-9861?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]

Benjamin Lerer updated CASSANDRA-9861:
--------------------------------------
    Attachment: 9861-2.2.txt

{quote}I guess the simplest way would be to add the -XX:+HeapDumpOnOutOfMemoryError command
line argument to our startup scripts.{quote}

Sorry, I initially misunderstood the problem.

The patch use {{jmap}} to generate a heap dump when the {{JVMStabilityInspector}} detects
an OOMError. 
One issue is to retrieve the process ID to be able to execute the {{jmap}} call. The Java
API will only provide a reliable way for retrieving the process ID in Java 9. The current
method used by the patch will work on the Oracle JVM but might not work on all the JVMs. If
the process ID cannot be retrieved no dump will be generated.

||utests||dtests||
|[2.2|http://cassci.datastax.com/view/Dev/view/blerer/job/blerer-9861-2.2-testall/]|[2.2|http://cassci.datastax.com/view/Dev/view/blerer/job/blerer-9861-2.2-dtest/]|
|[3.0|http://cassci.datastax.com/view/Dev/view/blerer/job/blerer-9861-3.0-testall/]|[3.0|http://cassci.datastax.com/view/Dev/view/blerer/job/blerer-9861-3.0-dtest/]|


> When forcibly exiting due to OOM, we should produce a heap dump
> ---------------------------------------------------------------
>
>                 Key: CASSANDRA-9861
>                 URL: https://issues.apache.org/jira/browse/CASSANDRA-9861
>             Project: Cassandra
>          Issue Type: Improvement
>            Reporter: Benedict
>            Assignee: Benjamin Lerer
>            Priority: Minor
>              Labels: lhf
>             Fix For: 2.2.x
>
>         Attachments: 9861-2.2.txt
>
>
> CASSANDRA-7507 introduced earlier termination on encountering an OOM, due to lack of
certainty about system state. However a side effect is that we never produce heap dumps on
OOM. We should ideally try to produce one forcibly before exiting.



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)

Mime
View raw message