[
https://issues.apache.org/jira/browse/COUCHDB-2217?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13956552#comment-13956552
]
Dave Cottlehuber commented on COUCHDB-2217:
-------------------------------------------
This is failing in here:
foldl(Fd, [Pos], Md5, Md5Acc, Fun, Acc) ->
{ok, Bin} = couch_file:pread_iolist(Fd, Pos),
Md5 = couch_util:md5_final(couch_util:md5_update(Md5Acc, Bin)),
Fun(Bin, Acc);
which implies that the calculated md5 doesn't match the supplied one for a
given doc/attachment. If replication stops at exactly the same point, likely
corrupted disk is the issue. It should be possible to compare update sequence
in the replication to work out which doc is failing.
Note to devs: would it make sense to have a worker process track a list of docs
that it's handling? so at least if it dies, we'll know which docs were underway
in the error trace.
> Replication Crash
> -----------------
>
> Key: COUCHDB-2217
> URL: https://issues.apache.org/jira/browse/COUCHDB-2217
> Project: CouchDB
> Issue Type: Bug
> Security Level: public(Regular issues)
> Components: HTTP Interface, Replication
> Reporter: Yang Zhewei
> Attachments: log.txt
>
>
> I want to run a replication to copy a database on the source server to the
> target server. The source server is Windows 7 OS, and the target server is
> Linux OS. The size of database is about 14GB.
> Everytime, when the replication runs and complete about 8.6GB, then it will
> crash. And there is no log on the target server.
> The attach file is the error message on the source server.
--
This message was sent by Atlassian JIRA
(v6.2#6252)
