To reply to your original question, when you soft commit
the top-level caches are thrown away. I.e. the filterCache,
documentResultCache, all the ones in solrconfig.xml.

And if you have a high autowarm count on them, you wind
up doing a lot of work for no gain. Say your soft commit
interval is 1 second. Only queries that come in during that
one second even _potentially_ use the caches.

Here's a long blog with lots of background:
http://searchhub.org/2013/08/23/understanding-transaction-logs-softcommit-and-commit-in-sorlcloud/

Try this:
1> set your soft commit interval to 1
2> set your cache sizes in solrconfig to 5
3> set your autowarm counts in <2> to 0.

try it. If you see unacceptable degradation in query performance,
then this is too aggressive and you need some caching.
If not, don't bother caching.

As always, it's a tradeoff between how fast docs are searchable
and how much you can improve things with caching.

Best,
Erick


On Mon, Oct 28, 2013 at 6:42 AM, Mugoma Joseph O. <mug...@yengas.com> wrote:

> Hello,
>
> > How do you add the documents to the index - one by one, batches of n ?
>
> Documents are added one by one using solrj
>
> > When do you do your commits ?
>
> We have the following settings in solrconfig.xml:
>
>
>      <autoCommit>
>        <maxTime>1800000</maxTime>
>        <openSearcher>false</openSearcher>
>      </autoCommit>
>
>
>        <autoSoftCommit>
>          <maxTime>15000</maxTime>
>        </autoSoftCommit>
>
>
>
> Thanks.
>
> Mugoma.
>
>
> On Mon, October 28, 2013 12:22 pm, michael.boom wrote:
> > How do you add the documents to the index - one by one, batches of n ?
> > When
> > do you do your commits ?
> > Because 8k docs per day is not a lot. Depending on the above, commiting
> > with
> > softCommit=true might also be a solution.
> >
> >
> >
> > -----
> > Thanks,
> > Michael
> > --
> > View this message in context:
> >
> http://lucene.472066.n3.nabble.com/Optimal-interval-for-soft-commit-tp4098016p4098022.html
> > Sent from the Solr - User mailing list archive at Nabble.com.
> >
>
>
>

Reply via email to