[0/3] multipath-tools: fixes for systemd watchdog

Message

Martin Wilck Nov. 14, 2024, 2:42 p.m. UTC

A set of small fixes to make multipathd better adhere to the conventions
for systemd's WatchdogSec= setting.

Martin Wilck (3):
  libmultipath: don't print error message if WATCHDOG_USEC is 0
  libmultipath: honor WATCHDOG_PID setting
  libmultipath: cut watchdog interval in half

 libmultipath/config.c             | 21 ++++++++++++++++++---
 libmultipath/config.h             |  2 ++
 libmultipath/libmultipath.version |  1 +
 3 files changed, 21 insertions(+), 3 deletions(-)

Comments

Benjamin Marzinski Nov. 16, 2024, 12:05 a.m. UTC | #1

On Thu, Nov 14, 2024 at 03:42:21PM +0100, Martin Wilck wrote:
> A set of small fixes to make multipathd better adhere to the conventions

I'm a little confused by this. Your patches make sense, but why do we
bother limiting max_checkint to the Watchdog timer? We send a notify
every loop in checkerloop() regardless. Changing max_checkint won't make
us any less likely to hang for too long if we're using a synchronous
chacker, and I can't figure out why the Watchdog time should have
anything to do with how frequently we check working paths.

Your code changes looks fine, so assuming there's a use for this code:
Reviewed-by: Benjamin Marzinski <[e-mail]>

But possibly we should be removing set_max_checkint_from_watchdog(), or
replacing max_checkint with something like checker_timeout, but only for
synchronous checkers. Or possibly I'm just missing something here.

-Ben

> for systemd's WatchdogSec= setting.
> 
> Martin Wilck (3):
>   libmultipath: don't print error message if WATCHDOG_USEC is 0
>   libmultipath: honor WATCHDOG_PID setting
>   libmultipath: cut watchdog interval in half
> 
>  libmultipath/config.c             | 21 ++++++++++++++++++---
>  libmultipath/config.h             |  2 ++
>  libmultipath/libmultipath.version |  1 +
>  3 files changed, 21 insertions(+), 3 deletions(-)
> 
> -- 
> 2.47.0