kafka-users mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Evan Huus <evan.h...@shopify.com>
Subject Re: MetadataResponse error code handling
Date Wed, 26 Nov 2014 19:29:26 GMT
Thanks, that's exactly what I needed to know.

Evan

On Wed, Nov 26, 2014 at 2:22 PM, Guozhang Wang <wangguoz@gmail.com> wrote:

> For the new producer, the only error code that will result in invalid
> partition metadata is LeaderNotAvailable, and as for the implementation
> details of the KAFKA-1609 patch, it handles this error code by setting the
> leader field to null.
>
> Guozhang
>
> On Wed, Nov 26, 2014 at 11:08 AM, Evan Huus <evan.huus@shopify.com> wrote:
>
> > Hi Guozhang, thanks for the information.
> >
> > I am implementing/maintaining the Golang producer [1] so I do need to
> worry
> > about these details.
> >
> > You mention LeaderNotAvailable as the only one, but the JIRA ticket I
> > linked also talks about ReplicaNotAvailable. Are both of these ignored?
> Or
> > is it that LeaderNotAvailable causes the metadata to be ignored, and
> *all*
> > other partition-specific errors (including ReplicaNotAvailable) still
> carry
> > legitimate data?
> >
> > Thanks,
> > Evan
> >
> > [1] https://github.com/Shopify/sarama
> >
> > On Wed, Nov 26, 2014 at 1:11 PM, Guozhang Wang <wangguoz@gmail.com>
> wrote:
> >
> > > Hi Evan,
> > >
> > > The error code is per-partition, and for new producer only
> > > LeaderNotAvailable error will cause it to ignore the metadata for that
> > > partition. If you are using the producer clients directly you would not
> > > need to worry about these error handling details; if you are fetching
> the
> > > metadata using a metadata request yourself and then you need to handle
> > the
> > > error code accordingly.
> > >
> > > I agree that such documentation is kind of lacking for now, and we
> would
> > > better add that in the response class Java doc.
> > >
> > > Guozhang
> > >
> > >
> > > On Tue, Nov 25, 2014 at 12:29 PM, Evan Huus <evan.huus@shopify.com>
> > wrote:
> > >
> > > > Hi folks,
> > > >
> > > > I was wondering in which cases the PartitionMetadata section of a
> > > > MetadataResponse [1] can contain useful information?
> > > >
> > > > I had been working under the assumption that when the
> > PartitionErrorCode
> > > > was 0 (NoError) then the rest of the data would be legitimate, and
> that
> > > > when the error was non-0 then the rest of the data would be
> > > > blank/empty/etc.
> > > >
> > > > However, I recently discovered a JIRA ticket [2] which suggests
> > > otherwise,
> > > > namely that there are certain error codes for which parts of the
> > > remainder
> > > > of the metadata may still be filled in. I have not found any explicit
> > > > documentation for which error codes this is true. Does anybody know?
> > > >
> > > > Thanks,
> > > > Evan
> > > >
> > > > P.S. I'm not 100% sure if this kind of question should go on users@
> or
> > > > dev@.
> > > > Please let me know if I'm in the wrong place.
> > > >
> > > > [1]
> > > >
> > > >
> > >
> >
> https://cwiki.apache.org/confluence/display/KAFKA/A+Guide+To+The+Kafka+Protocol#AGuideToTheKafkaProtocol-MetadataResponse
> > > > [2] https://issues.apache.org/jira/browse/KAFKA-1609
> > > >
> > >
> > >
> > >
> > > --
> > > -- Guozhang
> > >
> >
>
>
>
> --
> -- Guozhang
>

Mime
  • Unnamed multipart/alternative (inline, None, 0 bytes)
View raw message