[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: Bug#191036: create /run for programs that run before /var is mounted



I agree with Manoj when he says:
> I think it is premature to [r]atify into policy an action
> that has not been fully decided upon, and has not yet had
> all the kinks ironed out [...]

However, there does need to be some discussion of what
properties /run/ will have.  Early in the debian-devel 
discussion it was suggested that /run/ be required to be
RAM-based, that it be initialised with a skeleton, etc.

As the discussion went on there seemed to be agreement 
that /run/'s properties should be determined by the 
problem it is solving, namely, the problem of there being
nowhere to store runtime state prior to the availability of
/var/run/ on a system that mounts /var/ from the network.
To solve this problem, /run/ needs to be available locally
so that it can be mounted early, probably by
/etc/rcS.d/S35mountall.sh .

I would argue that for the present /run/ should be so defined,
and otherwise should be implemented as simply as possible.

Proposed list of properties of /run/:
* Available and empty after /etc/rcS.d/S35mountall.sh runs
* No skeleton
* Files stored in /run/ are not "reaped"
* A program should *only* use /run/ if it *must* do so because
  it has to store something before /var/run/ becomes available
  on systems that mount /var/ over NFS.
-- 
Thomas Hood <jdthood0@yahoo.co.uk>



Reply to: