As Selva mentions, we want to , as part of error handling , sometimes want
to  create core dump within Traf code  for certain conditions wehre we
current assert and  return with no information. We sometimes simply cannot
debug without a corefile expecially for some extremem and unusual error
conditions. For these, Trafodion  seabaed layer provides this API
msg_mon_dump_process_name().
This is different from the cases where the process aborts and dies.

So the suggestion from Hans/Gunnar is that  we follow the same pattern as
set by proc/sys/kernel/core_pattern file for these cases  as well that will.
be used mainly by us for debugging . I think that sounds fine to me.  Today
it creates them usually on the current directory and that has to change.

Sandhya
-----Original Message-----
From: Hans Zeller [mailto:[email protected]]
Sent: Thursday, November 12, 2015 4:43 PM
To: dev <[email protected]>
Subject: Re: Trafodion and core file location ?

Here are some recommendations that I think many of us adopted over the
years:

   - Include process name, pid, user in the core file name.
   - Collect all the cores in a single directory or directory tree (at
   least on production systems).
   - Consider mounting a separate disk partition for that directory tree,
   so that cores can't fill up the main disk and bring down the system.
   - As Gunnar and Selva mentioned, the rules for core file names and
   directories should be established in /proc/sys, not in our code.


Hans

On Thu, Nov 12, 2015 at 4:37 PM, Selva Govindarajan <
[email protected]> wrote:

> When an application dumps core by issuing abort() or
> exit(non_zero_value), the core location is usually determined by
> /proc/sys/kernel/core_pattern.
> When Trafodion engine explicitly dump core for certain conditions like
> dumping ssmp core before bringing down the node, we might be able to
> control the location. Ideally, we would want these dumps also to
> conform to core_pattern, but might be bit involved to do that.  We
> might have chosen an easy route to dump the core at some location.
>
> Selva
>
> -----Original Message-----
> From: Dave Birdsall [mailto:[email protected]]
> Sent: Thursday, November 12, 2015 4:21 PM
> To: [email protected]
> Subject: RE: Trafodion and core file location ?
>
> Hi,
>
> So, with the current situation, where do core files land? I sometimes
> see them in $MY_SQROOT/sql/scripts, and I've also seen them in
> directories where I was running sqlci. Is this the norm for the
> current situation?
>
> Yes, I think it would be great to have a knob to place them where we
> want them. Would be easier to find. And ultimately easier to manage
> both from a user perspective and a support perspective.
>
> Dave
>
> -----Original Message-----
> From: Sandhya Sundaresan [mailto:[email protected]]
> Sent: Thursday, November 12, 2015 4:05 PM
> To: [email protected]
> Subject: Trafodion and core file location ?
>
> Hi ,
>
> Has anyone thought of a standard location for trafodion core files to
> reside ?  There is a programmatic interface we currently have to
> create a core file if needed. Today it doesn’t take a location param .
> I was going to change it to allow a location to be specified.
> Specifying a location for the core files we create may good so we
> taking up space have a way of not inadvertently where we shouldn’t be.
>
>  Today it just puts it on ‘pwd’ i.e current directory.
>
> We could  have an envvar to specify a location  for all core files
> with current dir being the default.
>
> Any thoughts on this topic ?
>
> Thanks
>
> Sandhya
>

Reply via email to