On 3/4/2016 5:46 AM, Cool Techi wrote:
> We are using solr  4.8.1 cloud and have a single shard core and 3 replicas in 
> total.  We are noticing the following issues,
> 1) Master propagates slow to to the other nodes/replicas.2) This increase 
> over a period of time, i.e the data propagation time increase as the time 
> from which leader is elected increases?
> What could be causing this, our commit settings are given below.
>  <autoCommit> 
>        <maxTime>12000</maxTime> 
>        <maxDocs>25000</maxDocs>
>        <openSearcher>false</openSearcher> 
>      </autoCommit>
>
>     <!-- softAutoCommit is like autoCommit except it causes a
>          'soft' commit which only ensures that changes are visible
>          but does not ensure that data is synced to disk.  This is
>          faster and more near-realtime friendly than a hard commit.
>       -->
>        <autoSoftCommit> 
>          <maxTime>1000</maxTime> 
>        </autoSoftCommit>

There's very little information to go on here.

This information will help troubleshoot performance problems:

* How many Solr instances are running?
* How many machines are running those Solr instances?
* What is the max heap setting on each of those instances?
* Are you running with the included jetty (start.jar) or a separate
container?
* How big is each replica in terms of disk space?
* How many documents are in each replica?
* How much memory is installed in each machine?
* How frequently are you sending updates, and how many updates per batch?
* How long is Solr taking to do commits that open a new searcher?
* What is your query rate?

An autoSoftCommit value of one second is a major source of performance
problems.  A commit that opens a new Searcher (which soft commits do)
will usually take longer than one second to occur.

http://lucidworks.com/blog/understanding-transaction-logs-softcommit-and-commit-in-sorlcloud/

The first thing I would try is increasing the maxTime on both autoCommit
and autoSoftCommit.  I would use five minutes for autoCommit.  For
autoSoftCommit, I wouldn't go lower than one minute, and a value of five
to 15 minutes would be better.

Thanks,
Shawn

Reply via email to