You are viewing a plain text version of this content. The canonical link for it is here.
Posted to solr-user@lucene.apache.org by Nathan Neulinger <nn...@neulinger.org> on 2014/01/25 03:52:59 UTC
Replica not consistent after update request?
How can we issue an update request and be certain that all of the replicas in the SolrCloud cluster are up to date?
I found this post:
http://comments.gmane.org/gmane.comp.jakarta.lucene.solr.user/79886
which seems to indicate that all replicas for a shard must finish/succeed before it returns to client that the operation
succeeded - but we've been seeing behavior lately (until we configured automatic soft commits) where the replicas were
almost always "not current" - i.e. the replicas were missing documents/etc.
Is this something wrong with our cloud setup/replication, or am I misinterpreting the way that updates in a cloud
deployment are supposed to function?
If it's a problem with our cloud setup, do you have any suggestions on diagnostics?
Alternatively, are we perhaps just using it wrong?
-- Nathan
------------------------------------------------------------
Nathan Neulinger nneul@neulinger.org
Neulinger Consulting (573) 612-1412
Re: Replica not consistent after update request?
Posted by Nathan Neulinger <nn...@neulinger.org>.
Wow, the detail in that jira issue makes my brain hurt... Great to see it's got a quick answer/fix!
Thank you!
-- Nathan
On 01/24/2014 09:43 PM, Joel Bernstein wrote:
> If you're on Solr 4.6 then this is likely the issue:
> https://issues.apache.org/jira/browse/SOLR-4260.
>
> The issue is resolved for Solr 4.6.1 which should be out next week.
>
>
> Joel Bernstein
> Search Engineer at Heliosearch
>
>
> On Fri, Jan 24, 2014 at 9:52 PM, Nathan Neulinger <nn...@neulinger.org>wrote:
>
>> How can we issue an update request and be certain that all of the replicas
>> in the SolrCloud cluster are up to date?
>>
>> I found this post:
>>
>> http://comments.gmane.org/gmane.comp.jakarta.lucene.
>> solr.user/79886
>>
>> which seems to indicate that all replicas for a shard must finish/succeed
>> before it returns to client that the operation succeeded - but we've been
>> seeing behavior lately (until we configured automatic soft commits) where
>> the replicas were almost always "not current" - i.e. the replicas were
>> missing documents/etc.
>>
>> Is this something wrong with our cloud setup/replication, or am I
>> misinterpreting the way that updates in a cloud deployment are supposed to
>> function?
>>
>> If it's a problem with our cloud setup, do you have any suggestions on
>> diagnostics?
>>
>> Alternatively, are we perhaps just using it wrong?
>>
>> -- Nathan
>>
>> ------------------------------------------------------------
>> Nathan Neulinger nneul@neulinger.org
>> Neulinger Consulting (573) 612-1412
>>
>
--
------------------------------------------------------------
Nathan Neulinger nneul@neulinger.org
Neulinger Consulting (573) 612-1412
Re: Replica not consistent after update request?
Posted by Joel Bernstein <jo...@gmail.com>.
If you're on Solr 4.6 then this is likely the issue:
https://issues.apache.org/jira/browse/SOLR-4260.
The issue is resolved for Solr 4.6.1 which should be out next week.
Joel Bernstein
Search Engineer at Heliosearch
On Fri, Jan 24, 2014 at 9:52 PM, Nathan Neulinger <nn...@neulinger.org>wrote:
> How can we issue an update request and be certain that all of the replicas
> in the SolrCloud cluster are up to date?
>
> I found this post:
>
> http://comments.gmane.org/gmane.comp.jakarta.lucene.
> solr.user/79886
>
> which seems to indicate that all replicas for a shard must finish/succeed
> before it returns to client that the operation succeeded - but we've been
> seeing behavior lately (until we configured automatic soft commits) where
> the replicas were almost always "not current" - i.e. the replicas were
> missing documents/etc.
>
> Is this something wrong with our cloud setup/replication, or am I
> misinterpreting the way that updates in a cloud deployment are supposed to
> function?
>
> If it's a problem with our cloud setup, do you have any suggestions on
> diagnostics?
>
> Alternatively, are we perhaps just using it wrong?
>
> -- Nathan
>
> ------------------------------------------------------------
> Nathan Neulinger nneul@neulinger.org
> Neulinger Consulting (573) 612-1412
>
Re: Replica not consistent after update request?
Posted by Nathan Neulinger <nn...@neulinger.org>.
Ok, so our issue sounds like a combination of not having softCommits properly done, combined with SOLR-4260.
Thanks everyone!
On 01/24/2014 11:04 PM, Erick Erickson wrote:
> Right. There updates are guaranteed to be on the replicas and in their
> transaction logs. That doesn't mean they're searchable, however. For a
> document to be found in a search there must be a commit, either soft,
> or hard with openSearcher=true. Here's a post that outlines all this.
>
>
>
> If you have discrepancies when after commits, that's a problem....
>
> Best,
> Erick
>
> On Fri, Jan 24, 2014 at 8:52 PM, Nathan Neulinger <nn...@neulinger.org> wrote:
>> How can we issue an update request and be certain that all of the replicas
>> in the SolrCloud cluster are up to date?
>>
>> I found this post:
>>
>> http://comments.gmane.org/gmane.comp.jakarta.lucene.solr.user/79886
>>
>> which seems to indicate that all replicas for a shard must finish/succeed
>> before it returns to client that the operation succeeded - but we've been
>> seeing behavior lately (until we configured automatic soft commits) where
>> the replicas were almost always "not current" - i.e. the replicas were
>> missing documents/etc.
>>
>> Is this something wrong with our cloud setup/replication, or am I
>> misinterpreting the way that updates in a cloud deployment are supposed to
>> function?
>>
>> If it's a problem with our cloud setup, do you have any suggestions on
>> diagnostics?
>>
>> Alternatively, are we perhaps just using it wrong?
>>
>> -- Nathan
>>
>> ------------------------------------------------------------
>> Nathan Neulinger nneul@neulinger.org
>> Neulinger Consulting (573) 612-1412
--
------------------------------------------------------------
Nathan Neulinger nneul@neulinger.org
Neulinger Consulting (573) 612-1412
Re: Replica not consistent after update request?
Posted by Erick Erickson <er...@gmail.com>.
Right. There updates are guaranteed to be on the replicas and in their
transaction logs. That doesn't mean they're searchable, however. For a
document to be found in a search there must be a commit, either soft,
or hard with openSearcher=true. Here's a post that outlines all this.
If you have discrepancies when after commits, that's a problem....
Best,
Erick
On Fri, Jan 24, 2014 at 8:52 PM, Nathan Neulinger <nn...@neulinger.org> wrote:
> How can we issue an update request and be certain that all of the replicas
> in the SolrCloud cluster are up to date?
>
> I found this post:
>
> http://comments.gmane.org/gmane.comp.jakarta.lucene.solr.user/79886
>
> which seems to indicate that all replicas for a shard must finish/succeed
> before it returns to client that the operation succeeded - but we've been
> seeing behavior lately (until we configured automatic soft commits) where
> the replicas were almost always "not current" - i.e. the replicas were
> missing documents/etc.
>
> Is this something wrong with our cloud setup/replication, or am I
> misinterpreting the way that updates in a cloud deployment are supposed to
> function?
>
> If it's a problem with our cloud setup, do you have any suggestions on
> diagnostics?
>
> Alternatively, are we perhaps just using it wrong?
>
> -- Nathan
>
> ------------------------------------------------------------
> Nathan Neulinger nneul@neulinger.org
> Neulinger Consulting (573) 612-1412
Re: Replica not consistent after update request?
Posted by Nathan Neulinger <nn...@neulinger.org>.
It's 4.6.0. Pair of servers with an external 3-node zk ensemble.
SOLR-4260 looks like a very promising answer. Will check it out as soon as 4.6.1 is released.
May also check out the nightly builds since this is still just development/prototype usage.
-- Nathan
On 01/24/2014 09:45 PM, Anshum Gupta wrote:
> Hi Nathan,
>
> It'd be great to have more information about your setup, Solr Version?
> Depending upon your version, you might want to also look at:
> https://issues.apache.org/jira/browse/SOLR-4260 (which is now fixed).
>
>
> On Fri, Jan 24, 2014 at 6:52 PM, Nathan Neulinger <nn...@neulinger.org>wrote:
>
>> How can we issue an update request and be certain that all of the replicas
>> in the SolrCloud cluster are up to date?
>>
>> I found this post:
>>
>> http://comments.gmane.org/gmane.comp.jakarta.lucene.
>> solr.user/79886
>>
>> which seems to indicate that all replicas for a shard must finish/succeed
>> before it returns to client that the operation succeeded - but we've been
>> seeing behavior lately (until we configured automatic soft commits) where
>> the replicas were almost always "not current" - i.e. the replicas were
>> missing documents/etc.
>>
>> Is this something wrong with our cloud setup/replication, or am I
>> misinterpreting the way that updates in a cloud deployment are supposed to
>> function?
>>
>> If it's a problem with our cloud setup, do you have any suggestions on
>> diagnostics?
>>
>> Alternatively, are we perhaps just using it wrong?
>>
>> -- Nathan
>>
>> ------------------------------------------------------------
>> Nathan Neulinger nneul@neulinger.org
>> Neulinger Consulting (573) 612-1412
>>
>
>
>
--
------------------------------------------------------------
Nathan Neulinger nneul@neulinger.org
Neulinger Consulting (573) 612-1412
Re: Replica not consistent after update request?
Posted by Anshum Gupta <an...@anshumgupta.net>.
Hi Nathan,
It'd be great to have more information about your setup, Solr Version?
Depending upon your version, you might want to also look at:
https://issues.apache.org/jira/browse/SOLR-4260 (which is now fixed).
On Fri, Jan 24, 2014 at 6:52 PM, Nathan Neulinger <nn...@neulinger.org>wrote:
> How can we issue an update request and be certain that all of the replicas
> in the SolrCloud cluster are up to date?
>
> I found this post:
>
> http://comments.gmane.org/gmane.comp.jakarta.lucene.
> solr.user/79886
>
> which seems to indicate that all replicas for a shard must finish/succeed
> before it returns to client that the operation succeeded - but we've been
> seeing behavior lately (until we configured automatic soft commits) where
> the replicas were almost always "not current" - i.e. the replicas were
> missing documents/etc.
>
> Is this something wrong with our cloud setup/replication, or am I
> misinterpreting the way that updates in a cloud deployment are supposed to
> function?
>
> If it's a problem with our cloud setup, do you have any suggestions on
> diagnostics?
>
> Alternatively, are we perhaps just using it wrong?
>
> -- Nathan
>
> ------------------------------------------------------------
> Nathan Neulinger nneul@neulinger.org
> Neulinger Consulting (573) 612-1412
>
--
Anshum Gupta
http://www.anshumgupta.net