If you want reasonably transparent failover, you should always float the IP address to the surviving node. If the service fails, deal with it locally with a monitoring/watchdog script of some sort. If the service isn't recoverable locally, have the machine fail itself, and the services will float to the surviving one. If you split each service to a separate IP, though, you can do partial service/IP failover, as I think you're saying. That means you don't have to fail the whole machine if a service fails and won't restart. It should also provide some load balancing between the nodes since not all services will run on all nodes. Depending on where the application bottlenecks, this may be advantageous. On Fri, 17 Apr 2009 14:23:40 +0200, ESGLinux <esggrupos@xxxxxxxxx> wrote: > Hi again, > > now we speak the same language ;-) what do your think about my question?: > > Has it sense to configure ip resources as shared? > I think it must be better to configure always as a private resource, isnt > it? > > ESG > > > 2009/4/17 Gordan Bobic <gordan@xxxxxxxxxx> > >> Right, I see what you mean now. No, you can't do that - the service >> always >> has to run on the node where it's IP is. But you can fail over the whole >> resource group (IP + services) together if required. >> >> Gordan >> >> On Fri, 17 Apr 2009 14:01:05 +0200, ESGLinux <esggrupos@xxxxxxxxx> wrote: >> > Hi Gordan, >> > >> > I´m talking about the floating IP. (I called it service IP, is wrog >> > the >> > name?) >> > >> > The real ip of the two nodes is diferent and share the floating IP but >> > in >> > that scenario when one service fails (no the whole node) I get one >> service >> > in one node and one service on the other node but I get the error I >> posted. >> > >> > for example >> > floating IP + BBDD is runing on node1 >> > floating IP + HTTPD is running on node2 >> > >> > is this possible? >> > >> > thanks >> > >> > ESG >> > >> > >> > >> > 2009/4/17 Gordan Bobic <gordan@xxxxxxxxxx> >> > >> >> You can't have the same IP on multiple machines at the same time. That >> >> much should be obvious before even thinking about clustering. >> >> >> >> You can, however configure floating IPs as resources that RHCS will >> >> fail over between cluster nodes. Only one node will have a particular >> >> IP address at any one time, but if that node fails, the floating IP >> >> will get migrated to one of the surviving nodes. >> >> >> >> Any number of services can run on a floating IP. There is no need >> >> to have one floating IP per service. >> >> >> >> Gordan >> >> >> >> On Fri, 17 Apr 2009 13:23:30 +0200, ESGLinux <esggrupos@xxxxxxxxx> >> wrote: >> >> > Hi, >> >> > >> >> > thanks for your answer >> >> > >> >> > I suspected so, >> >> > >> >> > can I say as an axiom: One service needs One IP? ( One man, One vote >> >> > ;-) >> >> ) >> >> > >> >> > and if is this true, has it sense to configure ip resources as >> >> > shared? >> >> > it >> >> > must be better to configure as a private resource, isnt it? >> >> > >> >> > Greetings, >> >> > >> >> > ESG >> >> > >> >> > 2009/4/17 Juan Ramon Martin Blanco <robejrm@xxxxxxxxx> >> >> > >> >> >> >> >> >> >> >> >> On Fri, Apr 17, 2009 at 12:57 PM, ESGLinux <esggrupos@xxxxxxxxx> >> >> >> wrote: >> >> >> >> >> >>> Hello all, >> >> >>> >> >> >>> I have a doubt about using the same service ip with two diferent >> >> >>> services >> >> >>> in two diferent nodes of my two-nodes-cluster. >> >> >> >> >> >> >> >> >>> >> >> >>> I´ll explain it a litle: >> >> >>> >> >> >>> I have two services: BBDD and HTTPD >> >> >>> >> >> >>> I have configured a shared IP: 192.168.1.100 >> >> >>> >> >> >> Hi, >> >> >> You _must_ use a different IP, cannot have the same IP on >> >> >> different >> >> >> machines. >> >> >> >> >> >> Greetings, >> >> >> Juanra >> >> >> >> >> >>> >> >> >>> two nodes: node1 and node2. >> >> >>> >> >> >>> When I run the two services on node1 all runs ok. If I try to >> >> >>> relocate >> >> >>> one >> >> >>> service, with >> >> >>> >> >> >>> clusvcadm -r BBDD -m node2 >> >> >>> >> >> >>> it fails with the error: >> >> >>> in.rdiscd[4805]: setsockopt (IP_ADD_MEMBERSHIP): Address already >> >> >>> in >> >> use >> >> >>> in.rdiscd[4805]: Failed joining addresses >> >> >>> >> >> >>> I suposse that is because one services is on node1 with the ip and >> >> >>> the >> >> >>> other trys to run in node2 with the same ip. >> >> >>> >> >> >>> So my question is if it is possible to run the services this way >> >> >>> or >> I >> >> >>> need >> >> >>> an ip per service? >> >> >>> (I have tested that with 2 diferent ips I can run BBDD on node1 an >> >> HTTPD >> >> >>> on node2) >> >> >>> >> >> >>> Thanks in advance >> >> >>> >> >> >>> ESG >> >> >>> >> >> >>> >> >> >>> -- >> >> >>> Linux-cluster mailing list >> >> >>> Linux-cluster@xxxxxxxxxx >> >> >>> https://www.redhat.com/mailman/listinfo/linux-cluster >> >> >>> >> >> >> >> >> >> >> >> >> -- >> >> >> Linux-cluster mailing list >> >> >> Linux-cluster@xxxxxxxxxx >> >> >> https://www.redhat.com/mailman/listinfo/linux-cluster >> >> >> >> >> >> >> -- >> >> Linux-cluster mailing list >> >> Linux-cluster@xxxxxxxxxx >> >> https://www.redhat.com/mailman/listinfo/linux-cluster >> >> >> >> -- >> Linux-cluster mailing list >> Linux-cluster@xxxxxxxxxx >> https://www.redhat.com/mailman/listinfo/linux-cluster >> -- Linux-cluster mailing list Linux-cluster@xxxxxxxxxx https://www.redhat.com/mailman/listinfo/linux-cluster
