Sorry hit send too fast. The shards were listed as active. Also the solr
instances were still running but the file system they wrote to had become
read only. I thought that would make replication fail and when the issue
was fixed and solr restarted replication would then succeed. Am I hitting
some fringe case?

On Saturday, May 12, 2012, Jamie Johnson <jej2...@gmail.com> wrote:
> I have not tried to reproduce as of yet but hope to do so Monday. The
machine that had the issue was a vm out of my control so I'm not certain
how it was restored. I am using a fairly recent nightly build within the
last few weeks
>
> On Friday, May 11, 2012, Mark Miller <markrmil...@gmail.com> wrote:
>> So it's easy to reproduce? What do you mean restored from a prior state?
>>
>> What snapshot are you on these days for future ref?
>>
>> You have double checked to make sure that shard is listed as ACTIVE
right?
>>
>> On May 11, 2012, at 4:55 PM, Jamie Johnson wrote:
>>
>>> I've had a few instances where a machine has needed to be restored
>>> from a prior state.  After doing so and firing up solr again I've had
>>> instances where replication doesn't seem to be working properly.  I
>>> have not seen any failures in logs (will have to keep a closer eye on
>>> this) but when this happens and I execute a query against each with
>>> distrib=false I am seeing the following counts
>>>
>>> Shard @ host1(shard1) returned 95150
>>> Shard @ host2(shard1) returned 95150
>>> Shard @ host2(shard4) returned 94311
>>> Shard @ host3(shard4) returned 8468
>>> Shard @ host3(shard5) returned 8303
>>> Shard @ host1(shard5) returned 96054
>>> Shard @ host1(shard2) returned 95620
>>> Shard @ host2(shard2) returned 95620
>>> Shard @ host2(shard3) returned 93195
>>> Shard @ host3(shard3) returned 8336
>>> Shard @ host3(shard6) returned 8309
>>> Shard @ host1(shard6) returned 96036
>>>
>>>
>>> in this case host3 is what failed and as you can see everything on
>>> host3 is significantly less than what the leader has.  Has anyone else
>>> experienced this?
>>
>> - Mark Miller
>> lucidimagination.com
>>
>>
>>
>>
>>
>>
>>
>>
>>
>>
>>
>>

Reply via email to