Hello Vladimir, comments in-lined.
On Wed, Jun 4, 2014 at 6:08 AM, Vladimir Tretyakov <
vladimir.tretya...@sematext.com> wrote:
> Hi again, few more questions from me:
>
> *1.*
>
> What I see in JMX:
>
>
> kafka.consumer:type="ZookeeperConsumerConnector",name="af_servers-af_servers-spm_new_cluster_topic-af_servers_wawanawna-Dell-1401353748289-fcaaea29-0-FetchQueueSize"
>
> From code:
>
> newGauge(
> config.clientId + "-" + config.groupId + "-" + topicThreadId._1 +
> "-" + topicThreadId._2 + "-FetchQueueSize",
> new Gauge[Int] {
> def value = q.size
> }
> )
>
> I've tried to parse part as I've understood they.
>
> config.clientId >> af_servers
> topicThreadId._1 >> af_servers-spm_new_cluster_topic
> topicThreadId._2 >> af_servers_wawanawna-Dell-1401353748289-fcaaea29-0
>
> Yes I can suppose that this topicThreadId._1 will always looks like
> GROUP_ID+TOPIC and topicThreadId._2 will contain CONSUMER HOST, but will it
> always true?
With the new consumer coming soon, the metrics naming schemes would very
likely to be refined. So I cannot say this will always be true.
>
> *2.*
>
> From code I see that sometimes Kafka uses "_" as separator, not only "-":
>
> val consumerIdString = {
> var consumerUuid : String = null
> config.consumerId match {
> case Some(consumerId) // for testing only
> => consumerUuid = consumerId
> case None // generate unique consumerId automatically
> => val uuid = UUID.randomUUID()
> consumerUuid = "%s-%d-%s".format(
> InetAddress.getLocalHost.getHostName, System.currentTimeMillis,
> uuid.getMostSignificantBits().toHexString.substring(0,8))
> }
> config.groupId + "_" + consumerUuid
> }
>
> That means if user will use "_" as part of his host/topic/groupId name it
> maybe be a problem to parse string like:
>
>
> kafka.consumer:type="ZookeeperConsumerConnector",name="af_servers-af_servers-spm_new_cluster_topic-af_servers_wawanawna-Dell-1401353748289-fcaaea29-0-FetchQueueSize"
>
> Look at part: "spm_new_cluster_topic-af_servers_wawanawna-Dell", what is
> host name here "servers_wawanawna-Dell" or "wawanawna-Dell" ?
>
> So from one side if we want to be able parse name without any problems we
> have to avoid using "-" and "_" in host/topic/groupId/clientId, but at the
> same time I see (from
> http://grokbase.com/t/kafka/users/133xfsnpdh/cant-use-in-client-name):
>
> *"Client id is used for registering jmx beans for monitoring. Because of
> the*
> *restrictions in bean names, we limit the client id to be only
> alpha-numeric*
> *plus "-" and "_"."*
>
> Does that mean user can use only camelCase in his
> host/topic/groupId/clientId for distinguish one part of name from another?
>
> Is this a problem? Or I didn't understand something?
>
Yeah I agree this is a problem, and we should fix it in the new consumer.
>
> Best regards from Sematext.
>
>
>
>
>
>
> On Tue, Jun 3, 2014 at 3:24 AM, Otis Gospodnetic <
> otis.gospodne...@gmail.com
> > wrote:
>
> > Hi Guozhang,
> >
> > On Mon, Jun 2, 2014 at 7:18 PM, Guozhang Wang <wangg...@gmail.com>
> wrote:
> >
> > > That is indeed a problem, for now, we recommend group name and topic
> > names
> > > to use "_" when there is a need for "-", but this should be fixed
> > > systematically.
> > >
> >
> > Right!
> >
> > For you use case, could you change your topic/group name using "_"?
> >
> >
> > Our own Kafka doesn't use topics with "-" characters, so we don't have a
> > problem.
> >
> > The problem, in our case, is that we have a general (Kafka) monitoring
> tool
> > that other people use to monitor Kafka - see http://sematext.com/spm/ .
> > So
> > we can't really tell people "hey, our tool will work but only if you
> don't
> > have a dash in your topic names and hosts and ... because if you use
> dashes
> > we won't know how to parse your Kafka's MBean names" :)
> >
> >
> > > Also, do you mind to file a JIRA ticket to keep track of this issue?
> >
> >
> > Here it is: https://issues.apache.org/jira/browse/KAFKA-1481
> >
> > Otis
> > --
> > Performance Monitoring * Log Analytics * Search Analytics
> > Solr & Elasticsearch Support * http://sematext.com/
> >
> >
> >
> >
> >
> > >
> > > On Mon, Jun 2, 2014 at 5:18 AM, Vladimir Tretyakov <
> > > vladimir.tretya...@sematext.com> wrote:
> > >
> > > > Hello everyone,
> > > >
> > > > We are adding Kafka 0.8.x monitoring support to SPM
> > > > <http://sematext.com/spm/> here at Sematext. Unfortunately, we
> quickly
> > > hit
> > > > an issue caused by the new bean naming convention that embeds things
> > like
> > > > topic and host names in the beans along with metrics, separated by
> > > dashes,
> > > > making it hard to parse these beans.
> > > >
> > > > To put it simply: it is hard/impossible to automatically figure out
> > which
> > > > part of the bean name is e.g. consumer group, which is the topic,
> which
> > > is
> > > > the host name, and which is the name of the metric.
> > > >
> > > > Let me show you what I mean:
> > > >
> > > > kafka.consumer:type="ConsumerTopicMetrics",
> > > >
> > > > name="af_servers-spm_topic-BytesPerSec"
> > > >
> > > > Here we actually CAN extract:
> > > >
> > > > * consumer group ('af_servers')
> > > >
> > > > * topic ('spm_topic')
> > > >
> > > > * metric (‘BytesPerSec’)
> > > >
> > > > BUT what if the consumer group id and/or topic name contain '-'?
> > > >
> > > > Then how would we extract consumer group and topic?
> > > >
> > > > Here is a concrete example of this problem:
> > > >
> > > > kafka.consumer:type="ConsumerTopicMetrics",
> > > >
> > > > name="af-servers-spm-topic-BytesPerSec"
> > > >
> > > > How can we know what is group id or topic name here?
> > > >
> > > > This looks like a problem to me, but maybe I’m missing something?
> > > >
> > > > Is it possible to have all these values (group id, topic name) as
> > > separate
> > > > attributes inside JMX bean?
> > > >
> > > > Or maybe the problem could be solved if a different delimiter was
> used,
> > > > such as the pipe (“I”)?
> > > >
> > > > It is really needed things and will be nice to have it to build good
> > tool
> > > > for monitoring.
> > > >
> > > > Thx and best regards from Sematext.
> > > >
> > >
> > >
> > >
> > > --
> > > -- Guozhang
> > >
> >
>
--
-- Guozhang
