[Info-ingres] Weird problem in Ingres 10

Thu May 17 12:05:36 UTC 2018

Can you attach the relevant sc930 output files with a note as to what timestamp is
the relevant one(s);  or alternatively email them to schendel at kbcomputer.com
and I'll take a look when I get a spare moment.  (Which isn't looking too likely today,
but tomorrow might work.)

Karl

> On May 17, 2018, at 7:57 AM, Mark <i at dontgetlotsofspamanymore.net> wrote:
> 
> On Thu, 17 May 2018 11:21:37 +0000, Martin Bowes
> <martin.bowes at ndph.ox.ac.uk> wrote:
> 
>> Twice...
>> 
>> Once where it updates the data and once where it puts it back the way it was?
> 
> No.  The data is the same for each.
> 
>> What do the Rowcount figures say on each statement?
> 
> 0.
> 
>> Marty
>> 
>> -----Original Message-----
>> From: Mark [mailto:i at dontgetlotsofspamanymore.net] 
>> Sent: 17 May 2018 12:12
>> To: info-ingres at lists.planetingres.org
>> Subject: Re: [Info-ingres] Weird problem in Ingres 10
>> 
>> On Wed, 16 May 2018 12:40:48 -0400, Karl and Betty Schendel
>> <schendel at kbcomputer.com> wrote:
>> 
>>> SC930 tracing is server wide and runs until you stop it with set notrace sc930.
>>> So you can connect to any database, do the set trace record '/some/ingres-writable/dir'
>>> and the set trace point sc930, go to the trace record directory you gave it and
>>> ensure that there's at least something there (should be one or more sessNNNNN files),
>>> and then you can run your esql program until the fault happens.
>>> 
>>> You don't actually need to put the sc930 into the esql program unless you want to
>>> do it that way.
>>> 
>>> Once you capture what you need to capture, set notrace point sc930 turns it off.
>>> 
>>> You'll need the trace point privilege, simplest is to do it as the installation owner
>>> (user ingres, traditionally).
>> 
>> Thanks again.  We now have traces but they don't help me.  The traces
>> include the SQL statement and the data but, pretty much, nothing else.
>> 
>> The only notable thing I observed is that the data is traced twice for
>> the failed transaction and only once for the successful ones.
>> 
>> There are no error messages.
>> 
>>> Karl
>>> 
>>>> On May 16, 2018, at 11:12 AM, Mark <i at dontgetlotsofspamanymore.net> wrote:
>>>> 
>>>> On Wed, 16 May 2018 12:11:08 +0000, Martin Bowes
>>>> <martin.bowes at ndph.ox.ac.uk> wrote:
>>>> 
>>>>> Just to expand on Karl's sc930...You may already know this...
>>>>> 
>>>>> To turn it on, make a recording directory...mkdir /full/path/to/directory'
>>>>> 
>>>>> And then...
>>>>> sql iidbdb << SQL_END
>>>>> set trace record '/full/path/to/directory';
>>>>> set trace point sc930 1;
>>>>> \p\g
>>>>> \q
>>>>> SQL_END
>>>> 
>>>> Do I need to run this on the server account?  I have implemented it
>>>> currently in the embedded SQL of my program.
>>>> 
>>>>> FYI. The digit after the sc930 indicates a tracing level, 1 should be sufficient.
>>>>> 
>>>>> Run the errant query.
>>>>> 
>>>>> And turn off the sc930.
>>>>> sql iidbdb << SQL_END
>>>>> set trace point sc930 0;
>>>>> \p\g
>>>>> \q
>>>>> SQL_END
>>>>> 
>>>>> You can now access the recording directory and start scanning the files for any sign of life from your query.
>>>>> 
>>>>> Marty
>>>>> 
>>>>> -----Original Message-----
>>>>> From: Karl and Betty Schendel [mailto:schendel at kbcomputer.com] 
>>>>> Sent: 16 May 2018 12:54
>>>>> To: Ingres and related product discussion forum
>>>>> Subject: Re: [Info-ingres] Weird problem in Ingres 10
>>>>> 
>>>>> On May 16, 2018, at 7:44 AM, Mark <i at dontgetlotsofspamanymore.net> wrote:
>>>>>> 
>>>>>> On Tue, 15 May 2018 12:27:09 -0400, Karl and Betty Schendel
>>>>>> <schendel at kbcomputer.com> wrote:
>>>>>> 
>>>>>>> It's certainly not something I have heard of or seen before.  Do you have any rules
>>>>>>> defined on the relevant tables?  Try enabling LOG_TRACE if the problem is
>>>>>>> sufficiently predictable, or do a logdump after the problem occurs if it's not;
>>>>>>> the idea being to try to see whether you actually got any PUT (insert) or
>>>>>>> REP (replace) log records that were then rolled back, or whether the insert / update
>>>>>>> was never executed at all.
>>>>>> 
>>>>>> After enabling log_trace all I got was:
>>>>>> 
>>>>>>  LOG: SAVEPOINT   Size written/reserved:      0/     0  Flags:
>>>>>> -------------------------------------------------------------------
>>>>> 
>>>>> So the insert/update isn't ever being executed.  Either it's failing with some sort of
>>>>> silent error, which would seem odd, or it's not reaching the backend at all, or
>>>>> it's being pre-empted by a before rule.  I think the next step would be to enable
>>>>> sc930 tracing and see if the backend is getting the insert-update, and what
>>>>> end-of-query status it's recording.  There should be KB articles on enabling
>>>>> SC930 tracing.
>>>>> 
>>>>> Karl
>>>>> 
>>>>> _______________________________________________
>>>>> Info-ingres mailing list
>>>>> Info-ingres at lists.planetingres.org
>>>>> http://lists.planetingres.org/mailman/listinfo/info-ingres
>>>> 
>>>> -- 
>>>> <insert witty sig here>
>>>> _______________________________________________
>>>> Info-ingres mailing list
>>>> Info-ingres at lists.planetingres.org
>>>> http://lists.planetingres.org/mailman/listinfo/info-ingres
> 
> -- 
> <insert witty sig here>
> _______________________________________________
> Info-ingres mailing list
> Info-ingres at lists.planetingres.org
> http://lists.planetingres.org/mailman/listinfo/info-ingres