You are viewing a plain text version of this content. The canonical link for it is here.
Posted to users@kafka.apache.org by Debraj Manna <su...@gmail.com> on 2018/06/18 06:11:52 UTC

Is 0.11 inter broker protocol and log message format backward compatible with 0.10.0 ?

Hi

I have seen the ugrade doc
<https://kafka.apache.org/0110/documentation.html#upgrade> for 0.11.

Let's say I am having a 3 node kafka broker cluster running 0.10 and no
inter.broker.protocol.version and log.message.format.version not added to
the server.properties in any of the brokers.

Can some one let me know if I just update the binaries in one of the broker
in a Kafka cluster 0.11 what is the expected behaviour?  Will the upgraded
broker node join the cluster? Will the cluster work as expected? Is 0.11
inter broker protocol and log message format backward compatible with
0.10.0 ?

Thanks,

Re: Is 0.11 inter broker protocol and log message format backward compatible with 0.10.0 ?

Posted by Ishwor Gurung <me...@ishworgurung.com.INVALID>.
On 18/6/18 4:11 pm, Debraj Manna wrote:
> Hi
> 
> I have seen the ugrade doc
> <https://kafka.apache.org/0110/documentation.html#upgrade> for 0.11.

Please don't *see* it, read it :) The documentation is written with some 
very intricate details so administrators do not mess up during upgrades.

> Let's say I am having a 3 node kafka broker cluster running 0.10 and no
> inter.broker.protocol.version and log.message.format.version not added to
> the server.properties in any of the brokers.
> 
> Can some one let me know if I just update the binaries in one of the broker
> in a Kafka cluster 0.11 what is the expected behaviour?  Will the upgraded

Yes

> broker node join the cluster? Will the cluster work as expected? Is 0.11

Yes

> inter broker protocol and log message format backward compatible with
> 0.10.0 ?

Yes. But with a performance penalty on the Kafka server because each 
message is down converted to 0.10.x format.

A successful upgrade in this case assumes that an upgraded broker had no 
prior issues (i.e., it was shutdown successfully) and that the other two 
brokers had no under replicated partitions, ISR is up to date on them etc..

Also to add, the wire format used in IBP is exactly the same format as 
used by a producer/fetcher. Lastly, the best way to test it out is to do 
these for yourself by spinning up Kafka cluster, fetch/produce on some 
topics and then perform a 0.10.x -> 0.11.x upgrade one by one.

YMMV; but do read up edge cases in the documentation.

[...]

Hope it helps.

-- 
Best regards,
Ishwor Gurung