Mailing-List: contact dev-help@lucene.apache.org; run by ezmlm
Precedence: bulk
Reply-To: dev@lucene.apache.org
Date: Wed, 18 Apr 2012 23:14:40 +0000 (UTC)
From: "Erick Erickson (Updated) (JIRA)" <jira@apache.org>
To: dev@lucene.apache.org
Message-ID: 
 <2015965523.3830.1334790880337.JavaMail.tomcat@hel.zones.apache.org>
In-Reply-To: 
 <460680525.3822.1334790641069.JavaMail.tomcat@hel.zones.apache.org>
Subject: [jira] [Updated] (SOLR-3376) SolrCloud: Specifying shardId not
 working correctly, although the failures are inconsistent.
MIME-Version: 1.0
Content-Type: text/plain; charset=utf-8
Content-Transfer-Encoding: 7bit


     [ https://issues.apache.org/jira/browse/SOLR-3376?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Erick Erickson updated SOLR-3376:
---------------------------------

    Description: 
I'm seeing some odd results when specifying "shardId" parameter. I'm trying the 4-node, 2-shard example from the Wiki and specifying shardIds like this:

dir       shardId     start order    runnng ZK   port
example     1           1               y        8983
example2    2           2               y        7574
example3    1           3               y        8900
example4    2           4               y        7500 

And I'm waiting a bit between starting various examples to let ZK settle down.

Once all of them are started, I was looking at http://localhost:8983/solr/#/~cloud?view=graph to check out what that looks like (pretty cool IMO, especially since I didn't have to do it). The problem was that shard 2 only reported a single instance, while shard1 showed the two instances I was expecting. I'm running with 3 embedded ZK instances, just for yucks. Interestingly the node that didn't show up was the only node that was NOT running ZK.

When I removed all the "shardId" parameters, nuked zoo_data from all directories and just started them up (with numShards=2 on the bootstrap ZK node), all 4 nodes showed up just fine.

When starting with shardId specified and trying to go straight to the admin interface on the node that wasn't showing up, I'd get odd errors like "This interface requires that you activate the admin request handlers, add the following configuration to your solrconfig.xml:". I also couldn't search directly on that machine, "http://localhost:7574/solr/select?q=*:*" returns a 404 error.

Command starting server that's giving me trouble: java -Xmx1G -Djetty.port=7500 -DzkHost=localhost:9983,localhost:8574,localhost:9900 -DshardId=2 -jar start.jar

Command for one that works fine:                   java -Xmx1G -Djetty.port=8900 -DzkRun -DzkHost=localhost:9983,localhost:8574,localhost:9900 -DshardId=1 -jar start.jar


Sami Siren and he reports similar issues via e-mail conversation. Sami says that ZK 3.3.5 apparently (without exhaustive tests) fixed the problem for him, but when I tried ZK 3.3.5 I saw the same issue. Of course with all the recent stuff with Ivy, I may have screwed up when/where the JARs were.

So then I went back to ZK 3.3.4 and couldn't reproduce the problem. Which seems highly suspicious to me. It was failing every time before with 3.3.4, so it sounds like gremlins.

And then I tried ZK 3.3.5 again (changed the ivy.xml in solrj, blew away the ZK 3.3.4, rebuilt, removed zoo_data, recopied example to three other directories)  and it works fine there too now. Siiiiggggh. Mostly this is a placeholder to insure we try this, I guarantee that sys admins will want to assign specific machines to specific shards, so this'll get used.

  was:

I'm seeing some odd results when specifying "shardId" parameter. I'm trying the 4-node, 2-shard example from the Wiki and specifying shardIds like this:

dir       shardId     start order    runnng ZK   port
example     1           1               y        8983
example2    2           2               y        7574
example3    1           3               y        8900
example4    2           4               y        7500 

And I'm waiting a bit between starting various examples to let ZK settle down.

Once all of them are started, I was looking at http://localhost:8983/solr/#/~cloud?view=graph to check out what that looks like (pretty cool IMO, especially since I didn't have to do it). The problem was that shard 2 only reported a single instance, while shard1 showed the two instances I was expecting. I'm running with 3 embedded ZK instances, just for yucks. Interestingly the node that didn't show up was the only node that was NOT running ZK.

When I removed all the "shardId" parameters, nuked zoo_data from all directories and just started them up (with numShards=2 on the bootstrap ZK node), all 4 nodes showed up just fine.

When starting with shardId specified and trying to go straight to the admin interface on the node that wasn't showing up, I'd get odd errors like "This interface requires that you activate the admin request handlers, add the following configuration to your solrconfig.xml:". I also couldn't search directly on that machine, "http://localhost:7574/solr/select?q=*:*" returns a 404 error.

Command starting server that's giving me trouble: java -Xmx1G -Djetty.port=7500 -DzkHost=localhost:9983,localhost:8574,localhost:9900 -DshardId=2 -jar start.jar

Command for one that works fine:                   java -Xmx1G -Djetty.port=8900 -DzkRun -DzkHost=localhost:9983,localhost:8574,localhost:9900 -DshardId=1 -jar start.jar


Sami Siren and he reports similar issues via e-mail conversation. Sami says that ZK 3.5 apparently (without exhaustive tests) fixed the problem for him, but when I tried ZK 3.5 I saw the same issue. Of course with all the recent stuff with Ivy, I may have screwed up when/where the JARs were.

So then I went back to ZK 3.4 and couldn't reproduce the problem. Which seems highly suspicious to me. It was failing every time before with 3.4, so it sounds like gremlins.

And then I re-did the stuff with ZK 3.5 and it works fine there too now. Siiiiggggh.   Mostly this is a placeholder to insure we try this, I guarantee that sys admins will want to assign specific machines to specific shards, so this'll get used.

    
> SolrCloud: Specifying shardId not working correctly, although the failures are inconsistent.
> --------------------------------------------------------------------------------------------
>
>                 Key: SOLR-3376
>                 URL: https://issues.apache.org/jira/browse/SOLR-3376
>             Project: Solr
>          Issue Type: Bug
>          Components: SolrCloud
>    Affects Versions: 4.0
>            Reporter: Erick Erickson
>
> I'm seeing some odd results when specifying "shardId" parameter. I'm trying the 4-node, 2-shard example from the Wiki and specifying shardIds like this:
> dir       shardId     start order    runnng ZK   port
> example     1           1               y        8983
> example2    2           2               y        7574
> example3    1           3               y        8900
> example4    2           4               y        7500 
> And I'm waiting a bit between starting various examples to let ZK settle down.
> Once all of them are started, I was looking at http://localhost:8983/solr/#/~cloud?view=graph to check out what that looks like (pretty cool IMO, especially since I didn't have to do it). The problem was that shard 2 only reported a single instance, while shard1 showed the two instances I was expecting. I'm running with 3 embedded ZK instances, just for yucks. Interestingly the node that didn't show up was the only node that was NOT running ZK.
> When I removed all the "shardId" parameters, nuked zoo_data from all directories and just started them up (with numShards=2 on the bootstrap ZK node), all 4 nodes showed up just fine.
> When starting with shardId specified and trying to go straight to the admin interface on the node that wasn't showing up, I'd get odd errors like "This interface requires that you activate the admin request handlers, add the following configuration to your solrconfig.xml:". I also couldn't search directly on that machine, "http://localhost:7574/solr/select?q=*:*" returns a 404 error.
> Command starting server that's giving me trouble: java -Xmx1G -Djetty.port=7500 -DzkHost=localhost:9983,localhost:8574,localhost:9900 -DshardId=2 -jar start.jar
> Command for one that works fine:                   java -Xmx1G -Djetty.port=8900 -DzkRun -DzkHost=localhost:9983,localhost:8574,localhost:9900 -DshardId=1 -jar start.jar
> Sami Siren and he reports similar issues via e-mail conversation. Sami says that ZK 3.3.5 apparently (without exhaustive tests) fixed the problem for him, but when I tried ZK 3.3.5 I saw the same issue. Of course with all the recent stuff with Ivy, I may have screwed up when/where the JARs were.
> So then I went back to ZK 3.3.4 and couldn't reproduce the problem. Which seems highly suspicious to me. It was failing every time before with 3.3.4, so it sounds like gremlins.
> And then I tried ZK 3.3.5 again (changed the ivy.xml in solrj, blew away the ZK 3.3.4, rebuilt, removed zoo_data, recopied example to three other directories)  and it works fine there too now. Siiiiggggh. Mostly this is a placeholder to insure we try this, I guarantee that sys admins will want to assign specific machines to specific shards, so this'll get used.

--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators: https://issues.apache.org/jira/secure/ContactAdministrators!default.jspa
For more information on JIRA, see: http://www.atlassian.com/software/jira

        
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@lucene.apache.org
For additional commands, e-mail: dev-help@lucene.apache.org