kafka-users mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Nicolas Berthet <nicolasbert...@maaii.com>
Subject RE: OffsetOutOfRangeException with 0 retention
Date Thu, 14 Mar 2013 02:59:16 GMT
Sadly, I don't have access to those logs anymore, I don't have access to
environment. Though I remember seeing some exception during offset writing,
most probably due to zookeeper connection issue.

What would be side effects of not being able to write the consumer offset,
beside seeing this exception ? As long as my service doesn't restart and I
do not recreate the consumer, would the consumer continue to work ? Get
duplicates of messages when it's getting connected to ZK again ?

Basically, I'm interested in whatever could go wrong with our kafka
consumers, what would be the symptoms, what would be the possible
workaround. 

Kindly,

Nicolas

-----Original Message-----
From: Neha Narkhede [mailto:neha.narkhede@gmail.com] 
Sent: Wednesday, March 13, 2013 9:19
To: users@kafka.apache.org
Subject: Re: OffsetOutOfRangeException with 0 retention

Looks like your consumers have never updated their offsets and are unable to
reset offset to the earliest/latest on startup. Can you pass around the
entire consumer log ?

Thanks,
Neha


On Mon, Mar 11, 2013 at 6:34 PM, Nicolas Berthet
<nicolasberthet@maaii.com>wrote:

> Neha,
>
> Thanks for the reply. I'm using the high level consumer, btw, I'm 
> using kafka 0.7.2 (we built it with scala 2.10) the consumer is using 
> default values with an high ZK timeout value.
>
> As far as I know, my consumers didn't restart, they're running on 
> services that were not restarted (unless the consumer itself would 
> reconnect after sometime).
>
> Don't know if it could be part of the reason, some of my consumers are 
> in remote sites, they have high latency and experience ZK timeouts 
> here and there. I've ZK observers on the remote sites with rather high 
> timeout values, they still disconnect from time to time from the main 
> site due to timeout.
> Due to the ZK timeouts I noticed the consumers fail to write their
offsets.
>
>
> PS: Sorry for the previous spamming, my mail client went crazy and by 
> the time I realized it was too late.
>
> Kindly,
>
> Nicolas
>
> -----Original Message-----
> From: Neha Narkhede [mailto:neha.narkhede@gmail.com]
> Sent: Monday, March 11, 2013 23:52
> To: users@kafka.apache.org
> Subject: Re: OffsetOutOfRangeException with 0 retention
>
> Nicolas,
>
> It seems that you started a consumer from the earliest offset, then 
> shut it down for a long time, and tried restarting it again. At this 
> time, you will see OffsetOutOfRange exceptions, since the offset that 
> your consumer is trying to fetch has been garbage collected from the 
> server (due to it being too old). If you are using the high level 
> consumer (ZookeeperConsumerConnector), the consumer will automatically  
> reset the offset to the earliest or latest depending on the 
> autooffset.reset config value.
>
> Which consumer are you using in this test ?
>
> Thanks,
> Neha
>
>
> On Mon, Mar 11, 2013 at 2:12 AM, Nicolas Berthet
> <nicolasberthet@maaii.com>wrote:
>
> > Hi,
> >
> >
> >
> > I'm currently seeing a lot of OffsetOutOfRangeException in my server 
> > logs (it's not something that appeared recently, I simply didn't use 
> > Kafka before). I tried to find information on the mailing-list, but 
> > nothing seems to match my case.
> >
> >
> >
> > ERROR error when processing request FetchRequest(topic:test-topic,
> > part:0
> > offset:3004960 maxSize:1048576) (kafka.server.KafkaRequestHandlers)
> >
> >      kafka.common.OffsetOutOfRangeException: offset 3004960 is out 
> > of range
> >
> >
> >
> > I understand that, at startup, consumers will ask for a MAX_VALUE 
> > offset to trigger this exception and detect the correct offset, right ?
> >
> >
> >
> > In my case, it's just too often (much more than the number of 
> > consumer connections), but I also noticed it seems to happen in 
> > particular for topics with a "0" retention. Did anybody else suffer 
> > from the same symptoms ?
> >
> >
> >
> > Although it seems not critical (everything seems to work), it's 
> > probably far from optimal, and the log is just full of those.
> >
> >
> >
> > Regards,
> >
> >
> >
> > Nicolas
> >
> >
>
>


Mime
View raw message