Date: Mon, 22 May 2000 14:54:47 -0300 (BRST)
From: Luis Claudio R. Goncalves <lclaudio@conectiva.com.br>
To: Alan Robertson <alanr@unix.sh>
Subject: Re: Patch


Hi!

   I've read you comments about the patch and figured out a possible
mistake I did. Well, let me explain a bit about the way this patch works:

   Let's imagine the possible starting conditions:

1) Node 1 is starting and node 2 is down (or vice-versa):
   Take the resources. (If the primary is dead, do that via
mark_node_dead(). Else do that via req_our_resources())

2) Node 1 and node two are starting at the same time:
   Let's make both machine req_our_resources(). The primary (defined
in /etc/ha.d/haresources) will get his resources. If both machines
have resources defined in the file, each one will hold his own resources.

3) Node 1 is starting and node 2 has no resources:
   Just like the above (#2).

4) Node 1 is starting and node 2 has (his) local resources:
   Let's ask for our local resources. (req_our_resources()).

5) Node 1 is starting and node 2 has both local and foreign (all) resources:
   Do nothing. :)

   Note that if you have more than two nodes, this may work. But as I
said before, this is just to use before the API is ready. :)

   The possible resources_held messages are: "I don't hold resources",
"I hold local resources", "I hold foreign resources" and "I hold all
resources". I don't create lists of resources anymore. Just what kind
of resources I hold, if so.

   I agree with you about the sequence number in the messages. But if
a message take long to be retransmitted, it may confuse the
cluster. Anyway as this stuff is used only for the starting process,
it doesn't hurt anything.

   The mistake I found relates to resources_timer. It may fit in the
same place but may not be related to send_starting_now that happens in
the first ten seconds (only while ((now - starttime) <
RQSTDELAY) ). To fix it I've splitted the if into two ifs. :)

   That's the way the new stuff works. I sent it to you to have a more
experienced oppinion. Thanks again! :)

							Luis

[ Luis Claudio R. Goncalves                  lclaudio@conectiva.com.br ]
[ BSc in Computer Science -- MSc coming soon -- Gospel User -- Linuxer ]
[ Fault Tolerance - Real-Time - Distributed Systems - IECLB - IS 40:31 ]
[ LateNite Programmer --  Jesus Is The Solid Rock On Which I Stand  -- ]