lucene-solr-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Erick Erickson <erickerick...@gmail.com>
Subject Re: Planning Solr migration to production: clean and autoSoftCommit
Date Fri, 10 Jul 2015 17:36:32 GMT
You're confusing a couple of things here.

First, I'm assuming that you are NOT using SolrCloud, but older-style
master/slave.
If that's not true, disregard the rest of this.

autoSoftCommit is _local_ and has nothing to do with changing the
Lucene segments.
And since you're not searching on the master, you might as well set
this to -1. It
really has no effect on the slaves since they're not doing any indexing.

Slaves periodically poll the master to see if the index has changed. What you're
describing is not having that process occur until after DIH is done in
order not to
get partial views of the data. What I'd do is just use the replication
API to force
replication after you know DIH is done. Especially in a situation
where re-indexing
is rare, this may be your best option. See:
https://cwiki.apache.org/confluence/display/solr/Index+Replication

Or just disable polling on the slaves, do the DIH thing, then re-enable polling.

Best,
Erick

On Fri, Jul 10, 2015 at 10:21 AM, wwang525 <wwang525@gmail.com> wrote:
> Hi,
>
> The following questions are about the basic configuration options in
> production.
>
> We will have three machines: one indexing instance (master) and two Solr
> instances (in different machines) for searching purpose. This way, we will
> always have two Solr instances dedicated for executing search requests.
>
> Right now, we are only considering re-build full index every once in a
> while, so there will be no incremental indexing.
>
> I understand that the indexing instance can have the indexing parameter
> "clean" to be set as true or false. If I set it as true, the search index in
> the indexing instance will be cleaned up and anytime when I check the index,
> it is going to grow.
>
> The question is :
>
> (1) Will the slave instance (for executing requests) get in sync with the
> master if we set the "clean" to true? This is not what we would like it to
> be since the search index will be clean up and grow. Customers will need to
> wait for some period of time to search for the entire data pending the
> completion of the indexing job
>
> (2) The "autoSoftCommit" is supposed to make the update visible to search. I
> also configured "autoSoftCommit" in solrconfig.xml in the master. When I set
> the "clean" to true in the indexing job, what is the impact of this
> parameter to the search requests executed in slave machine?
>
> Thanks
>
>
>
>
>
>
> --
> View this message in context: http://lucene.472066.n3.nabble.com/Planning-Solr-migration-to-production-clean-and-autoSoftCommit-tp4216736.html
> Sent from the Solr - User mailing list archive at Nabble.com.

Mime
View raw message