[Spread-users] Re: Bug in spread - message counter
John Lane Schultz
jschultz at spreadconcepts.com
Wed Sep 3 13:25:12 EDT 2008
Sorry this hasn't been fixed comprehensively yet. We are going to post a rough fix to this problem in the next couple of days. Basically, if the counter gets too high, then we will "manually" invoke network membership change, which will reset the counters back to zero.
The reason why starting, killing or restarting a daemon eliminates the issue is because when the membership of daemons changes they start a new membership that resets all the counters back down to zero.
John Lane Schultz
Spread Concepts LLC
Phn: 443 838 2200
Fax: 301 560 8875
Wednesday, September 3, 2008, 9:24:08 AM, you wrote:
> As mentioned before spread will become totally unusable after about
> 2^31 messages have been sent.
> This can simply be reproduced using spflooder with small messages.
> Knowing that the spread system will become blocked after a certain
> amount of time (e.g. number of messages) is not acceptable for me.
> Is somebody already working on the spread message counter enlargement
> topic (int64 instead of int32)?
> I observed restarting _one_ of my spread daemons or starting an
> additional spread daemon will cause the message counter to start from
> zero again, if this is done _before_ the blocking counter limit is
> Thus it should be possible to achieve a counter reset in some way
> _without_ configuring DangerousMonitor=true.
> Can you tell me what steps are performed when a new spread daemon
> comes up and joins an already running spread system? Why will this
> lead to a counter reset?
> Of course solving the counter length problem would be better as a workaround.
> Spread-users mailing list
> Spread-users at lists.spread.org
More information about the Spread-users