[Linux-HA] strange heartbeat shutdown
) standalone mode STARTUP >Jan 27 06:07:12 swordfish ResourceManager[5458]: info: Running >/etc/init.d/httpd start >Jan 27 06:07:13 swordfish ResourceManager[5458]: info: Running >/etc/ha.d/resource.d/mon start >Jan 27 06:07:13 swordfish mach_down[5432]: info: >/usr/share/heartbeat/mach_down: nice_failback: foreign resources acquired >Jan 27 06:07:13 swordfish mach_down[5432]: info: mach_down takeover complete >for node moray.bmrb.wisc.edu. ... >Jan 27 06:09:14 swordfish heartbeat: [2778]: info: Heartbeat shutdown in >progress. (2778) >Jan 27 06:09:14 swordfish heartbeat: [5843]: info: Giving up all HA resources. >Jan 27 06:09:14 swordfish ResourceManager[5856]: info: Releasing resource >group: moray.bmrb.wisc.edu 144.92.217.20 proftpd httpd mon >Jan 27 06:09:14 swordfish ResourceManager[5856]: info: Running >/etc/ha.d/resource.d/mon stop >Jan 27 06:09:14 swordfish ResourceManager[5856]: info: Running >/etc/init.d/httpd stop >Jan 27 06:09:14 swordfish ResourceManager[5856]: info: Running >/etc/init.d/proftpd stop >Jan 27 06:09:14 swordfish proftpd[5705]: swordfish.bmrb.wisc.edu - ProFTPD >killed (signal 15) >Jan 27 06:09:14 swordfish proftpd[5705]: swordfish.bmrb.wisc.edu - ProFTPD >1.3.1 standalone mode SHUTDOWN >Jan 27 06:09:14 swordfish ResourceManager[5856]: info: Running >/etc/ha.d/resource.d/IPaddr 144.92.217.20 stop >Jan 27 06:09:14 swordfish IPaddr[5981]: INFO: ifconfig eth0:0 down >Jan 27 06:09:14 swordfish IPaddr[5964]: INFO: Success >Jan 27 06:09:14 swordfish heartbeat: [5843]: info: All HA resources >relinquished. >Jan 27 06:09:16 swordfish ntpd[2383]: Deleting interface #9 eth0:0, >144.92.217.20#123, interface stats: received=0, sent=0, dropped=0, >active_time=163 secs >Jan 27 06:09:16 swordfish heartbeat: [2778]: info: killing HBREAD process 2805 >with signal 15 >Jan 27 06:09:16 swordfish heartbeat: [2778]: info: killing HBFIFO process 2803 >with signal 15 >Jan 27 06:09:16 swordfish heartbeat: [2778]: info: killing HBWRITE process >2804 with signal 15 >Jan 27 06:09:16 swordfish heartbeat: [2778]: info: Core process 2803 exited. 3 >remaining >Jan 27 06:09:16 swordfish heartbeat: [2778]: info: Core process 2805 exited. 2 >remaining >Jan 27 06:09:16 swordfish heartbeat: [2778]: info: Core process 2804 exited. 1 >remaining >Jan 27 06:09:16 swordfish heartbeat: [2778]: info: swordfish.bmrb.wisc.edu >Heartbeat shutdown complete. --- -- Dimitri Maziuk Programmer/sysadmin BioMagResBank, UW-Madison -- http://www.bmrb.wisc.edu ___ Linux-HA mailing list Linux-HA@lists.linux-ha.org http://lists.linux-ha.org/mailman/listinfo/linux-ha See also: http://linux-ha.org/ReportingProblems
Re: [Linux-HA] strange heartbeat shutdown
On Friday 30 January 2009 04:22:27 Dejan Muhamedagic wrote: > > Looked through your logs: no idea what happened. It's as if > somebody typed rcheartbeat stop. Ah, that's the kick my brane needed -- mon would do that if it thought httpd/ftpd weren't running. Now to figure out why it'd think that. > What about the configuration? Pretty basic: bcast eth1, two nodes, auto-failback is on, and "host IP/26 proftpd httpd mon" in haresources. It's gotta be mon. Thanks for the push Dima -- Dimitri Maziuk Programmer/sysadmin BioMagResBank, UW-Madison -- http://www.bmrb.wisc.edu ___ Linux-HA mailing list Linux-HA@lists.linux-ha.org http://lists.linux-ha.org/mailman/listinfo/linux-ha See also: http://linux-ha.org/ReportingProblems
Re: [Linux-HA] strange heartbeat shutdown - SOLUTION
(Or at least I'm pretty sure that's what happened) If you tell mon to monitor, say, httpd on a hostname, make sure hostname resolves. In my case, I gave it the hostname of cluster ip and it wasn't in /etc/hosts. On top of that, servers are in a dmz subnet that has no local dns server. So when the gateway went down mon couldn't resolve the hostname and assumed httpd was dead. (The other mistake was a combination of brain fart and insufficient RTFM -- of course I wanted mon to run hb_standby, not shut heartbeat down. D'oh.) Dima -- Dimitri Maziuk Programmer/sysadmin BioMagResBank, UW-Madison -- http://www.bmrb.wisc.edu ___ Linux-HA mailing list Linux-HA@lists.linux-ha.org http://lists.linux-ha.org/mailman/listinfo/linux-ha See also: http://linux-ha.org/ReportingProblems
Re: [Linux-HA] Outgoing IP address
On Wednesday 04 February 2009 15:35:21 Lee Hobart wrote: > It's Unidata's LDM as far as I can tell I can specify a interface within > LDM. Then tell it to use eth0:0 or whatever it is on your system. Typically if there's more than one interface associated with the route to destination, the client will pick the first one -- for some value of "first". Most of the time "eth0" would come before "eth0:0" in the list, so that's the one it'll pick. It's probably possible to mark the packets with iptables and route them to eth0:0 with iproute (google for iptables mark and traffic shaping/policy routing), but option 1 is way simpler. Dima -- Dimitri Maziuk Programmer/sysadmin BioMagResBank, UW-Madison -- http://www.bmrb.wisc.edu ___ Linux-HA mailing list Linux-HA@lists.linux-ha.org http://lists.linux-ha.org/mailman/listinfo/linux-ha See also: http://linux-ha.org/ReportingProblems
Re: [Linux-HA] Outgoing IP address
James R. Leu wrote: For locally originated connections that do not bind to an interface you can use the SNAT target of iptables. iptables -A POSTROUTING -o eth0 -j SNAT --to-source 192.168.1.3 There's another problem with using cluster ip for outgoing address: if it fails over in the middle of a connection. The other node will start getting packets for a connection that doesn't exist on that node. First, the default iptables rules: "accept established" won't let them through since it's not "established" on the new node. If you get around that, client software isn't there or isn't in the state to process those packets. Dima ___ Linux-HA mailing list Linux-HA@lists.linux-ha.org http://lists.linux-ha.org/mailman/listinfo/linux-ha See also: http://linux-ha.org/ReportingProblems
Re: [Linux-HA] How to setup a 2 node active/passive apache2 cluster for Proof of Concept
Bernie Wu wrote: > The company I work for wants us to start investigating HA. > My first POC setup was a 2 node cluster with a floating IP and that worked > out quite well. > Now the second POC was to work with an application, in this case, apache2 in > a active/passive configuration. > My question is this. Do I need to setup a third node to serve as the quorum > node or can I work with 2 nodes. If your setup is v1-style active/passive, all you need to do is add httpd to haresources line (and make sure it's not started by init). That's for proof of concept. IRL you may want to throw in at least mon, possibly stonith -- although I haven't seen a split brain problem in my setup (but then again, mine usually fail over when I upgrade the kernel). Dima ___ Linux-HA mailing list Linux-HA@lists.linux-ha.org http://lists.linux-ha.org/mailman/listinfo/linux-ha See also: http://linux-ha.org/ReportingProblems
Re: [Linux-HA] How to setup a 2 node active/passive apache2 cluster for Proof of Concept
On Friday 29 May 2009 14:44:55 Bernie Wu wrote: > Hi Les, > Currently, there is no database or filesystem on the active system that > needs to be sync'ed. Apache currently just serves up static pages. It's > too early in our POC for something fancy. Syncing only really works if people upload files and expect them to persist past their current "session". With throw-away uploads (i.e. they click on upload and get back some output) it's not going to do anything, and with fancier stuff like data entry via multiple CGI forms you'll want DRBD. And if you have stateful stuff like applet-servlet apps, those will break during failover. Dima -- Dimitri Maziuk Programmer/sysadmin BioMagResBank, UW-Madison -- http://www.bmrb.wisc.edu ___ Linux-HA mailing list Linux-HA@lists.linux-ha.org http://lists.linux-ha.org/mailman/listinfo/linux-ha See also: http://linux-ha.org/ReportingProblems
Re: [Linux-HA] questions about using HA
blue_hmq wrote: > hi, i am sorry to disturb you because i have some questions about using HA > > i had configured the HA on two computer(HA01 as master,HA02),using apache2 > service. > if i stop apache2 service on HA01,but the Ethernet and heartbeat is Ok. as this situation ,will the ha02 take over the apache2 service ? if taken ,how does it do that(how does ha02 know the apache service on HA02 was stoped) ? > It doesn't, you'll need mon. Dima ___ Linux-HA mailing list Linux-HA@lists.linux-ha.org http://lists.linux-ha.org/mailman/listinfo/linux-ha See also: http://linux-ha.org/ReportingProblems
Re: [Linux-HA] Stonith with APCSmart UPS1000 +Network ManagementCard
Ehlers, Kolja wrote: > Could somebody explain how the APC Smart shutdown command works? Does it > actually allow to take away the power from only one of the > connected servers or does it just take away the power from everything that is > connected? IIRC we have the previous model of that network management card. It can shutdown the whole UPS, or it can send the powerfail signal to connected machines. I'm pretty sure ours can't shut off an individual outlet. You'll probably need their net-managed power strip for what you're trying to do. Dima ___ Linux-HA mailing list Linux-HA@lists.linux-ha.org http://lists.linux-ha.org/mailman/listinfo/linux-ha See also: http://linux-ha.org/ReportingProblems
Re: [Linux-HA] Stonith with APCSmart UPS1000 +Network ManagementCard
Ehlers, Kolja wrote: > Hello Dima, > > can you explain how you get that card to shutdown the whole UPS or how to > send the powerfail command to connected machines (are you > talking about the PowerChute agent?). Is any of that possible through a > stonith plugin? I never got a 'round tuit: we have pretty reliable power and all important stuff is on journaled raids. As I recall you can send powerfail either to a monitoring daemon (apcupsd?) or as snmp trap; with recent firmware you can also group upses to send the signal only when both of them lose power (with 2-psu machines connected to 2 upses on independent circuits). However, this is the opposite of what you want. Their networked power strip lets you turn individual outlets on and off -- dunno if you can do it via snmp, though. I'm sure you can wrap telnet/expect or http/curl script into a plugin if it comes to that. Dima ___ Linux-HA mailing list Linux-HA@lists.linux-ha.org http://lists.linux-ha.org/mailman/listinfo/linux-ha See also: http://linux-ha.org/ReportingProblems
Re: [Linux-HA] Raid and drdb
On Monday 27 July 2009 12:14:57 Claus Denk wrote: > Hi all, > > we have now almost configured our hardware for a HA cluster. We will use > two IBM 3550 M2 servers, and I would like to ask you if you could share > some experience about different disk configurations. > > 1) Using drdb, is having some hardware RAID on both machines good > practice (for example 3 disks with RAID 5) or is it not necessary to do > so as drdb is our network raid? I guess in case of failure of the drdb > communication it would be good to have some additional raid security on > each server. Does it affect the write performance a lot? Actually, I'd like to know: what happens when a disk fails? Would that propagate to DRBD and trigger a failover? > 2) This machine has a maximum of 4 SAS slots, what would be good > practice to install OS and data (2+2 disks with raid 1, or everything on > 3 disks raid 5, o 4 disks raid 10?) Of course disk price is an issue, we > would like to have 500 GB of data, which requires 3 300 Gb disks for raid > 5. With SAS drives you should be OK with raid 5 (SATA is where you'l want raid 1/10, apparently). Personally, I wouldn't worry too much about raid'ing the root drives: HA cluster should give you enough time to reinstall the OS if one fails. I would keep a spare disk handy. Dimitri -- Dimitri Maziuk Programmer/sysadmin BioMagResBank, UW-Madison -- http://www.bmrb.wisc.edu ___ Linux-HA mailing list Linux-HA@lists.linux-ha.org http://lists.linux-ha.org/mailman/listinfo/linux-ha See also: http://linux-ha.org/ReportingProblems
Re: [Linux-HA] Redundant MailMan
On Friday 11 September 2009 06:34:39 maike wrote: > Hello people, I wonder if anyone has implemented a list of Milaman > redundant with Heartbeat. I've been searching the list but have not found > anything, my network. > > correioa [postfix-mailman] -> smtpa > correiob [postfix-mailman] -> smtp > > if correioa or b stay out the mailman service is not compromised. If > someone can help me I am grateful. You should put /var/lib/mailman on drbd so that archives are shared between the 2 nodes. If you're going to let people create mailing lists (as opposed to creating a couple at install time and not adding any more ever), you should also put /etc/mail on drbd to share mailman aliases. I assume you'll be running httpd as well, you should make ServerName in apache config the same on both nodes: that of your cluster ip. Other than that I can't think of any gotchas. Dima -- Dimitri Maziuk Programmer/sysadmin BioMagResBank, UW-Madison -- http://www.bmrb.wisc.edu ___ Linux-HA mailing list Linux-HA@lists.linux-ha.org http://lists.linux-ha.org/mailman/listinfo/linux-ha See also: http://linux-ha.org/ReportingProblems
Re: [Linux-HA] postfix recover failed.
On Monday 09 November 2009 07:00:42 Dejan Muhamedagic wrote: > It would, but the stop action failed: > > Nov 9 11:07:12 Server1 lrmd: [4682]: info: RA output: > (Postfix:stop:stderr) 2009/11/09_11:07:12 ERROR: Postfix returned an error > while stopping. 1 > > This error comes from /usr/sbin/postfix. Not an expert on > postfix, so CC the author of the RA, perhaps he can take a look. Actually, this > > Failed actions: > > Postfix_monitor_5 (node=Server1, call=16, rc=7, > > status=complete): not running > > Postfix_stop_0 (node=Server1, call=17, rc=1, status=complete): > > unknown error looks like it's using a "restart" action, usually coded as "stop" followed by "start". If the daemon isn't running to begin with, "stop" will fail. Dima -- Dimitri Maziuk Programmer/sysadmin BioMagResBank, UW-Madison -- http://www.bmrb.wisc.edu ___ Linux-HA mailing list Linux-HA@lists.linux-ha.org http://lists.linux-ha.org/mailman/listinfo/linux-ha See also: http://linux-ha.org/ReportingProblems
Re: [Linux-HA] postfix recover failed.
On Tuesday 10 November 2009 03:57:25 Dejan Muhamedagic wrote: > Hi, > > The patch looks fine to me. Looks like /usr/sbin/postfix can't > handle this itself. Raoul: Is that actually expected? Why not: if "postfix stop" fails to stop postfix for whatever reason, it should return a non-zero value. Note that "stop" will wait for the daemons to shut down gracefully. Which means it could potentially block the failover for a while (unlikely worst case: deadlock). Dima -- Dimitri Maziuk Programmer/sysadmin BioMagResBank, UW-Madison -- http://www.bmrb.wisc.edu ___ Linux-HA mailing list Linux-HA@lists.linux-ha.org http://lists.linux-ha.org/mailman/listinfo/linux-ha See also: http://linux-ha.org/ReportingProblems