Hi Stephan,
Sorry for the slow response.

I added some logging inside of my DeserializationSchema’s `deserialize(byte[] 
message)` method.

I see the extra bytes appearing in that method.

If there’s another place I should add logging, please let me know and I’m happy 
to do so.

Additionally (and this is weird), I write all my messages to the DB, so I was 
looking for what messages didn’t make it (i.e. input message 1->10,000 which of 
those isn’t in the DB).  Turns out all 10k are in the DB.  I’m not sure if that 
indicates this message is read and then retried, or what.  I would have guessed 
that somehow extra data got written to my topic, but kafka tool tell me 
otherwise.  So from my application’s perspective it just looks like I get extra 
garbage data every now and then.

This is actually a big relief, I toss out the garbage data and keep rolling.

I hope this helps, thank you.

From: Stephan Ewen <se...@apache.org>
Date: Thursday, March 1, 2018 at 9:26 AM
To: "user@flink.apache.org" <user@flink.apache.org>
Cc: Philip Doctor <philip.doc...@physiq.com>
Subject: Re: Flink Kafka reads too many bytes .... Very rarely

Can you specify exactly where you have that excess of data?

Flink uses basically Kafka's standard consumer and passes byte[] unmodified to 
the DeserializationSchema. Can you help us check whether the "too many bytes" 
happens already before or after the DeserializationSchema?

  - If the "too many bytes" already arrive at the DeserializationSchema, then 
we should dig into the way that Kafka's consumer is configured

  - If the "too many bytes" appears after the  DeserializationSchema, then we 
should look into the DeserializationSchema, for example whether it is stateful, 
accidentally shared across threads, etc.

On Thu, Mar 1, 2018 at 11:08 AM, Fabian Hueske 
<fhue...@gmail.com<mailto:fhue...@gmail.com>> wrote:
Hi Phil,
I've created a JIRA ticket for the problem that you described and linked it to 
this thread: FLINK-8820.
Thank you, Fabian

[1] https://issues.apache.org/jira/browse/FLINK-8820

2018-02-28 5:13 GMT+01:00 Philip Doctor 

  *   The fact that I seem to get all of my data is currently leading me to 
discard and ignore this error

Please ignore this statement, I typed this email as I was testing a theory, I 
meant to delete this line.  This is still a very real issue for me.  I was 
looking to try a work around tomorrow, I saw that the Kafka 11 consumer 
supported transactions for exactly once processing, I was going to read about 
that and see if I could somehow fail a read that I couldn’t deserialize and try 
again, and if that might make a difference (can I just retry this ?).  I’m not 
sure how that’ll go.  If you’ve got an idea for a work around, I’d be all ears 

From: Philip Doctor <philip.doc...@physiq.com<mailto:philip.doc...@physiq.com>>
Date: Tuesday, February 27, 2018 at 10:02 PM
To: "Tzu-Li (Gordon) Tai" <tzuli...@apache.org<mailto:tzuli...@apache.org>>, 
Fabian Hueske <fhue...@gmail.com<mailto:fhue...@gmail.com>>
Cc: "user@flink.apache.org<mailto:user@flink.apache.org>" 
Subject: Re: Flink Kafka reads too many bytes .... Very rarely

Honestly this has been a very frustrating issue to dig in to.  The fact that I 
seem to get all of my data is currently leading me to discard and ignore this 
error, it’s rare, flink still seems to work, but something is very hard to 
debug here and despite some confusing observations, most of my evidence 
suggests that this originates in the flink kafka consumer.

Reply via email to