[PATCH 00/12] Standardize c/r error reporting (v3)

Oren Laadan orenl at librato.com
Tue Nov 3 08:51:57 PST 2009



serue at us.ibm.com wrote:
> From: Serge E. Hallyn <serue at us.ibm.com>
> 
> Here is a completely new approach, basically verbatim implementing
> Oren's recipe from last Friday.  It only implements the error part
> to replace ckpt_write_err.  The intent would be for ckpt_debug to
> also be replaced by ckpt_msg() which would be similar to ckpt_err()
> except adding a prefix to the message.
> 
> My goal is for users to be able to get real errors printed to a logfile
> for checkpoint and restart (not have to check dmesg) in ckpt-v19,
> especially for things like 'oh you're on a btrfs which is not supported'.
> Having to check dmesg seems to walk right into the 'toy implementation'
> argument.
> 
> Thanks Oren for the detailed explanation of what you want to see, and
> Matt for several excellent cleanup suggestions.
> 
> If there are no major objections then I'll add ckpt_err()s at restart
> on top of this patchset, but probably hold off on converting ckpt_debugs
> for a bit.  Note that the heavyweight semaphore means that for real
> debugging we're only increasing the chances of hiding bugs with the
> debugging, so I'm open to the idea of keeping ckpt_debug() as is.

"Heavyweight semaphore" ?

Maybe also mention that during checkpoint only one process uses this
interface - no contention;  During restart, for the most part, one
process runs at a time, except when transferring control to the the
next process in line - contention unlikely too.

Oren.



More information about the Containers mailing list