Thanks for the responses. Based on the document https://storm.incubator.apache.org/documentation/Running-topologies-on-a-production-cluster.html, I am able to set the TOPOLOGY_MAX_SPOUT_PENDING, but how can you set the TOPOLOGY_AKERS? I see TOPOLOGY_AKERS_EXECUTORS. Does that mean the same? How do you determine the TOPOLOGY_AKERS for a batch load of more than 100K data at 20 messages per seconds? Can any of you give me an idea on this?
-- Kushan Maskey 817.403.7500 On Wed, Sep 10, 2014 at 4:18 AM, Spico Florin <spicoflo...@gmail.com> wrote: > Hello! > I'll consider to slow down the spout. Set up a value > for Config.TOPOLOGY_MAX_SPOUT_PENDING. It can happen that Cassanda, Solr > and CouchDatabase do not cope with the requency that you emit your messages > and thus you have backpressure. > Also, the spout should emit the messages anchored (( not sure here see > my post about this))in order that the set up of > Config.TOPOLOGY_MAX_SPOUT_PENDING to take effect. > Check this blog: > http://brianoneill.blogspot.ro/2012/08/a-big-data-trifecta-storm-kafka-and.html > > Hope that these help. > Best regards, > Florin > > > On Wed, Sep 10, 2014 at 6:25 AM, Vikas Agarwal <vi...@infoobjects.com> > wrote: > >> Kafka still contains the logs and they would be there upto the configured >> time of log retention period. Check server.properties of kafka and update >> the log retention period to 5 min and restart kafka and when kafka >> stablizes, shut down it and restart the it with original value of log >> retentions period property. >> >> >> On Tue, Sep 9, 2014 at 10:40 PM, Kushan Maskey < >> kushan.mas...@mmillerassociates.com> wrote: >> >>> I hope it did because I dont see the multiple tuple failure error. But I >>> see another issue. >>> I have stopped loading the batch process that sends messages to Kafka. >>> I killed my topology and then restarted again. I still see that message are >>> been loaded into Cassandra. Does that mean that storm still trying to >>> process the failed messages? Is htere a way to flush the old message out >>> from storm so I can fresh start it? >>> >>> -- >>> Kushan Maskey >>> 817.403.7500 >>> >>> On Tue, Sep 9, 2014 at 10:09 AM, Naresh Kosgi <nareshko...@gmail.com> >>> wrote: >>> >>>> Yes, that is what I was talking about. Hopefully that fixes it. >>>> >>>> On Tue, Sep 9, 2014 at 10:59 AM, Kushan Maskey < >>>> kushan.mas...@mmillerassociates.com> wrote: >>>> >>>>> Just realized that the tuple timeout you are talking about is the >>>>> "topology.message.timeout.secs" >>>>> which was set to 30 sec and now I made to to 120. >>>>> >>>>> -- >>>>> Kushan Maskey >>>>> 817.403.7500 >>>>> >>>>> On Tue, Sep 9, 2014 at 9:43 AM, Kushan Maskey < >>>>> kushan.mas...@mmillerassociates.com> wrote: >>>>> >>>>>> >>>>>> Thanks and apologies, I should I mentioned that in my question >>>>>> earlier. I am using storm 0.9.2 and using the inbuilt KafkaSpout. I do >>>>>> not >>>>>> implement any failure my self. Do I need to create my own custom >>>>>> KafkaSpout? >>>>>> >>>>>> I have not set timeout for tuples. In fact I dont know where to set >>>>>> that. Here is my storm config if that is where I need to set the time >>>>>> out. >>>>>> But non of them say anything about tuple timeout. >>>>>> >>>>>> dev.zookeeper.path/tmp/dev-storm-zookeeperdrpc.childopts-Xmx768m >>>>>> drpc.invocations.port3773drpc.port3772drpc.queue.size128 >>>>>> drpc.request.timeout.secs600drpc.worker.threads64java.library.path >>>>>> /usr/local/lib:/opt/local/lib:/usr/liblogviewer.appender.nameA1 >>>>>> logviewer.childopts-Xmx128mlogviewer.port8000nimbus.childopts >>>>>> -Xmx1024mnimbus.cleanup.inbox.freq.secs600 >>>>>> nimbus.file.copy.expiration.secs600nimbus.hostnmcxstrmd001 >>>>>> nimbus.inbox.jar.expiration.secs3600nimbus.monitor.freq.secs10 >>>>>> nimbus.reassigntruenimbus.supervisor.timeout.secs60 >>>>>> nimbus.task.launch.secs120nimbus.task.timeout.secs30 >>>>>> nimbus.thrift.max_buffer_size1048576nimbus.thrift.port6627 >>>>>> nimbus.topology.validator >>>>>> backtype.storm.nimbus.DefaultTopologyValidatorstorm.cluster.mode >>>>>> distributedstorm.local.dir/data/disk00/storm/localdir >>>>>> storm.local.mode.zmqfalsestorm.messaging.netty.buffer_size5242880 >>>>>> storm.messaging.netty.client_worker_threads1 >>>>>> storm.messaging.netty.flush.check.interval.ms10 >>>>>> storm.messaging.netty.max_retries30storm.messaging.netty.max_wait_ms >>>>>> 1000storm.messaging.netty.min_wait_ms100 >>>>>> storm.messaging.netty.server_worker_threads1 >>>>>> storm.messaging.netty.transfer.batch.size262144 >>>>>> storm.messaging.transportbacktype.storm.messaging.netty.Context >>>>>> storm.thrift.transport >>>>>> backtype.storm.security.auth.SimpleTransportPlugin >>>>>> storm.zookeeper.connection.timeout15000storm.zookeeper.port2181 >>>>>> storm.zookeeper.retry.interval1000 >>>>>> storm.zookeeper.retry.intervalceiling.millis30000 >>>>>> storm.zookeeper.retry.times5storm.zookeeper.root/storm >>>>>> storm.zookeeper.serversnmcxstrmd001storm.zookeeper.session.timeout >>>>>> 20000supervisor.childopts-Xmx256msupervisor.enabletrue >>>>>> supervisor.heartbeat.frequency.secs5supervisor.monitor.frequency.secs >>>>>> 3supervisor.slots.ports >>>>>> 6700,6701,6702,6703,6704,6705,6706,6707,6708,6709,6710,6711,6712,6713,6714,6715,6716,6717,6718,6719,6720,6721,6722,6723,6724,6725,6726,6727,6728 >>>>>> supervisor.worker.start.timeout.secs120supervisor.worker.timeout.secs >>>>>> 30task.heartbeat.frequency.secs3task.refresh.poll.secs10 >>>>>> topology.acker.executorstopology.builtin.metrics.bucket.size.secs60 >>>>>> topology.debugfalsetopology.disruptor.wait.strategy >>>>>> com.lmax.disruptor.BlockingWaitStrategy >>>>>> topology.enable.message.timeoutstrue >>>>>> topology.error.throttle.interval.secs10 >>>>>> topology.executor.receive.buffer.size1024 >>>>>> topology.executor.send.buffer.size1024 >>>>>> topology.fall.back.on.java.serializationtruetopology.kryo.factory >>>>>> backtype.storm.serialization.DefaultKryoFactory >>>>>> topology.max.error.report.per.interval5topology.max.spout.pending >>>>>> topology.max.task.parallelismtopology.message.timeout.secs30 >>>>>> topology.multilang.serializerbacktype.storm.multilang.JsonSerializer >>>>>> topology.receiver.buffer.size8 >>>>>> topology.skip.missing.kryo.registrationsfalse >>>>>> topology.sleep.spout.wait.strategy.time.ms1 >>>>>> topology.spout.wait.strategy >>>>>> backtype.storm.spout.SleepSpoutWaitStrategy >>>>>> topology.state.synchronization.timeout.secs60 >>>>>> topology.stats.sample.rate0.05topology.tasks >>>>>> topology.tick.tuple.freq.secstopology.transfer.buffer.size1024 >>>>>> topology.trident.batch.emit.interval.millis500 >>>>>> topology.tuple.serializer >>>>>> backtype.storm.serialization.types.ListDelegateSerializer >>>>>> topology.worker.childoptstopology.worker.receiver.thread.count1 >>>>>> topology.worker.shared.thread.pool.size4topology.workers1 >>>>>> transactional.zookeeper.porttransactional.zookeeper.root >>>>>> /transactionaltransactional.zookeeper.serversui.childopts-Xmx768m >>>>>> ui.port8080worker.childopts-Xmx768mworker.heartbeat.frequency.secs1 >>>>>> zmq.hwm0zmq.linger.millis5000zmq.threads1 >>>>>> >>>>>> -- >>>>>> Kushan Maskey >>>>>> 817.403.7500 >>>>>> >>>>>> On Tue, Sep 9, 2014 at 9:23 AM, Naresh Kosgi <nareshko...@gmail.com> >>>>>> wrote: >>>>>> >>>>>>> What is your timeout setting for failing a tuple? Its hard to say >>>>>>> what is causing this issue without more information but the default >>>>>>> timeout >>>>>>> on tuples is 30 seconds and for some tuples it maybe taking longer then >>>>>>> 30 >>>>>>> seconds to process. Try increasing the timeout to 1 or 2 min? >>>>>>> >>>>>>> >>>>>>> "Why the ack/failure ack counts are so much higher than the number >>>>>>> of records I am trying to process?" >>>>>>> >>>>>>> how are you implementing the fail() method in your spout? on >>>>>>> failure, this method is called by the framework. It could be you are >>>>>>> reemitting the tuple to be processed and its failing again. This could >>>>>>> be >>>>>>> a reason why u have more failed tuples then records >>>>>>> >>>>>>> On Tue, Sep 9, 2014 at 10:06 AM, Kushan Maskey < >>>>>>> kushan.mas...@mmillerassociates.com> wrote: >>>>>>> >>>>>>>> I have a batch job where I process more than 100k records from >>>>>>>> file. I post all these message to Kafka topic. I have a topology that >>>>>>>> goes >>>>>>>> and fetches these records and dumps them into Cassandra database and >>>>>>>> also >>>>>>>> updates solr and couch databases. >>>>>>>> >>>>>>>> I have been trying to run the process multiple times to make sure >>>>>>>> that the process completes successfully. It does run successfully >>>>>>>> sometimes >>>>>>>> and sometimes it errors out saying the following error that says "Too >>>>>>>> many >>>>>>>> tuple failures" in the storm UI. >>>>>>>> >>>>>>>> java.lang.RuntimeException: java.lang.RuntimeException: Too many >>>>>>>> tuple failures at >>>>>>>> backtype.storm.utils.DisruptorQueue.consumeBatchToCursor(DisruptorQueue.java:128) >>>>>>>> at >>>>>>>> backtype.storm.utils.DisruptorQueue.consumeBatch(DisruptorQueue.java:87) >>>>>>>> at backtype.storm.disruptor$consume_batch.invoke(disruptor.clj:76) at >>>>>>>> backtype.storm.daemon.executor$fn__5573$fn__5588$fn__5617.invoke(executor.clj:540) >>>>>>>> at backtype.storm.util$async_loop$fn__457.invoke(util.clj:431) at >>>>>>>> clojure.lang.AFn.run(AFn.java:24) at >>>>>>>> java.lang.Thread.run(Thread.java:744) >>>>>>>> Caused by: java.lang.RuntimeException: Too many tuple failures at >>>>>>>> storm.kafka.PartitionManager.fail(PartitionManager.java:210) at >>>>>>>> storm.kafka.KafkaSpout.fail(KafkaSpout.java:174) at >>>>>>>> backtype.storm.daemon.executor$fail_spout_msg.invoke(executor.clj:370) >>>>>>>> at >>>>>>>> backtype.storm.daemon.executor$fn$reify__5576.expire(executor.clj:430) >>>>>>>> at >>>>>>>> backtype.storm.utils.RotatingMap.rotate(RotatingMap.java:73) at >>>>>>>> backtype.storm.daemon.executor$fn__5573$tuple_action_fn__5579.invoke(executor.clj:435) >>>>>>>> at >>>>>>>> backtype.storm.daemon.executor$mk_task_receiver$fn__5564.invoke(executor.clj:402) >>>>>>>> at >>>>>>>> backtype.storm.disruptor$clojure_handler$reify__745.onEvent(disruptor.clj:58) >>>>>>>> at >>>>>>>> backtype.storm.utils.DisruptorQueue.consumeBatchToCursor(DisruptorQueue.java:125) >>>>>>>> ... 6 more >>>>>>>> >>>>>>>> once this failure happens, i also see that the number of records >>>>>>>> stored in Cassandra database if way much higher than the actual batch >>>>>>>> records count. How do I handle this error? Also when there is any kind >>>>>>>> of >>>>>>>> error/exception occurs then the ack failed values goes up form 0 to >>>>>>>> thousands. Why the ack/failure ack counts are so much higher thank the >>>>>>>> number of records I am trying to process? >>>>>>>> >>>>>>>> >>>>>>>> -- >>>>>>>> Kushan Maskey >>>>>>>> >>>>>>> >>>>>>> >>>>>> >>>>> >>>> >>> >> >> >> -- >> Regards, >> Vikas Agarwal >> 91 – 9928301411 >> >> InfoObjects, Inc. >> Execution Matters >> http://www.infoobjects.com >> 2041 Mission College Boulevard, #280 >> Santa Clara, CA 95054 >> +1 (408) 988-2000 Work >> +1 (408) 716-2726 Fax >> >> >
