[
https://issues.apache.org/jira/browse/KAFKA-1112?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13823829#comment-13823829
]
Jay Kreps commented on KAFKA-1112:
----------------------------------
Jun, this is true.
However, if you think about it recovery of the log has the same problem. We
read a message and then compare it to its CRC. The CRC is a 32 bit number. The
crc could certainly match the message by chance.
In this case we compare to a 64 bit number so this should be less likely. But
in reality there are many rare events here: (1) we hard crash, (2) hard crash
leads to corruption, (3) corruption of index points to a location that exactly
matches the recovery offset.
In general I think peoples concern with this approach is that it is just kind
of hacky. I agree with this complaint and am sort of disappointed with this set
of changes overall.
I will post a slightly more paranoid version of the check, and then let's
discuss that.
> broker can not start itself after kafka is killed with -9
> ---------------------------------------------------------
>
> Key: KAFKA-1112
> URL: https://issues.apache.org/jira/browse/KAFKA-1112
> Project: Kafka
> Issue Type: Bug
> Components: log
> Affects Versions: 0.8, 0.8.1
> Reporter: Kane Kim
> Assignee: Jay Kreps
> Priority: Critical
> Attachments: KAFKA-1112-v1.patch, KAFKA-1112-v2.patch, KAFKA-1112.out
>
>
> When I kill kafka with -9, broker cannot start itself because of corrupted
> index logs. I think kafka should try to delete/rebuild indexes itself without
> manual intervention.
--
This message was sent by Atlassian JIRA
(v6.1#6144)
