Kafka, mail # user - Re: Copy kafka data between servers? - 2013-03-01, 01:11
Solr & Elasticsearch trainings in New York & San Francisco [more info][hide]
 Search Hadoop and all its subprojects:

Switch to Plain View
Jason Huang 2013-02-27, 10:11
Jun Rao 2013-02-27, 17:48
Neha Narkhede 2013-02-27, 17:58
Jason Huang 2013-02-27, 21:31
Neha Narkhede 2013-02-27, 23:59
Jason Huang 2013-02-28, 11:21
Neha Narkhede 2013-02-28, 17:10
Copy link to this message
Re: Copy kafka data between servers?

I actually tried to load the data back with the same instance of kafka
on server A so the broker id must be the same. The reason I brought
this up at the first place is because we've had some issues
recognizing the messages on a server stop/restart. I was able to
reproduce our issue with following steps:

(1) servers start:
nohup sudo /opt/kafka-0.8.0/bin/zookeeper-server-start.sh
/opt/kafka-0.8.0/config/zookeeper.properties >
/opt/kafka-0.8.0/data/kafka-logs/zook.out 2>&1 &
nohup sudo /opt/kafka-0.8.0/kafka-server-start.sh
/opt/kafka-0.8.0/config/server.properties >
/opt/kafka-0.8.0/data/kafka-logs/kafka.out 2>&1 &

(2) create some messages

(3) stop server
sudo /opt/kafka-0.8.0/bin/kafka-server-stop.sh
sudo /opt/kafka-0.8.0/bin/zookeeper-server-start.sh

Notice that kafka-server-stop.sh uses kill -SIGTERM and
zookeeper-server-start.sh uses kill -SIGINT. My observation is that on
our server kill -SIGINT doesn't actually kill the zookeeper process.
(I can still see that running when I check the processes).

Start from this state (running kill -SIGTERM for kafka server and kill
-SIGINT for zookeeper server), we restart the zookeeper and kafka
nohup sudo /opt/kafka-0.8.0/bin/zookeeper-server-start.sh
/opt/kafka-0.8.0/config/zookeeper.properties >
/opt/kafka-0.8.0/data/kafka-logs/zook.out 2>&1 &
nohup sudo /opt/kafka-0.8.0/kafka-server-start.sh
/opt/kafka-0.8.0/config/server.properties >
/opt/kafka-0.8.0/data/kafka-logs/kafka.out 2>&1 &

Then when we tried to fetch the messages from existing topics and
partitions, we get the following error:
WARN [KafkaApi-1] Error while responding to offset request
kafka.common.UnknownTopicOrPartitionException: Topic topic_general
partition 0 doesn't exist on 1
at kafka.server.ReplicaManager.getLeaderReplicaIfLocal(ReplicaManager.scala:163)

I am not sure if anyone has experienced this before. It appears to me
that because kill -SIGINT didn't actually kill the previous zookeeper
process, running from that state messes up the partition/topic
information with zookeeper? And maybe because of that, copying the log
files and trying to reload them won't work (because somehow
information were corrupted)?



On Thu, Feb 28, 2013 at 12:10 PM, Neha Narkhede <[EMAIL PROTECTED]> wrote:

Neha Narkhede 2013-03-01, 22:41
Jason Huang 2013-03-03, 11:44
NEW: Monitor These Apps!
elasticsearch, apache solr, apache hbase, hadoop, redis, casssandra, amazon cloudwatch, mysql, memcached, apache kafka, apache zookeeper, apache storm, ubuntu, centOS, red hat, debian, puppet labs, java, senseiDB