You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@kudu.apache.org by "Mike Percy (JIRA)" <ji...@apache.org> on 2016/10/21 10:10:58 UTC

[jira] [Created] (KUDU-1717) Provide tablet-specific context on crash

Mike Percy created KUDU-1717:
--------------------------------

             Summary: Provide tablet-specific context on crash
                 Key: KUDU-1717
                 URL: https://issues.apache.org/jira/browse/KUDU-1717
             Project: Kudu
          Issue Type: Bug
          Components: supportability, tablet
            Reporter: Mike Percy


It would be helpful to add tablet-specific crash reporting information for cases where a particular tablet is corrupted and this results in a tablet server crash.

An idea from Todd is as follows:

For example, I've often found in these kinds of cases that we get a CHECK failure, but we have no idea which tablet caused it. So even if we have this tool, it's not obvious which tablet should be quarantined/deleted to fix the issue, unless we get a core and are able to figure it out by looking up the stack frames. (Even then it can be tricky in release builds).

The idea of the crash context is some macros like:

CRASH_INFO("tablet", tablet_id_);

which would register as part of a threadlocal map/stack of sorts, and then a SEGV handler could dump such information before exiting.




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