lucene-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Mark Miller (JIRA)" <>
Subject [jira] [Commented] (SOLR-4519) corrupt tlog causes fullCopy download index files every time reboot a node
Date Fri, 01 Mar 2013 13:59:12 GMT


Mark Miller commented on SOLR-4519:

bq. the tlog should be fixed.

Currently, when you replicate, you get nothing in the tlog. Yonik has brought up perhaps doing
a little trick on replication to populate the tlog a bit, but nothing has been started on
that front. So once you replicate, unless some docs are then added, the next fail will require
another replication.

However, we may actually be able to take advantage of replication itself noticing that it
doesn't need to do a full replicate. Currently, in SolrCloud we force a replication every
time no matter what when we call replicate - now that std replication has had some bugs fixed
and has better tests, we may not have to force that anymore - and so the next full replication
would not actually move any files.
> corrupt tlog causes fullCopy download index files every time reboot a node
> --------------------------------------------------------------------------
>                 Key: SOLR-4519
>                 URL:
>             Project: Solr
>          Issue Type: Bug
>    Affects Versions: 4.0
>         Environment: The solrcloud is implemented on three servers. There are three solr
instance on each server. The collection has three shards. Every shard has three replica. Replicas
in same shard run in solr instance on different server.
>            Reporter: Simon Scofield
> There are two questions:
> 1. The tlog of one replica of shard1 is damaged by some reason. We are still looking
for the reason. Please give some clue if you are familia with this problem.
> 2. The error replica successed to recovery by fullcopy download index files from leader.
Then I killed the instance and started it again, the recovery process still is fullcopy download.
In my opinion, after the first time fullcopy recovery, the tlog should be fixed. Here is some
> 2013-02-28 15:04:58,622 INFO - Core needs to recover:metadata
> 2013-02-28 15:04:58,622 INFO org.apache.solr.update.DefaultSolrCoreState:214 - Running
recovery - first canceling any ongoing recovery
> 2013-02-28 15:04:58,625 INFO - Starting recovery
process.  core=metadata recoveringAfterStartup=true
> 2013-02-28 15:04:58,626 INFO - Updating
cloud state from ZooKeeper...
> 2013-02-28 15:04:58,628 ERROR org.apache.solr.update.UpdateLog:957 - Exception reading
versions from log
>         at org.apache.solr.common.util.FastInputStream.readUnsignedByte(
>         at org.apache.solr.common.util.FastInputStream.readInt(
>         at org.apache.solr.update.TransactionLog$
>         at org.apache.solr.update.UpdateLog$RecentUpdates.update(
>         at org.apache.solr.update.UpdateLog$RecentUpdates.access$000(
>         at org.apache.solr.update.UpdateLog.getRecentUpdates(
>         at
>         at
> 2013-02-28 15:05:01,857 INFO - Begin buffering
updates. core=metadata
> 2013-02-28 15:05:01,857 INFO org.apache.solr.update.UpdateLog:1015 - Starting to buffer
updates. FSUpdateLog{state=ACTIVE, tlog=null}
> 2013-02-28 15:05:01,857 INFO - Attempting
to replicate from core=metadata
> 2013-02-28 15:05:02,882 INFO org.apache.solr.handler.SnapPuller:305 - Master's generation:
> 2013-02-28 15:05:02,882 INFO org.apache.solr.handler.SnapPuller:306 - Slave's generation:
> 2013-02-28 15:05:02,882 INFO org.apache.solr.handler.SnapPuller:307 - Starting replication
> 2013-02-28 15:05:02,893 INFO org.apache.solr.handler.SnapPuller:312 - Number of files
in latest index in master: 422
> 2013-02-28 15:05:02,897 INFO org.apache.solr.handler.SnapPuller:325 - Starting download
to /solr/nodes/node1/bin/../solr/metadata/data/index.20130228150502893 fullCopy=true
> 2013-02-28 15:33:55,848 INFO org.apache.solr.handler.SnapPuller:334 - Total time taken
for download : 1732 secs (The size of index files is 94G)

This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see:

To unsubscribe, e-mail:
For additional commands, e-mail:

View raw message