pagg
[Top] [All Lists]

Re: Feedback requested on 'task notifier' patch

To: kingsley@xxxxxxxxxx
Subject: Re: Feedback requested on 'task notifier' patch
From: Erik Jacobson <erikj@xxxxxxx>
Date: Fri, 28 Oct 2005 15:54:38 -0500
Cc: Erik Jacobson <erikj@xxxxxxx>, pagg@xxxxxxx
In-reply-to: <20051028061159.GH16575@aurema.com>
References: <20051027140125.GA337@sgi.com> <20051028061159.GH16575@aurema.com>
Sender: pagg-bounce@xxxxxxxxxxx
User-agent: Mutt/1.5.6i
My guess is a semaphore for these locks just simply won't cost much but
we have to test it.

A couple of the ideas floating around seem to support the idea of layering
pagg/pnotify on top some how.

I'm honestly confused as to what the best next step is.  I think Matt
Helsley's patch that includes per-task and global task notification
has promise but has locking issues even for the per-task pieces (no
protection of the task watcher list per task).  

Matt's patch is close to working I think.  If he gets a working version
posted, maybe I'll try to write up a pnotify-like-thing for it.  I may need
to add list protections to Matt's patch first though. I just don't know :)

On the other hand, Chandra's idea is more simple but some may dislike the 
global notification feature (of course, that's present in Matt's patch
too I guess, but in Chandra's, it's the onliy option).  Chandra's idea is
to implement pnotify on top of the patch and add pieces to the task struct
for the per-task data.  In effect, I believe this means adding a notification
layer.  pnotify gets notified and then needs to notify it's users.

Erik

<Prev in Thread] Current Thread [Next in Thread>