Hi,

A few items our operations teams monitor today in our
JMS infrastructure are 'number of messages processed' and 'number of
messages in queue'.

Since Kafka changes the paradigm away from 'messages in queue' how do we
get operations an idea if our consumers are running behind?

One thought is to query the zookeeper storage and get the current offset
for each topic/partition/consumer group and compare it to the latest offset
created by the broker. That will tell us if one or more consumer groups and
specific partition consumers are not keeping up.

Only problem, I can't figure out where the broker is storing the high water
mark. I looked at the 0.8 Zookeeper document on the wiki and didn't see it
there (which is where I see the consumer group information).

First question, is the watermark stored in zookeeper?

Second, is there a different way of monitoring how consumers are doing
relative to what has been stored by the broker?

Thanks,

Chris

 

All zk paths in 0.8 are documented in
https://cwiki.apache.org/confluence/display/KAFKA/Kafka+data+structures+in+Zookeeper

In 0.8, there is a jmx bean (ConsumerLag) in the consumer under
kafka.server that monitors the lag of each partition in terms of messages.
We also have a command line tool ConsumerOffsetChecker.

Thanks,

Jun

On Mon, Feb 18, 2013 at 7:54 AM, Chris Curtin <[EMAIL PROTECTED]>wrote:

> Hi,
>
> A few items our operations teams monitor today in our
> JMS infrastructure are 'number of messages processed' and 'number of
> messages in queue'.
>
> Since Kafka changes the paradigm away from 'messages in queue' how do we
> get operations an idea if our consumers are running behind?
>
> One thought is to query the zookeeper storage and get the current offset
> for each topic/partition/consumer group and compare it to the latest offset
> created by the broker. That will tell us if one or more consumer groups and
> specific partition consumers are not keeping up.
>
> Only problem, I can't figure out where the broker is storing the high water
> mark. I looked at the 0.8 Zookeeper document on the wiki and didn't see it
> there (which is where I see the consumer group information).
>
> First question, is the watermark stored in zookeeper?
>
> Second, is there a different way of monitoring how consumers are doing
> relative to what has been stored by the broker?
>
> Thanks,
>
> Chris
>

 

Thanks Jun.

Looks like a couple of ways to do this based on how operations wants to
manage things.

Thanks,

Chris
On Mon, Feb 18, 2013 at 12:04 PM, Jun Rao <[EMAIL PROTECTED]> wrote:

> All zk paths in 0.8 are documented in
>
> https://cwiki.apache.org/confluence/display/KAFKA/Kafka+data+structures+in+Zookeeper
>
> In 0.8, there is a jmx bean (ConsumerLag) in the consumer under
> kafka.server that monitors the lag of each partition in terms of messages.
> We also have a command line tool ConsumerOffsetChecker.
>
> Thanks,
>
> Jun
>
> On Mon, Feb 18, 2013 at 7:54 AM, Chris Curtin <[EMAIL PROTECTED]
> >wrote:
>
> > Hi,
> >
> > A few items our operations teams monitor today in our
> > JMS infrastructure are 'number of messages processed' and 'number of
> > messages in queue'.
> >
> > Since Kafka changes the paradigm away from 'messages in queue' how do we
> > get operations an idea if our consumers are running behind?
> >
> > One thought is to query the zookeeper storage and get the current offset
> > for each topic/partition/consumer group and compare it to the latest
> offset
> > created by the broker. That will tell us if one or more consumer groups
> and
> > specific partition consumers are not keeping up.
> >
> > Only problem, I can't figure out where the broker is storing the high
> water
> > mark. I looked at the 0.8 Zookeeper document on the wiki and didn't see
> it
> > there (which is where I see the consumer group information).
> >
> > First question, is the watermark stored in zookeeper?
> >
> > Second, is there a different way of monitoring how consumers are doing
> > relative to what has been stored by the broker?
> >
> > Thanks,
> >
> > Chris
> >
>