You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@lucene.apache.org by Ted Cao <te...@gmail.com> on 2015/06/30 19:22:47 UTC

solr 4.5 replicated slaves don't delete old files in /index directory

Hi, I hope someone is aware of this bug and could point me to
either
old Jira ticket
or advice on how it can be safely fixed ...

In our production environment, with solr 4.5, we often see slave
replicating down
data from upstream into /index directory but NOT deleting older segment
files ...

We have enormous index, 1T+, so that adds to our problem, because we
sometimes end up with 2T /index directory when fully optimized index is
coming down ... the problem usually resolves itself in the next replication
(or 2), but once we run out of disk, it gets in a vicious cycle

Replicating this issue in dev/staging is very tricky so I hope someone
knows this issue already.

Possible solutions
1. manually deleting old segment files (need to write some code and don't
know if safe)
2. would restart or issuing commits trigger deletion of the old files?

Thanks, much appreciated!!!

Re: solr 4.5 replicated slaves don't delete old files in /index directory

Posted by "david.w.smiley@gmail.com" <da...@gmail.com>.
Ted,
Please ask again on the solr-user list; this list is for development of
Lucene/Solr.
Good luck,
~ David

On Tue, Jun 30, 2015 at 1:23 PM Ted Cao <te...@gmail.com> wrote:

> Hi, I hope someone is aware of this bug and could point me to
> either
> old Jira ticket
> or advice on how it can be safely fixed ...
>
> In our production environment, with solr 4.5, we often see slave
> replicating down
> data from upstream into /index directory but NOT deleting older segment
> files ...
>
> We have enormous index, 1T+, so that adds to our problem, because we
> sometimes end up with 2T /index directory when fully optimized index is
> coming down ... the problem usually resolves itself in the next replication
> (or 2), but once we run out of disk, it gets in a vicious cycle
>
> Replicating this issue in dev/staging is very tricky so I hope someone
> knows this issue already.
>
> Possible solutions
> 1. manually deleting old segment files (need to write some code and don't
> know if safe)
> 2. would restart or issuing commits trigger deletion of the old files?
>
> Thanks, much appreciated!!!
>