Hi Tim,

Thanks for your inputs. I am working as per your suggestions and will
update to you the findings.

Regards,
Valencia



From:   Tim Armstrong <[email protected]>
To:     Valencia Serrao/Austin/Contr/IBM@IBMUS
Cc:     Jeszy <[email protected]>, "dev@impala"
            <[email protected]>, Manish
            Patil/Austin/Contr/IBM@IBMUS, Nishidha
            Panpaliya/Austin/Contr/IBM@IBMUS, Sudarshan
            Jagadale/Austin/Contr/IBM@IBMUS
Date:   01/20/2017 03:16 AM
Subject:        Re: Fw: impalad crashes at load_nested step in dataloading.



I had a look through our bug tracker and I don't see any known crashes in
that part of the code. That code has also been rewritten pretty extensively
so the function that it crashed in no longer exists.

It's a null pointer dereference so your next step should really be to
inspect the core dump and figure out which pointer is NULL.

I'd also recommend reproducing it on a DEBUG build so that the core file is
easier to debug. It may also hit a DCHECK somewhere earlier that will give
you a better diagnostic of what the problem is.

- Tim

On Thu, Jan 19, 2017 at 6:41 AM, Valencia Serrao <[email protected]>
wrote:
  Hi Jeszy,

  Thanks for your inputs. I've got the backtrace generated. Here it is:
  (See attached file: bt.txt)
  I am also analyzing it.

  It would be great if I could get some pointers.

  Regards,
  Valencia

  Inactive hide details for Nishidha Panpaliya---01/19/2017 04:35:50
  PM-------- Forwarded by Nishidha Panpaliya/Austin/Contr/IBM Nishidha
  Panpaliya---01/19/2017 04:35:50 PM-------- Forwarded by Nishidha
  Panpaliya/Austin/Contr/IBM on 01/19/2017 04:35 PM ----- From: Jeszy <jes

  From: Nishidha Panpaliya/Austin/Contr/IBM
  To: Valencia Serrao/Austin/Contr/IBM@IBMUS
  Date: 01/19/2017 04:35 PM
  Subject: Fw: impalad crashes at load_nested step in dataloading.



  ----- Forwarded by Nishidha Panpaliya/Austin/Contr/IBM on 01/19/2017
  04:35 PM -----

  From: Jeszy <[email protected]>
  To: [email protected]
  Cc: Tim Armstrong <[email protected]>, Manish
  Patil/Austin/Contr/IBM@IBMUS, Nishidha Panpaliya/Austin/Contr/IBM@IBMUS,
  Sudarshan Jagadale/Austin/Contr/IBM@IBMUS
  Date: 01/19/2017 04:33 PM

  Subject: Re: impalad crashes at load_nested step in dataloading.



  Hello Valencia,

  Yes, this is a crash, but it's still not likely to be related to the
  "Could not codegen" message.
  Looks like a core dump was written:
  # Core dump written. Default location: /home/test/ImpalaPPC/core or
  core.3848
  #
  # An error report file with more information is saved as:
  # /home/test/ImpalaPPC/hs_err_pid3848.log
  The best would be to get a backtrace from that. This should do:
  gdb <path/to/impalad> <path/to/core> --batch -ex 'thread apply all bt' >
  bt

  We can then check if it matches any known issues.

  Thanks


  On Thu, Jan 19, 2017 at 11:48 AM, Valencia Serrao <[email protected]>
  wrote:
        Hi Tim,

        Thanks for the quick response.

        I've run the query (See attached file: tmp_orders_string_query.txt)
        in impala shell and I get the "Connection closed by peer" message.
        Next, if I start the impala-cluster again and check in impala-shell
        prompt the table is created in tpch_nested_parquet database.
        Checking the impalad logs for the failures, I again got the message
        I mentioned in earlier email..
        I'm attaching the logs again for you reference: (See attached file:
        load_nested.py.log), (See attached file: hs_err_pid3848.log)(See
        attached file: impalad.testvm.test.log.INFO
        .20170118-211515.3848.txt)

        Your comments will help us understand the issue even better.

        Regards,
        Valencia

        Inactive hide details for Tim Armstrong ---01/19/2017 01:33:27
        PM---Hi Valencia,  That error message should not directly cause Tim
        Armstrong ---01/19/2017 01:33:27 PM---Hi Valencia, That error
        message should not directly cause a crash - the query should

        From: Tim Armstrong <[email protected]>
        To: "dev@impala" <[email protected]>
        Cc: Valencia Serrao/Austin/Contr/IBM@IBMUS, Nishidha
        Panpaliya/Austin/Contr/IBM@IBMUS, Sudarshan
        Jagadale/Austin/Contr/IBM@IBMUS, Manish
        Patil/Austin/Contr/IBM@IBMUS
        Date: 01/19/2017 01:33 PM
        Subject: Re: impalad crashes at load_nested step in dataloading.



        Hi Valencia,
         That error message should not directly cause a crash - the query
        should continue to execute.

        I don't see any attached logs. My guess without any additional info
        is that there's a bug on your branch of Impala or related to
        PowerPC - I don't think we've seen any failures like that in recent
        memory.

        Thanks,
        Tim

        On Wed, Jan 18, 2017 at 11:12 PM, Valencia Serrao <
        [email protected]> wrote:
                    Hi All,

                    While executing the load_nested step in dataloading for
                    Impala2.6 on Ubuntu16.04 ppc64le, impalad crashes with
                    following message in logs:

                    "Could not codegen
                    PartitionedAggregationNode::UpdateTuple because
                    intermediate type STRING is not yet supported for
                    aggregate function "group_concat()""

                    Here are the logs: (See attached file:
                    impalad.testvm.test.log.INFO.20170118-211515.3848),
                    (See attached file: load_nested.py.log), (See attached
                    file: hs_err_pid3848.log)

                    Any guidance on this issue will be helpful.

                    Regards,
                    Valencia














Reply via email to