Integrate BLCR to GDB

Integrate BLCR to GDB

[Zhiyong]

Hi,
  If my understanding is correct, current gdb checkpoint/restart is using
fork() and does not support multi-threading. 

  I am interested to know if  someone has done some work to integrate BLCR
(Berkely Lab Checkpoint/Restart ,
https://ftg.lbl.gov/CheckpointRestart/CheckpointRestart.shtml) into gdb? 

  As a standalone, BLCR is quite helpful in certain debugging scenarios. But
it is a bit hard to use with gdb , i.e. breakpoints can not be saved and
restart is not convienent to stop at certain point etc. 

  BLCR's limitation is it is only for Linux x86 and x86_64. It also needs to
load kernel modules which may or may not be an issue. It is not targeting
debugging but rather process management/fault tolerance. 
 

  Thanks,
  Zhiyong
  
  
  
-- 
View this message in context: http://www.nabble.com/Integrate-BLCR-to-GDB-tp22891922p22891922.html
Sent from the Sourceware - gdb list mailing list archive at Nabble.com.

Re: Integrate BLCR to GDB

[Michael Snyder]

Zhiyong wrote:
> Hi,
>   If my understanding is correct, current gdb checkpoint/restart is using
> fork() and does not support multi-threading. 
> 
>   I am interested to know if  someone has done some work to integrate BLCR
> (Berkely Lab Checkpoint/Restart ,
> https://ftg.lbl.gov/CheckpointRestart/CheckpointRestart.shtml) into gdb? 
> 
>   As a standalone, BLCR is quite helpful in certain debugging scenarios. But
> it is a bit hard to use with gdb , i.e. breakpoints can not be saved and
> restart is not convienent to stop at certain point etc. 
> 
>   BLCR's limitation is it is only for Linux x86 and x86_64. It also needs to
> load kernel modules which may or may not be an issue. It is not targeting
> debugging but rather process management/fault tolerance. 


Sounds like a very interesting project.

Would you be interested in spearheading it?

Re: Integrate BLCR to GDB

[Zhiyong]

Michael Snyder-6 wrote:
> 
> 
> Sounds like a very interesting project.
> 
> Would you be interested in spearheading it?
> 
> 
> 


Hi Michael,

   I am totally new to gdb so I am afraid I will need quite some help from
someone here with experience. But I am more than happy to spend some time on
it.

   I am also new to BLCR. My previous short description of BLCR is
incorrect. For example BLCR does support x86, x86-64, ppc, ppc64 and ARM and
Xen dom0/domU. This recent presentation may give a more complete picture
about BLCR : https://ftg.lbl.gov/CheckpointRestart/Pubs/WTTC2008-BKK.pdf. 

   I see there are other approach/solutions out there to achieve similar
goals in terms of debugging: like using visualization VM or effort on
reversible debugging. In my working area, I have to deal with programs with
huge memory footprint ( > 16G ),  long runtime  like in days and
multi-threaded. I feel BLCR approach fits the requirement nicely. Actually I
can use BLCR as standalone for my debugging purpose. I just wish I could
easily do the checkpoint/restart inside gdb. 

   I probably will  also contact BLCR folks  to see if they already have
something done for gdb or not.

  Thanks,
  Zhiyong. 

 


-- 
View this message in context: http://www.nabble.com/Integrate-BLCR-to-GDB-tp22891922p22917611.html
Sent from the Sourceware - gdb list mailing list archive at Nabble.com.

Re: Integrate BLCR to GDB

[Michael Snyder]

Zhiyong wrote:
> 
> Michael Snyder-6 wrote:
>>
>> Sounds like a very interesting project.
>>
>> Would you be interested in spearheading it?
>>
>>
>>
> 
> 
> Hi Michael,
> 
>    I am totally new to gdb so I am afraid I will need quite some help from
> someone here with experience. But I am more than happy to spend some time on
> it.
> 
>    I am also new to BLCR. My previous short description of BLCR is
> incorrect. For example BLCR does support x86, x86-64, ppc, ppc64 and ARM and
> Xen dom0/domU. This recent presentation may give a more complete picture
> about BLCR : https://ftg.lbl.gov/CheckpointRestart/Pubs/WTTC2008-BKK.pdf. 
> 
>    I see there are other approach/solutions out there to achieve similar
> goals in terms of debugging: like using visualization VM or effort on
> reversible debugging. In my working area, I have to deal with programs with
> huge memory footprint ( > 16G ),  long runtime  like in days and
> multi-threaded. I feel BLCR approach fits the requirement nicely. Actually I
> can use BLCR as standalone for my debugging purpose. I just wish I could
> easily do the checkpoint/restart inside gdb. 
> 
>    I probably will  also contact BLCR folks  to see if they already have
> something done for gdb or not.

Thanks.

This is an area that interests me personally -- unfortunately I am
very busy these days.

With respect to doing the checkpoints from within gdb --
do you know if there are shell commands for doing that?
If so, you might be able to write a combination of gdb
scripts and shell commands.

Michael

Re: Integrate BLCR to GDB

[Thiago Jung Bauermann]

El lun, 06-04-2009 a las 15:12 -0700, Michael Snyder escribió:
> With respect to doing the checkpoints from within gdb --
> do you know if there are shell commands for doing that?
> If so, you might be able to write a combination of gdb
> scripts and shell commands.

But would we ship those scripts and shell commands then? It'd be better
if GDB could integrate nicely with BLCR if it finds it is available.
-- 
[]'s
Thiago Jung Bauermann
IBM Linux Technology Center

Re: Integrate BLCR to GDB

[Zhiyong]

Thiago Jung Bauermann-2 wrote:
> 
> 
> But would we ship those scripts and shell commands then? It'd be better
> if GDB could integrate nicely with BLCR if it finds it is available.
> -- 
> []'s
> Thiago Jung Bauermann
> IBM Linux Technology Center
> 
> 
> 

By default BLCR does not do checkpoint on a process that is beining ptraced.
If forcing it, at least I see right now I have to hit "cont" multiple times
in gdb to let the process continue. So it would be better to have  a tighter
integration of BLCR . Another point is to save breakpoints etc that is
related to gdb process into the checkpointed data file, for the convince of
restarting on a totally different gdb session.  

Zhiyong 
-- 
View this message in context: http://www.nabble.com/Integrate-BLCR-to-GDB-tp22891922p22919013.html
Sent from the Sourceware - gdb list mailing list archive at Nabble.com.