[Top][All Lists]

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

Re: [HEADS-UP] Rawhide: /tmp is now on tmpfs

From: Pádraig Brady
Subject: Re: [HEADS-UP] Rawhide: /tmp is now on tmpfs
Date: Thu, 13 Sep 2012 01:56:41 +0100
User-agent: Mozilla/5.0 (X11; Linux x86_64; rv:6.0) Gecko/20110816 Thunderbird/6.0

On 06/01/2012 02:05 PM, Pádraig Brady wrote:
It seems like we should default to /var/tmp in sort(1) and tac(1),
as debian and fedora at least are probably moving to tmpfs for /tmp.
I'll monitor the situation a bit before making the change.

Thinking a bit more about this, as I see it:

  /tmp =     stateless
  /var/tmp = stateful

What is actually used to back those paths is system specific
and inconsequential to `sort`. `sort` logically just needs stateless
storage, as after sort exits, its tmp files are no longer useful.
If `sort` could restart operations then that would be a different story.

So `sort` should stick to /tmp I think, and not worry
about system implementation details.

As an aside on tech for backing those paths:
Ideally, tmpfs could be used for /tmp even for large files as
swap could be used as needed. This swap could even be
over nfs/nbd for nodes without stateful storage.
  If the swap (over net) implementation isn't up to scratch,
then as a pragmatic solution one might need to configure
paths considering 'size' as well as 'statefulness' and
have /large_tmp which is backed by a traditional file system
on disk and auto cleaned on boot. Disadvantages of that are:
  1. require boot cleanup scripts
  2. traditional file system on disk doesn't take advantage
     of the fact that we don't _need_ to persist the data
     to disk/flash/net (which can be slower and less reliable).
Traditional disk backed /tmp avoids possible swap issues
with large files, but suffers from the above 2 disadvantages.

Another pertinent piece of info is that Debian
have reverted the default setting of tmpfs backed /tmp:


reply via email to

[Prev in Thread] Current Thread [Next in Thread]