lucene-java-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Michael McCandless" <>
Subject Re: Recovering from a Crash
Date Wed, 25 Jul 2007 09:49:41 GMT
"Simon Wistow" <> wrote:
> On Wed, Jul 25, 2007 at 05:19:31AM -0400, Michael McCandless said:
> > It's somewhat spooky that you have a write.lock present because that
> > means you backed up while a writer was actively writing to the index
> > which is a bit dangerous because if the timing is unlucky (backup does
> > an "ls" but before it can copy the segments_N file a commit has
> > happened) you could fail to copy a segments_N file.  It's best to
> > either pause the writer for backpus to occur (simplest) or make a
> > custom deletion policy that safely allows the backup to slowly copy
> > even while indexing is continuing (advanced).
> Sorry, I should have been clearer - I took the backup of the state of
> the index when the machine restarted after the crash. I did have another
> backup from a day or so ago but I was hoping to not have to reindex a
> days worth of data (which is alot).

Ahhh, OK.  But do you have a segments_N file?

> Our backup strategy is currently -
> 1) Stop the writer (and let write tasks queue up)
> 2) cp -lr indexes indexes-<date>
> 3) Restart the writer
> Is this something approximating best practice?

Yes, this is perfect.  This is the "simple" option I described.  The
more complex option is to use a custom deletion policy which enables
you to safely do backups (even if the copy process is slow) without
pausing the write task (indexing).


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

View raw message