Dejan Muhamedagic schrieb am 04.04.2011 14:35:34: > On Fri, Mar 18, 2011 at 04:15:16PM +0100, [email protected] wrote: > > Hi, > > > > Dejan Muhamedagic schrieb am 18.03.2011 14:31:08: > > > Hi, > > > > > > On Wed, Mar 16, 2011 at 04:58:25PM +0100, Corvus Corax wrote: > > > > > > > > IPAddr2 puts the interface up on start and down on stop. > > > > But its not able to detect an UP or DOWN change in status or monitor. > > > > > > > > Therefore an "ifconfig <interface> down" from a thrird program or a > > > > careless administrator would drop the link without pacemaker noticing! > > > > > > Hmm, careless administrator is somewhat of a paradox, right? > > > > > > Really, what was your motivation for this? It makes me wonder, > > > since this RA has existed for many years and so far nobody > > > bothered to test this. > > > > Hm, maybe the idea behind is not totally new. Remember this thread: > > http://lists.community.tummy.com/pipermail/linux-ha-dev/2011-February/018184.html
> > > > I would go with the remarks of LMB, that this is something closer to > > the pingd than to Ipaddr2. Isn't the real intention of both post, that you > > want to know, if your network interface is vital ? > > Yes. > > > You may use pingd for that, but someone may be concerned to ping the right > > remote device (also a default-gateway might not be a very static thing in > > a modern network). > > > > My imagination is currently an agent (let's call it ethmonitor) that > > monitors > > a network interface with a combination of the fine methods that Robert > > Euhus > > has posted in his patch. Than you could define some rules in CIB how to > > react on the event of a failed network interface. Sure this assumes that > > you > > do your heartbeats over more than one interface. > > > > It would check: > > 1. interface link up ? > > 2. does the RX counter of the interface increase during a certain amout > > of time ? > > 3. do I have some other nodes in my arp-cache which I could arping ? > > 4. maybe retry all checks to overcome short outages > > If all questions are answered with NO - the interface is dead. > > > > I would add my vote for such a feature. > > Just took a look at the thread you referenced above. > Unfortunately, the author didn't get back with the new code > after review and short discussion. > Now I took the code from Robert in the above referenced thread and put it into a complete new RA. It is based very much on the existing pind agent, but implements the monitoring like discussed above. Please let me know, what you think about it. Cheers, Alex
ethmonitor
Description: Binary data
_______________________________________________________ Linux-HA-Dev: [email protected] http://lists.linux-ha.org/mailman/listinfo/linux-ha-dev Home Page: http://linux-ha.org/
