Re: LeaderElection bugfix

2012-06-27 Thread Sami Siren
On Wed, Jun 27, 2012 at 10:32 AM, Trym R. Møller t...@sigmat.dk wrote:
 Hi

Hi,

 The behaviour of this can be verified using the below test in the
 org.apache.solr.cloud.LeaderElectionIntegrationTest

Can you reproduce the failure in your test every time or just rarely?
I added the test method to LeaderElectionIntegrationTest and ran it
few times but I can't get it to fail.

--
 Sami Siren


Re: LeaderElection bugfix

2012-06-27 Thread Trym R. Møller

Hi Sami

Thanks for your rapid reply.

Regarding 1) This seems to be time dependent but it is seen on my local 
windows running the unit test and on a linux server running Solr.
Regarding 2) The test does not show the number of Watchers are 
increasing, but this can be observed either by dumping the memory from 
the jvm or by looking at the debug statements (if debug is enabled).


I don't know how to make assert statements regarding the number of 
watchers in zookeeper, so the test is not quite informative, but more 
confirming that the fix doesn't destroy anything.


Best regards Trym

Den 27-06-2012 10:06, Sami Siren skrev:

On Wed, Jun 27, 2012 at 10:32 AM, Trym R. Møller t...@sigmat.dk wrote:

Hi

Hi,


The behaviour of this can be verified using the below test in the
org.apache.solr.cloud.LeaderElectionIntegrationTest

Can you reproduce the failure in your test every time or just rarely?
I added the test method to LeaderElectionIntegrationTest and ran it
few times but I can't get it to fail.

--
  Sami Siren





Re: LeaderElection bugfix

2012-06-27 Thread Trym R. Møller

Hi Sami

Regarding 2) A simple way to inspect the number of watchers, is to add 
an error log statement to the process method of the watcher

  public void process(WatchedEvent event) {
log.error(seq +  watcher received event:  + event);
and see that the number of logs doubles for each call to
containerMap.get(replicaPort).getZkController().getZkClient().getSolrZooKeeper().pauseCnxn(2000);

Best regards Trym

Den 27-06-2012 10:14, Trym R. Møller skrev:

Hi Sami

Thanks for your rapid reply.

Regarding 1) This seems to be time dependent but it is seen on my 
local windows running the unit test and on a linux server running Solr.
Regarding 2) The test does not show the number of Watchers are 
increasing, but this can be observed either by dumping the memory from 
the jvm or by looking at the debug statements (if debug is enabled).


I don't know how to make assert statements regarding the number of 
watchers in zookeeper, so the test is not quite informative, but more 
confirming that the fix doesn't destroy anything.


Best regards Trym

Den 27-06-2012 10:06, Sami Siren skrev:
On Wed, Jun 27, 2012 at 10:32 AM, Trym R. Møller t...@sigmat.dk 
wrote:

Hi

Hi,


The behaviour of this can be verified using the below test in the
org.apache.solr.cloud.LeaderElectionIntegrationTest

Can you reproduce the failure in your test every time or just rarely?
I added the test method to LeaderElectionIntegrationTest and ran it
few times but I can't get it to fail.

--
  Sami Siren








Re: LeaderElection bugfix

2012-06-27 Thread Sami Siren
ok,

I see what you mean. Looks to me that you're right. I am not too
familiar with the LeaderElector so I'll let Mark take a second look.

--
 Sami Siren

On Wed, Jun 27, 2012 at 11:32 AM, Trym R. Møller t...@sigmat.dk wrote:
 Hi Sami

 Regarding 2) A simple way to inspect the number of watchers, is to add an
 error log statement to the process method of the watcher
  public void process(WatchedEvent event) {
log.error(seq +  watcher received event:  + event);
 and see that the number of logs doubles for each call to
 containerMap.get(replicaPort).getZkController().getZkClient().getSolrZooKeeper().pauseCnxn(2000);

 Best regards Trym

 Den 27-06-2012 10:14, Trym R. Møller skrev:

 Hi Sami

 Thanks for your rapid reply.

 Regarding 1) This seems to be time dependent but it is seen on my local
 windows running the unit test and on a linux server running Solr.
 Regarding 2) The test does not show the number of Watchers are increasing,
 but this can be observed either by dumping the memory from the jvm or by
 looking at the debug statements (if debug is enabled).

 I don't know how to make assert statements regarding the number of
 watchers in zookeeper, so the test is not quite informative, but more
 confirming that the fix doesn't destroy anything.

 Best regards Trym

 Den 27-06-2012 10:06, Sami Siren skrev:

 On Wed, Jun 27, 2012 at 10:32 AM, Trym R. Møller t...@sigmat.dk
 wrote:

 Hi

 Hi,

 The behaviour of this can be verified using the below test in the
 org.apache.solr.cloud.LeaderElectionIntegrationTest

 Can you reproduce the failure in your test every time or just rarely?
 I added the test method to LeaderElectionIntegrationTest and ran it
 few times but I can't get it to fail.

 --
  Sami Siren