Home | About | Sematext search-lucene.com search-hadoop.com
 Search Hadoop and all its subprojects:

Switch to Plain View
Zookeeper, mail # user - stunnel and zookeeper and timeouts


+
Prakash 2012-12-14, 19:56
+
Patrick Hunt 2012-12-17, 08:25
Copy link to this message
-
Re: stunnel and zookeeper and timeouts
Matt Wise 2012-12-22, 05:56
I am actually curious about how you setup the stunnel-to-stunnel connections between the servers. Do you literally have individual stunnel connections on every single host to every other host ? ie..

hostA:2000 -> hostB:3888
hostA:2001 -> hostC:3888
hostB:2000 -> hostA:3888
hostB:2001 -> hostC:3888
hostC:2000 -> hostA:3888
hostC:2001 -> hostB:3888

I'm mainly curious how you run this operationally ... it seems rather complicated. In our case we use stunnel between our Zookeeper clients and the Servers -- but since the Zookeeper quorum is all in one Amazon Region, we just leverage security groups for the inter-server-communication.

--Matt

(ps, stunnel+zookeeper does work amazingly well on the client-side of things. we've been extremely happy with that model)

On Dec 17, 2012, at 12:25 AM, Patrick Hunt <[EMAIL PROTECTED]> wrote:

> On Fri, Dec 14, 2012 at 11:56 AM, Prakash <[EMAIL PROTECTED]> wrote:
>> I am running a 3 cluster setup where the zookeeper connections (both client
>> - server and server - server ) run through stunnel. Stunnel has an idle
>> timeout setting, where if it sees a connection that is idle (default 12
>> hours) it tries to close it. I see that for the server to server, there are
>> two ports/connections configured in zookeeper. One for followers to be
>> connected to the leader and a separate one for leader election.
>>
>> My question is, after the cluster is bought up, and leader election is
>> done, is there any activity (keep-alive etc) on the leader election
>> connections, if things are operating normally?
>>
>
> iirc the answer is no. Can anyone confirm?
>
> Patrick
>
>> If not, i am worried stunnel will try to close this port as there is no
>> activity.