replica.lag.time.max.ms or doco bugged?

[Sven Stark <sv...@m-square.com.au>]

Howdy!

I've got a three node 0.8 cluster with replication factor three and
replica.lag.time.max.ms set to 2000.

Typical list-topic output looks like

topic: user.publish partition: 0 leader: 1 replicas: 2,1,3 isr: 1,2,3

I noticed though that when I kill (kill -9 PID) one of the brokers the list
of the ISR does not change. The doco states for replica.lag.time.max.ms  "If
a follower hasn't sent any fetch requests for this window of time, the
leader will remove the follower from ISR and treat it as dead." According
to this I would expect that after 2000ms the killed broker would be removed
from the ISR list. Which doesn't happen. Not even after minutes.

Is this

a/ a kafka bug?
b/ a doco bug?
c/ a bug in my understanding of how it works?

Thanks in advance,
Sven

Re: replica.lag.time.max.ms or doco bugged?

[Guozhang Wang <wa...@gmail.com>]

Hello Sven,

After you killed the broker, do you keep sending data to Kafka? If not then
the ISR may not change in 0.8

Guozhang


On Tue, Jan 21, 2014 at 4:44 PM, Sven Stark <sv...@m-square.com.au>wrote:

> Howdy!
>
> I've got a three node 0.8 cluster with replication factor three and
> replica.lag.time.max.ms set to 2000.
>
> Typical list-topic output looks like
>
> topic: user.publish partition: 0 leader: 1 replicas: 2,1,3 isr: 1,2,3
>
> I noticed though that when I kill (kill -9 PID) one of the brokers the list
> of the ISR does not change. The doco states for replica.lag.time.max.ms "If
> a follower hasn't sent any fetch requests for this window of time, the
> leader will remove the follower from ISR and treat it as dead." According
> to this I would expect that after 2000ms the killed broker would be removed
> from the ISR list. Which doesn't happen. Not even after minutes.
>
> Is this
>
> a/ a kafka bug?
> b/ a doco bug?
> c/ a bug in my understanding of how it works?
>
> Thanks in advance,
> Sven
>



-- 
-- Guozhang