Question about replication problem

[Fred Drake <fd...@gmail.com>]

I have a multi-core master/slave configuration that's showing
unexpected replication behavior for one core; other cores are
replicating without problems.

The master is running Solr 4.1; one slave is running 4.1 under Tomcat,
and another (for testing) is running 4.6 under Jetty.  These are
exhibiting similar problems.

The core on the master is at generation 4; this is reported as both
the generation for searching and replication using the replication API
on the slave.

When the slave attempts to replicate, it fails to load the file list
required for replication, and logs the message:

    INFO: Master's generation: 4
    INFO: Slave's generation: 5
    INFO: Starting replication process
    SEVERE: No files to download for index generation: 4

(Timestamps removed for readability.)

I have verified that the slave returns empty results where the master
returns non-empty results; so this isn't a matter of reporting being
incorrect.

I'm not sure how to diagnose this; what can I check for to understand
the actual problem?

(I have Googled for relevant issues or mailing list traffic, but
nothing actually explained what's happening or how to correct it.)



  -Fred

-- 
Fred L. Drake, Jr.    <fdrake at gmail.com>
"A storm broke loose in my mind."  --Albert Einstein

Re: Question about replication problem

[Fred Drake <fr...@fdrake.net>]

On Fri, Dec 20, 2013 at 1:33 PM, Mark Miller <ma...@gmail.com> wrote:
> What Solr version?

I've seen the first problem in the thread with Solr 4.1, and the
second with both 4.1 and 4.6.


   -Fred

-- 
Fred L. Drake, Jr.    <fred at fdrake.net>
"A storm broke loose in my mind."  --Albert Einstein

Re: Question about replication problem

[Mark Miller <ma...@gmail.com>]

What Solr version?

- Mark

On Dec 20, 2013, at 1:14 PM, Fred Drake <fd...@gmail.com> wrote:

> Here's another sequence of messages I frequently see where replication
> isn't happening with no clearly identified cause:
> 
> INFO  org.apache.solr.handler.SnapPuller; Starting replication process
> INFO  org.apache.solr.handler.SnapPuller; Master's generation: 6
> INFO  org.apache.solr.handler.SnapPuller; Slave's generation: 1
> INFO  org.apache.solr.handler.SnapPuller; Starting replication process
> INFO  org.apache.solr.handler.SnapPuller; Number of files in latest
> index in master: 94
> ERROR org.apache.solr.handler.SnapPuller; No files to download for
> index generation: 6
> This frequently happens with a new slave, so it's never able to get
> any data for the core.  This doesn't normally happen for *all* cores;
> at least one core will replicate successfully (and continues to stay
> up to date).
> 
> I don't think I've seen this with multiple deployments of single-core
> Solr instances in a container, but it seems odd that multi-core
> configurations would face as many replication problems as I've seen
> unless I'm configuring something incorrectly.
> 
> Are others having better success with multi-core master/slave
> configurations?  Or is this not a well-supported approach to
> configuration?
> 
> 
>  -Fred
> 
> -- 
> Fred L. Drake, Jr.  <fdrake at gmail.com>
> "A storm broke loose in my mind."  --Albert Einstein

Re: Question about replication problem

[Fred Drake <fd...@gmail.com>]

Here's another sequence of messages I frequently see where replication
isn't happening with no clearly identified cause:

INFO  org.apache.solr.handler.SnapPuller; Starting replication process
INFO  org.apache.solr.handler.SnapPuller; Master's generation: 6
INFO  org.apache.solr.handler.SnapPuller; Slave's generation: 1
INFO  org.apache.solr.handler.SnapPuller; Starting replication process
INFO  org.apache.solr.handler.SnapPuller; Number of files in latest
index in master: 94
ERROR org.apache.solr.handler.SnapPuller; No files to download for
index generation: 6
This frequently happens with a new slave, so it's never able to get
any data for the core.  This doesn't normally happen for *all* cores;
at least one core will replicate successfully (and continues to stay
up to date).

I don't think I've seen this with multiple deployments of single-core
Solr instances in a container, but it seems odd that multi-core
configurations would face as many replication problems as I've seen
unless I'm configuring something incorrectly.

Are others having better success with multi-core master/slave
configurations?  Or is this not a well-supported approach to
configuration?


  -Fred

-- 
Fred L. Drake, Jr.  <fdrake at gmail.com>
"A storm broke loose in my mind."  --Albert Einstein