Re: Re: Re: Re: Re: Kafka lost data issue

2015-11-16 Thread Gwen Shapira
If producer doesn't get a response, retries but both produce-requests succeeded, you will get duplicates. Kafka does not have a Idempotent Producer. On Fri, Nov 13, 2015 at 4:35 AM, Prabhjot Bharaj wrote: > Hi Gwen, > > If producer cant get a response but the message got committed, Because of >

Re: Re: Re: Re: Re: Kafka lost data issue

2015-11-13 Thread Prabhjot Bharaj
Hi Gwen, If producer cant get a response but the message got committed, Because of retries, can this producer be committing messages more than once?? I'm trying to see this problem from the point of view of Idempotent Producer Thanks, Prabhjot On Nov 13, 2015 1:26 PM, "Hawin Jiang" wrote: > Tha

Re: Re: Re: Re: Re: Kafka lost data issue

2015-11-12 Thread Hawin Jiang
Thanks Gwen for your excellent slides I will test it again based on your suggestions. Best regards Hawin On Thu, Nov 12, 2015 at 6:35 PM, Gwen Shapira wrote: > Hi, > > First, here's a handy slide-deck on avoiding data loss in Kafka: > > http://www.slideshare.net/gwenshap/kafka-reliability-w

Re: Re: Re: Re: Re: Kafka lost data issue

2015-11-12 Thread Gwen Shapira
Hi, First, here's a handy slide-deck on avoiding data loss in Kafka: http://www.slideshare.net/gwenshap/kafka-reliability-when-it-absolutely-positively-has-to-be-there Note configuration parameters like the number of retries. Also, it looks like you are sending data to Kafka asynchronously, but

Re: Re: Re: Re: Re: Kafka lost data issue

2015-11-12 Thread Hawin Jiang
Hi Pradeep Here is my configuration # Producer Basics # # list of brokers used for bootstrapping knowledge about the rest of the cluster # format: host1:port1,host2:port2 ... metadata.broker.list=localhost:9092 # name of the partitioner c

Re:Re: Re: Re: Re: Kafka lost data issue

2015-11-12 Thread jinxing
i have 3 brokers; the ack configuration is -1(all), meaning a message is sent successfully only after getting every broker's ack; is this a bug? At 2015-11-12 21:08:49, "Pradeep Gollakota" wrote: >What is your producer configuration? Specifically, how many acks are you >requesting from Ka

Re: Re: Re: Re: Kafka lost data issue

2015-11-12 Thread Pradeep Gollakota
What is your producer configuration? Specifically, how many acks are you requesting from Kafka? On Thu, Nov 12, 2015 at 2:03 AM, jinxing wrote: > in kafka_0.8.3.0: > kafkaProducer = new KafkaProducer<>(properties, new ByteArraySerializer(), > new ByteArraySerializer()); > kafkaProducer.flush();

Re:Re: Re: Re: Kafka lost data issue

2015-11-12 Thread jinxing
in kafka_0.8.3.0: kafkaProducer = new KafkaProducer<>(properties, new ByteArraySerializer(), new ByteArraySerializer()); kafkaProducer.flush(); you can call the flush after sending every few messages; At 2015-11-12 17:36:24, "Hawin Jiang" wrote: >Hi Prabhjot > >The messages are "Thread1_ka

Re: Re: Re: Kafka lost data issue

2015-11-12 Thread Hawin Jiang
Hi Prabhjot The messages are "Thread1_kafka_1" and "Thread2_kafka_1". Something like that. For GetOffsetShell report below: [kafka@dn-01 bin]$ ./kafka-run-class.sh kafka.tools.GetOffsetShell --broker-list dn-01:9092 --time -1 --topic kafka-test kafka-test:0:12529261 @Jinxing Can you share

Re:Re: Re: Kafka lost data issue

2015-11-12 Thread jinxing
there is a flush api of the producer, you can call this to prevent messages lost; maybe it can help; At 2015-11-12 16:43:54, "Hawin Jiang" wrote: >Hi Jinxing > >I don't think we can resolve this issue by increasing producers. if I >increased more producers, it should lost more message

Re:Re: Re: Kafka lost data issue

2015-11-12 Thread jinxing
yes, increase producer will trigger more messages lost; that's what I wanted to say; At 2015-11-12 16:43:54, "Hawin Jiang" wrote: >Hi Jinxing > >I don't think we can resolve this issue by increasing producers. if I >increased more producers, it should lost more messages. > >I just test two

Re:Re: Re: Kafka lost data issue

2015-11-12 Thread jinxing
for example the issue can be as below: from the producer side, say 1000 messages is sent, and no exception found from the callback; from the kafka broker side, the log size only increase 900; as a result only 900 messages received from the consumer; conclusion: the problem is we cannot

Re: Re: Kafka lost data issue

2015-11-12 Thread Prabhjot Bharaj
Hi, Just to confirm that the number of messages produced are what you are seeing, What does GetOffsetShell report for this topic ? Regards, Prabhjot On Thu, Nov 12, 2015 at 2:13 PM, Hawin Jiang wrote: > Hi Jinxing > > I don't think we can resolve this issue by increasing producers. if I > in

Re: Re: Kafka lost data issue

2015-11-12 Thread Hawin Jiang
Hi Jinxing I don't think we can resolve this issue by increasing producers. if I increased more producers, it should lost more messages. I just test two producers. Thread Producer 1 has 83064 messages in producer side and 82273 messages in consumer side Thread Producer 2 has 89844 messages in p

Re: Kafka lost data issue

2015-11-12 Thread Hawin Jiang
Hi Jinxing I am using kafka_2.10-0.9.0.0-SNAPSHOT. I have downloaded source code and installed it last week. I saw 97446 messages have been sent to kafka successfully. So far, I have not found any failed messages. Best regards Hawin On Thu, Nov 12, 2015 at 12:07 AM, jinxing wrote: > Hi,

Re: Kafka lost data

2014-10-27 Thread Guozhang Wang
Hello Chen, You can look into brokers for "message size too large" exceptions if you cannot access the producer logs (both of them should have this in their log files). Also which ack mode are your producer using? Guozhang On Mon, Oct 27, 2014 at 10:31 AM, Mark Roberts wrote: > I think it will

Re: Kafka lost data

2014-10-27 Thread Mark Roberts
I think it will depend on how your producer application logs things, but yes I have historically seen exceptions in the producer logs when they exceed the max message size. -Mark On Mon, Oct 27, 2014 at 10:19 AM, Chen Wang wrote: > Hello folks, > I recently noticed our message amount in kafka s