You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@lucene.apache.org by "Ramkumar R. Aiyengar" <an...@gmail.com> on 2015/03/14 00:27:40 UTC

Could temporary ZK election / connection loss stall indexing due to LIR?

I haven't tested this, but something which I observed looking at the LIR
code. If there's a ZK election or connection loss, and the leader is unable
to reach a replica, would it stall till the ZK connection is established,
due to the LIR process? I can't see it happening in the background, may be
it should?

Re: Could temporary ZK election / connection loss stall indexing due to LIR?

Posted by "Ramkumar R. Aiyengar" <an...@gmail.com>.
Let me raise an issue for this, ideally all of LIR should happen in the
background..

On Sat, Mar 14, 2015 at 12:10 PM, Shalin Shekhar Mangar <
shalinmangar@gmail.com> wrote:

> Yes, I have observed this during jepsen tests. The current LIR code tries
> ZK operations with retryOnConnectionLoss set to true which makes the
> indexing thread hang during connection loss events.
>
> On Sat, Mar 14, 2015 at 4:57 AM, Ramkumar R. Aiyengar <
> andyetitmoves@gmail.com> wrote:
>
>> I haven't tested this, but something which I observed looking at the LIR
>> code. If there's a ZK election or connection loss, and the leader is unable
>> to reach a replica, would it stall till the ZK connection is established,
>> due to the LIR process? I can't see it happening in the background, may be
>> it should?
>>
>
>
>
> --
> Regards,
> Shalin Shekhar Mangar.
>



-- 
Not sent from my iPhone or my Blackberry or anyone else's

Re: Could temporary ZK election / connection loss stall indexing due to LIR?

Posted by Shalin Shekhar Mangar <sh...@gmail.com>.
Yes, I have observed this during jepsen tests. The current LIR code tries
ZK operations with retryOnConnectionLoss set to true which makes the
indexing thread hang during connection loss events.

On Sat, Mar 14, 2015 at 4:57 AM, Ramkumar R. Aiyengar <
andyetitmoves@gmail.com> wrote:

> I haven't tested this, but something which I observed looking at the LIR
> code. If there's a ZK election or connection loss, and the leader is unable
> to reach a replica, would it stall till the ZK connection is established,
> due to the LIR process? I can't see it happening in the background, may be
> it should?
>



-- 
Regards,
Shalin Shekhar Mangar.