Actually, reassigning the replica does work, even if the broker the
partition resides on is dead. My problem was that there were some unknown
issue with the leader. When I restarted the leader broker, it worked.

Paul

On 10/6/14, 11:41 AM, "Joe Stein" <[email protected]> wrote:

>Agreed, I think it is also a replacement scenario since we may want to
>bring in a new broker for the dead broker.  We should support explicit
>remove also. We could add to the ReassignPartitionsCommand for an explicit
>"make new broker just like the old broker".  This is important very much
>in
>cloud environments where the broker id is the ip address and instances
>come
>and go.  Many scripts get written to do a bunch of steps when it could
>just
>be one.  This makes automatically "in-servicing" a new broker and having
>it
>take up the balance of the work for the broker that it is replacing much
>simpler and straightforward. It also makes it more consistent across the
>community. I created a JIRA
>https://issues.apache.org/jira/browse/KAFKA-1678
>for those just now.
>
>On Mon, Oct 6, 2014 at 2:10 PM, Gwen Shapira <[email protected]>
>wrote:
>
>> Do we have a jira to support removal of dead brokers without having to
>> start a new broker with the same id?
>>
>> I think its something we'll want to allow.
>>
>> On Thu, Oct 2, 2014 at 7:45 AM, Jun Rao <[email protected]> wrote:
>> > The reassign partition process only completes after the new replicas
>>are
>> > fully caught up and the old replicas are deleted. So, if the old
>>replica
>> is
>> > down, the process can never complete, which is what you observed. In
>>your
>> > case, if you just want to replace a broker host with a new one,
>>instead
>> of
>> > using the reassign partition tool, simply start a new broker with the
>> same
>> > broker id as the old one, the new broker will replicate all the data
>> > automatically.
>> >
>> > Thanks,
>> >
>> > Jun
>> >
>> > On Wed, Oct 1, 2014 at 3:43 PM, Lung, Paul <[email protected]> wrote:
>> >
>> >> Hi All,
>> >>
>> >> I had a 0.8.1.1 Kafka Broker go down, and I was trying to use the
>> reassign
>> >> partition script to move topics off that broker. When I describe the
>> >> topics, I see the following:
>> >>
>> >> Topic: mini__022____active_120__33__mini Partition: 0 Leader: 2131118
>> >> Replicas: 2131118,2166601,2163421 Isr: 2131118,2166601
>> >>
>> >> This shows that the broker ³2163421² is down. So I create the
>>following
>> >> file /tmp/move_topic.json:
>> >> {
>> >>     "version": 1,
>> >>     "partitions": [
>> >>         {
>> >>             "topic": "mini__022____active_120__33__mini",
>> >>             "partition": 0,
>> >>             "replicas": [
>> >>                 2131118, 2166601,  2156998
>> >>             ]
>> >>         }
>> >>     ]
>> >> }
>> >>
>> >> And then do this:
>> >>
>> >> ./kafka-reassign-partitions.sh --execute --reassignment-json-file
>> >> /tmp/move_topic.json
>> >> Successfully started reassignment of partitions
>> >>
>> 
>>{"version":1,"partitions":[{"topic":"mini__022____active_120__33__mini","
>>partition":0,"replicas":[2131118,2166601,2156998]}]}
>> >>
>> >> However, when I try to verify this, I get the following error:
>> >> ./kafka-reassign-partitions.sh --verify --reassignment-json-file
>> >> /tmp/move_topic.json
>> >> Status of partition reassignment:
>> >> ERROR: Assigned replicas (2131118,2166601,2156998,2163421) don't
>>match
>> the
>> >> list of replicas for reassignment (2131118,2166601,2156998) for
>> partition
>> >> [mini__022____active_120__33__mini,0]
>> >> Reassignment of partition [mini__022____active_120__33__mini,0]
>>failed
>> >>
>> >> If I describe the topics, I now see there are 4 replicas. This has
>>been
>> >> like this for many hours now, so it seems to have permanently moved
>>to 4
>> >> replicas for some reason.
>> >> Topic:mini__022____active_120__33__mini PartitionCount:1
>> >> ReplicationFactor:4 Configs:
>> >> Topic: mini__022____active_120__33__mini Partition: 0 Leader: 2131118
>> >> Replicas: 2131118,2166601,2156998,2163421 Isr: 2131118,2166601
>> >>
>> >> If I re-execute and re-verify, I get the same error. So it seems to
>>be
>> >> wedged.
>> >>
>> >> Can someone help?
>> >>
>> >> Paul Lung
>> >>
>> >>
>> >>
>>

Reply via email to