Host-HA offers investigation, fencing and recovery mechanisms for host that for
any reason are malfunctioning. It uses Activity and Health checks to determine
current host state based on which it may degrade a host or try to recover it. On
failing to recover it, it may try to fence the host.
The core feature is implemented in a hypervisor agnostic way, with two separate
implementations of the driver/provider for Simulator and KVM hypervisors. The
framework also allows for implementation of other hypervisor specific provider
implementation in future.
The Host-HA provider implementation for KVM hypervisor uses the out-of-band
management sub-system to issue IPMI calls to reset (recover) or poweroff (fence)
a host.
The Host-HA provider implementation for Simulator provides a means of testing
and validating the core framework implementation.
Signed-off-by: Abhinandan Prateek <abhinandan.prateek@shapeblue.com>
Signed-off-by: Rohit Yadav <rohit.yadav@shapeblue.com>
This fixes the agreed upon url on download.cloudstack.org in various
sql files and misc scripts.
Signed-off-by: Rohit Yadav <rohit.yadav@shapeblue.com>
more detailed error if host file not found or cannot be opened
using mkstemp and mkdtemp for improved security
improve resource cleanup in error conditions in unit test
As requested here: https://github.com/apache/cloudstack/pull/1495
No scripts are using perl so that install requirement can be removed.
The new scripts are using standard python packages only.
Includes extensive unit test.
This setting works on CentOS 6 / RHEL 6 but does nothing, as
"cpu" cgroup is not mounted. On CentOS 7 / RHEL 7 systemd does
mount cgroups and "cpu" is co-mounted with "cpuacc". Hence, if
we specify "cpu" then this results in an error because it can
only use them both, or none.
By removing the setting, we rely on the default of qemu, which
is:
cgroup_controllers = ["cpu", "devices", "memory", "blkio", "cpuacct", "net_cls"]
Only if they are really mounted, they will be used. So, this will
work on both version 6 and 7.
The 'fix script' didn't work well, as after a reboot you'd still have qemu
throwing errors. Now we can handle the co-mountedcgroups.
The recent discussed improvement has the risk that if 'sync' hangs, the reboot may be delayed in the same way as the 'reboot' command would do. To work around, we're adding a 5 second timeout. If it cannot sync in 5 seconds, it will not succeed anyway and we should proceed the reset.
@snuf: Could we use your OVM3 heartbeat script for other hypervisors as well? One way to do it seems like a nice idea :-)
As discussed with @wido @pyr and @nuxro added an extra log line.
Tested it and it logs fine (tested to local disk) when syncing first:
Apr 3 15:31:23 mcctest2 heartbeat: kvmheartbeat.sh system because it was unable to write the heartbeat to the storage
By the way, it did also log to the agent.log but this extra log has the benefit of ending up in the system log so you'll probably find it easier there. Existing logs:
2015-04-03 15:27:23,943 WARN [kvm.resource.KVMHAMonitor] (Thread-24:null) write heartbeat failed: timeout, retry: 0
2015-04-03 15:28:23,944 WARN [kvm.resource.KVMHAMonitor] (Thread-24:null) write heartbeat failed: timeout, retry: 1
2015-04-03 15:29:23,946 WARN [kvm.resource.KVMHAMonitor] (Thread-24:null) write heartbeat failed: timeout, retry: 2
2015-04-03 15:30:23,948 WARN [kvm.resource.KVMHAMonitor] (Thread-24:null) write heartbeat failed: timeout, retry: 3
2015-04-03 15:31:23,950 WARN [kvm.resource.KVMHAMonitor] (Thread-24:null) write heartbeat failed: timeout, retry: 4
2015-04-03 15:31:23,950 WARN [kvm.resource.KVMHAMonitor] (Thread-24:null) write heartbeat failed: timeout; reboot the host
This closes#145
Signed-off-by: Rohit Yadav <rohit.yadav@shapeblue.com>
When storage cannot be reached, it does not make sense to reboot as it will try to flush buffers, umount NFS mounts, etc. This will not work and thus cause a long delay. With this change, the box will reboot immediately (like pressing the reset button).
Detail: This gets rid of the patchdisk method of passing cmdline and
authorized_keys to KVM system VMs. It instead passes them to a virtio socket,
which the KVM guest reads from the character device /dev/vport0p1 during
cloud-early-config. Tested to work on CentOS 6.3 and Ubuntu 12.04. Should
work with even older versions of libvirt.
Signed-off-by: Marcus Sorensen <marcus@betterservers.com> 1362691685 -0700
host if there are multiple storage pools in a cluster.
The issue is as follows:
1. When CloudStack detects that a host is not responding to ping
requests it'll send a fence command for this host to another host in the
cluster.
2. The agent takes a long time to respond to this check if the storage
is fenced. This is because the agent checks if the first host is writing
to its heartbeat file on all pools in the cluster. It is doing this in a
sequential manner on all storage pool.
Making a fix to get rid of sleep, wait during HA. The behavior is now
similar to Xenserver.
RB: https://reviews.apache.org/r/6133/
Send-by:devdeep.singh@citrix.com
The sequence:
1. add host in UI
2. scp setup_agent.sh to agent host, and execute it. This script receives hostip,zoneid, podid and guid, then runs "cloud-setup-agent" and "cloud-setup-console-proxy". Here, we assume that network/hostname and cloud-agent are already configed and installed.
3. Write a dummy kvm resource into the database, then wait for agent connects to server, by polling the database for every 1 minutes. If it finds the agent is in UP state in database, then return, or wait for at least 10 minutes.
