Note: "permalinks" may not be as permanent as we would like,
direct links of old sources may well be a few messages off.
Lars Ellenberg ha scritto: > On Thu, Sep 27, 2007 at 01:06:39PM +0200, Matteo Campana wrote: >> Hi Lars and all the guys in ML :). >> My problem is to understand the behaviur of my HA system: I have read in >> the documentation about heartbeat that at least 2 channels of >> communication between 2 nodes are required to preserve the system from >> the "split brain" problem (in my system I have 1 ethernet + 1 serial). >> But....in my tests I see however that DRBD says the "infamous" :) >> message "split-brain detected" ecc.. . >> My test: >> >> * I disconnect primary node from the ethernet >> * secondary ----> primary (ok) >> * I reconnect primary >> * secondary still remain primary (ok for my config) >> * the problem is that the two nodes are standalone and not >> re-syncronize theirs drbd partition. >> >> DRBD does not use also the serial channel in order to communicate >> between the nodes? > > um. it does use the heartbeat communication infrastructure, > if you configure the "outdate-peer handler", in which case it had > outdated the other node, before you try to make it primary, > thus preventing you from going online with stale data. I need to recompile heartbeat to use the "outdate-peer" plugin? (currently I'm using the heartbeat package for Debian). > > or did you just suggest that we should mirror block io > via 9600 baud serial line? > , > Oh...no Lars :):) But it was no clear for me how to tell drbd to use heartbeat communication, that is the "outdate-peer handler" tips that you have explain before. Thank you! Matteo. -- * Ing. Matteo Campana System Engineer Klarya s.r.l.* **Open Source IP Communications* __________________________________________________