One common issue of lost messages is that consumer auto-committing (related config is "auto.commit.enabled", "commit.interval.ms"): from the Kafka consumer point of view, once the messages are returned from the "poll" call they are considered "consumed", and if committing offsets is called it will set the offset to the last message's offset. So if you have some exceptions / errors in the middle of processing some of your consumed records, upon restarting they will not be re-fetched again, hence possibly causing "data loss" on your end.
Guozhang On Mon, Oct 10, 2016 at 3:33 AM, <yangy...@sina.com> wrote: > Hi Guozhang, > At first, thank you answer my question, and give me some suggest. > But, I'm sure I readed some introduction about kafka. > > In my producer, My Code is( c code): > res = rd_kafka_conf_set(kafka_conf, "request.required.acks", "-1", NULL, > 0); > res = rd_kafka_topic_conf_set( topic_conf, "produce.offset.report", > "true", errstr, sizeof(errstr) ); > > In my consumer, My Code is(kafka-python): > > self.__consumer = KafkaConsumer( bootstrap_servers=self.__ > kafka_config["hosts"], > group_id=self.__kafka_config[" > group"], > auto_offset_reset="earliest", > ) > self.__consumer.subscribe([self.__kafka_config["data_ > topic"]]) > > for message in self.__consumer: > > > I think these codes is common, What's your suggest about these codes? > > In the end, I must explain: Lost message is not often, some time, couple > days can find one or two lost messages. > But some day, maybe can find over 20 messages were losted. > Our message over 1,200,000 in one day. > > So, do your have any suggest? > Bye the way, can you speak Chinese? > Thank you very much & Best wishes > Jerry > > > -------------------------------- > > > ----- 原始邮件 ----- > 发件人：Guozhang Wang <wangg...@gmail.com> > 收件人："users@kafka.apache.org" <users@kafka.apache.org>, yangy...@sina.com > 主题：Re: I found kafka lsot message > 日期：2016年10月10日 01点25分 > > Jerry, > > Message lost scenarios usually are due to producer and consumer > mis-configured. Have you read about the client configs web docs? > > http://kafka.apache.org/documentation#producerconfigs > > http://kafka.apache.org/documentation#newconsumerconfigs > > > If not I'd suggest you reading those first and see if you can tune some of > these configs to have better delivery guarantees. > > Guozhang > > > On Fri, Oct 7, 2016 at 9:48 PM, <yangy...@sina.com> wrote: > > Hello every body,I build a kafka cluster(include 5 domains) use > kafka_2.11-0.10.0.0 and kafka-python-1.3.1.I create a topic by 100 > partitions and 2 replicate, then use 20 consumers to receive message. > But, I found sometime the kafka lost message! I found some partition's > offsite lost at consumer.After, I make a global index for every message > from producer for confirm this problem, and I also found the global index > had been break! > Why the kafka lost message? What can I do to fix the problem? > Thanks!Jerry > > -------------------------------- > > > > > -- > -- Guozhang > -- -- Guozhang
