[Devel] Re: [PATCH 00/17] Standardize c/r error reporting

[Oren Laadan]

serue at us.ibm.com wrote:
> From: Serge E. Hallyn <serue at us.ibm.com>
> 
> This patchset
> 	1. defines ckpt_error() 
> 	2. allows users to pass a logfd into sys_checkpoint and
> 	   sys_restart
> 	3. Switches ckpt_write_err() to accepting a single enhanced
> 	   format string, instead of two separate formats.
> 	4. Has ckpt_write_err() call ckpt_error() to also log the
> 	   error in the user-provided logfile and syslog.
> 
> Every ckpt_error() message is prefixed by current's global pid,
> current's virtual pid, number of bytes read/written, and the
> ctx->errno.

Hmmm... I thought that we'd have:

ckpt_error() will be used _only_ when there is an error.
(will always print ctx->errno, which is assumed to be set)

ckpt_debug() will be used to report "log-able" debugging info
(will not print ctx->errno, because there wasn't any error)

_ckpt_debug() will be used to throw debugging info that isn't
logged (only goes to dmesg) and will not be compiled unless
CONFIG_CHECKPOINT_DEBUG is selected. This is useful for those
debug messages that are a pain to add remove, but that we
don't want to spit to users' logs.

Maybe rename ckpt_debug() to ckpt_log() and then _ckpt_debug()
to ckpt_debug()...

I think it's confusing for ckpt_error() to be used everywhere.

Oren.

> 
> I'm sure we'll want another round if only to alter the format,
> but I'm finding it very useful as is.
> 
> (As more callers are added I expect we'll want to add %(A) to
> the format to indicate active_pid should be printed).
> 
> -serge
_______________________________________________
Containers mailing list
Containers at lists.linux-foundation.org
https://lists.linux-foundation.org/mailman/listinfo/containers