[jira] Commented: (TS-170) hostdb cores in reply_to_cont() when running out of space

["John Plevyak (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/TS-170?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12833210#action_12833210 ] 

John Plevyak commented on TS-170:
---------------------------------

This does fix the crash, but it doesn't prevent the HostDB Heap from filling.
One solution for that would be to have another water mark at say 90% where
you run through and decimate (delete 1 in 10) hostdb entries which use heap
space at the start of HeapGC.

This would prevent the Heap from overfilling and returning a failure back to
the HTTP state machine which could result in a page not getting returned.

However, this patch + a much larger hostdb size will probably prevent that
in most real world situations.

> hostdb cores in reply_to_cont() when running out of space
> ---------------------------------------------------------
>
>                 Key: TS-170
>                 URL: https://issues.apache.org/jira/browse/TS-170
>             Project: Traffic Server
>          Issue Type: Bug
>            Reporter: Bryan Call
>            Assignee: Bryan Call
>            Priority: Critical
>             Fix For: 2.0.0a
>
>         Attachments: hostdb_bcall_001.diff
>
>
> In HostDBContinuation::lookup_done() if we can't allocate memory in the cache we will free some space (delete_block())
> and then return a NULL.  Also, the error message "out of room in hostdb for reverse DNS data" gets printed
> out.
> This NULL pointer will be dereferenced later when we call reply_to_cont().
> bz 3191464

-- 
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.