| Summary: | sys-cluster/heartbeat does not correctly deactivate 802.1q interfaces on failover | ||
|---|---|---|---|
| Product: | Gentoo Linux | Reporter: | Alun Jones <auj> |
| Component: | [OLD] Server | Assignee: | Gentoo Cluster Team <cluster> |
| Status: | RESOLVED WONTFIX | ||
| Severity: | normal | CC: | loki_val |
| Priority: | High | ||
| Version: | unspecified | ||
| Hardware: | All | ||
| OS: | Linux | ||
| Whiteboard: | |||
| Package list: | Runtime testing required: | --- | |
Debian already has a bug and proposed workaround: http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=344633 -------- > I'd suggest switching the default ip script to IPaddr2 on linux platforms > which uses ip and doesn't have this limitation. Yes, please specify IPaddr2 in your haresources file instead of IPaddr. -------- Does this work for you? Yes, works fine. I did look at IPaddr2 - just not carefully enough. Thanks! + 18 Oct 2011; Kacper Kowalik <xarthisius@gentoo.org> -heartbeat-1.2.5.ebuild, + -heartbeat-1.2.5-r1.ebuild, -heartbeat-2.0.7-r2.ebuild, + -files/heartbeat-2.0.7-update-resources-failcount.patch, + -heartbeat-2.0.8.ebuild, -files/heartbeat-2.0.8-asneeded.patch, + -files/heartbeat-2.0.8-crm-assert.patch, + -files/heartbeat-2.0.8-crm-leaks.patch, -files/heartbeat-2.0.8-delay.patch, + -files/heartbeat-2.0.8-glibc.patch, -files/heartbeat-2.0.8-install_fix.patch, + -heartbeat-3.0.4-r1.ebuild: + Drop ancient and unusable versions of heartbeat-{1,2} since we have + heartbeat-3 stable |
We have two servers acting as an LVS cluster with heartbeat. They are running on 802.1q VLANs, and this means that the interface names for virtual IP addresses end up in the form eth0.116:0 However, ifconfig has a fixed field width and chops off the final character. This means that when heartbeat fails over, it doesn't correctly bring down the virtual IP off the original server. We've done some detective work and the problem lies in /usr/lib/ocf/heartbeat/IPaddr, in the function find_interface_generic, which parses the output of ifconfig to find the name of the interface. We've hacked around this for now by installing iproute2 and writing a simple find_interface_iproute2 function which is as follows: find_interface_iproute2() { ipaddr="$1" /sbin/ip addr show | /bin/fgrep "inet $1/" | /usr/bin/awk '{ print $NF; }' } We call this from find_interface() if we detect we're running Linux. This is a bit of a hack at the moment, as it assumes the existence of iproute2. Would appreciate feedback on this, as we intend to use it in front of a fairly large part of our infrastructure here. I would have reported this upstream, but their Bugzilla is in a dodgy state at the moment...