<html>
<head>
<meta http-equiv="Content-Type" content="text/html; charset=utf-8">
</head>
<body text="#000000" bgcolor="#FFFFFF">
<p>Many thanks for the excellent answer ,</p>
<p>Ken after investigation of the log files :</p>
<p>In our environment we have two drbd partitions one for
customer_vms and on for sigma_vms <br>
</p>
<p>For the customer_vms the active node is node2 and for the
sigma_vms the active node is node1 .</p>
<p>[root@sgw-01 drbd.d]# drbdadm status<br>
</p>
<p>customer_vms role:Secondary<br>
disk:UpToDate<br>
sgw-02 role:Primary<br>
peer-disk:UpToDate<br>
<br>
sigma_vms role:Primary<br>
disk:UpToDate<br>
sgw-02 role:Secondary<br>
peer-disk:UpToDate<br>
<br>
</p>
<p>when i create a new VM <b>i can't force the resource creation</b>
to take place on a specific node , the cluster places the resource
<br>
</p>
<p>spontaneously on one of the two nodes (if the node happens to be
the drbd Primary then is ok, else the pacemaker raise a failure
fro the node) .<br>
</p>
<p>My solution is the following :</p>
<p>pcs resource create windows_VM_res VirtualDomain
hypervisor="qemu:///system"
config="/opt/sigma_vms/xml_definitions/windows_VM.xml" <br>
</p>
<p>(the cluster arbitrarily try to place the above resource on node
2 who is currently the secondary for the corresponding partition.
Personally <br>
</p>
<p>i assume that the VirtualDomain agent should be able to read the
correct disk location from the xml defintion and then try to find
the correct drbd node) <br>
</p>
<p>pcs constraint colocation add windows_VM_res with
StorageDRBD_SigmaVMs INFINITY<br>
<br>
pcs constraint order start StorageDRBD_SigmaVMs_rers then start
windows_VM<br>
<br>
pcs resource cleanup windows_VM_res</p>
<p>After the above steps the VM is located on the correct node and
everything is ok.<br>
</p>
<p><br>
</p>
<p><b>Is my approach correct ?</b><br>
</p>
<p><br>
</p>
<p>Your opinion would be valuable,<br>
</p>
<p>Sincerely <br>
</p>
<p><br>
</p>
<br>
<div class="moz-cite-prefix">On 06/25/2018 07:15 PM, Ken Gaillot
wrote:<br>
</div>
<blockquote type="cite"
cite="mid:1529943316.5338.3.camel@redhat.com">
<pre wrap="">On Mon, 2018-06-25 at 09:47 -0500, Ken Gaillot wrote:
</pre>
<blockquote type="cite">
<pre wrap="">On Mon, 2018-06-25 at 11:33 +0300, Vaggelis Papastavros wrote:
</pre>
<blockquote type="cite">
<pre wrap="">Dear friends ,
We have the following configuration :
CentOS7 , pacemaker 0.9.152 and Corosync 2.4.0, storage with DRBD
and
stonith eanbled with APC PDU devices.
I have a windows VM configured as cluster resource with the
following
attributes :
Resource: WindowSentinelOne_res (class=ocf provider=heartbeat
type=VirtualDomain)
Attributes: hypervisor=qemu:///system
config=/opt/customer_vms/conf/WindowSentinelOne/WindowSentinelOne.x
ml
migration_transport=ssh
Utilization: cpu=8 hv_memory=8192
Operations: start interval=0s timeout=120s
(WindowSentinelOne_res-start-interval-0s)
stop interval=0s timeout=120s
(WindowSentinelOne_res-stop-interval-0s)
monitor interval=10s timeout=30s
(WindowSentinelOne_res-monitor-interval-10s)
under some circumstances (which i try to identify) the VM fails
and
disappears under virsh list --all and also pacemaker reports the VM
as
stopped .
If run pcs resource cleanup windows_wm everything is OK, but i
can't
identify the reason of failure.
For example when shutdown the VM (with windows shutdown) the
cluster
reports the following :
WindowSentinelOne_res (ocf::heartbeat:VirtualDomain): Started
sgw-
02
(failure ignored)
Failed Actions:
* WindowSentinelOne_res_monitor_10000 on sgw-02 'not running' (7):
call=67, status=complete, exitreason='none',
last-rc-change='Mon Jun 25 07:41:37 2018', queued=0ms,
exec=0ms.
My questions are
1) why the VM shutdown is reported as (FailedAction) from cluster ?
Its
a worthy operation during VM life cycle .
</pre>
</blockquote>
<pre wrap="">
Pacemaker has no way of knowing that the VM was intentionally shut
down, vs crashed.
When some resource is managed by the cluster, all starts and stops of
the resource have to go through the cluster. You can either set
target-
role=Stopped in the resource configuration, or if it's a temporary
issue (e.g. rebooting for some OS updates), you could set is-
managed=false to take it out of cluster control, do the work, then
set
is-managed=true again.
</pre>
</blockquote>
<pre wrap="">
Also, a nice feature is that you can use rules to set a maintenance
window ahead of time (especially helpful if the person who maintains
the cluster isn't the same person who needs to do the VM updates). For
example, you could set a rule that the resource's is-managed option
will be false from 9pm to midnight on Fridays. See:
<a class="moz-txt-link-freetext" href="http://clusterlabs.org/pacemaker/doc/en-US/Pacemaker/1.1/html-single/Pa">http://clusterlabs.org/pacemaker/doc/en-US/Pacemaker/1.1/html-single/Pa</a>
cemaker_Explained/index.html#idm140583511697312
particularly the parts about time/date expressions and using rules to
control resource options.
</pre>
<blockquote type="cite">
<pre wrap="">
</pre>
<blockquote type="cite">
<pre wrap="">2) why sometimes the resource is marked as stopped (the VM is
healthy)
and needs cleanup ?
</pre>
</blockquote>
<pre wrap="">
That's a problem. If the VM is truly healthy, it sounds like there's
an
issue with the resource agent. You'd have to look at the logs to see
if
it gave any more information (e.g. if it's a timeout, raising the
timeout might be sufficient).
</pre>
<blockquote type="cite">
<pre wrap="">3) I can't understand the corosync logs ... during the the VM
shutdown
corosync logs is the following
</pre>
</blockquote>
<pre wrap="">
FYI, the system log will have the most important messages.
corosync.log
will additionally have info-level messages -- potentially helpful but
definitely difficult to follow.
</pre>
<blockquote type="cite">
<pre wrap="">Jun 25 07:41:37 [5140] sgw-02 crmd: info:
process_lrm_event: Result of monitor operation for
WindowSentinelOne_res on sgw-02: 7 (not running) | call=67
key=WindowSentinelOne_res_monitor_10000 confirmed=false cib-
update=36
</pre>
</blockquote>
<pre wrap="">
This is really the only important message. It says that a recurring
monitor on the WindowSentinelOne_res resource on node sgw-02 exited
with status code 7 (which means the resource agent thinks the
resource
is not running).
'key=WindowSentinelOne_res_monitor_10000' is how pacemaker identifies
resource agent actions. The format is <resource-name>_<action-
name>_<action-interval-in-milliseconds>
This is the only information Pacemaker will get from the resource
agent. To investigate more deeply, you'll have to check for log
messages from the agent itself.
</pre>
<blockquote type="cite">
<pre wrap="">Jun 25 07:41:37 [5130] sgw-02 cib: info:
cib_process_request: Forwarding cib_modify operation for
section
status to all (origin=local/crmd/36)
Jun 25 07:41:37 [5130] sgw-02 cib: info:
cib_perform_op:
Diff: --- 0.4704.67 2
Jun 25 07:41:37 [5130] sgw-02 cib: info:
cib_perform_op:
Diff: +++ 0.4704.68 (null)
Jun 25 07:41:37 [5130] sgw-02 cib: info:
cib_perform_op:
+ /cib: @num_updates=68
Jun 25 07:41:37 [5130] sgw-02 cib: info:
cib_perform_op:
+ /cib/status/node_state[@id='2']: @crm-debug-
origin=do_update_resource
Jun 25 07:41:37 [5130] sgw-02 cib: info:
cib_perform_op:
++
/cib/status/node_state[@id='2']/lrm[@id='2']/lrm_resources/lrm_reso
ur
ce[@id='WindowSentinelOne_res']:
<lrm_rsc_op id="WindowSentinelOne_res_last_failure_0"
operation_key="WindowSentinelOne_res_monitor_10000"
operation="monitor"
crm-debug-origin="do_update_resource" crm_feature_set="3.0.10"
transition-key="84:3:0:f910c793-a714-4e24-80d1-b0ec66275491"
transition-magic="0:7;84:3:0:f910c793-a714-4e24-80d1-b0ec66275491"
on_node="sgw-02" cal
Jun 25 07:41:37 [5130] sgw-02 cib: info:
cib_process_request: Completed cib_modify operation for section
status: OK (rc=0, origin=sgw-02/crmd/36, version=0.4704.68)
</pre>
</blockquote>
<pre wrap="">
You can usually ignore the 'cib' messages. This just means Pacemaker
recorded the result on disk.
</pre>
<blockquote type="cite">
<pre wrap="">Jun 25 07:41:37 [5137] sgw-02 attrd: info:
attrd_peer_update: Setting fail-count-WindowSentinelOne_res[sgw-
02]:
(null) -> 1 from sgw-01
</pre>
</blockquote>
<pre wrap="">
Since the cluster expected the resource to be running, this result is
a
failure. Failures are counted using special node attributes that
start
with "fail-count-". This is what Pacemaker uses to determine if a
resource has reached its migration-threshold.
</pre>
<blockquote type="cite">
<pre wrap="">Jun 25 07:41:37 [5137] sgw-02 attrd: info:
write_attribute:
Sent update 10 with 1 changes for fail-count-
WindowSentinelOne_res,
id=<n/a>, set=(null)
Jun 25 07:41:37 [5130] sgw-02 cib: info:
cib_process_request: Forwarding cib_modify operation for
section
status to all (origin=local/attrd/10)
Jun 25 07:41:37 [5137] sgw-02 attrd: info:
attrd_peer_update: Setting
last-failure-WindowSentinelOne_res[sgw-02]: (null) -> 1529912497
from
sgw-01
</pre>
</blockquote>
<pre wrap="">
Similarly, the time the failure occurred is stored in a 'last-
failure-'
node attribute, which Pacemaker uses to determine if a resource has
reached its failure-timeout.
</pre>
<blockquote type="cite">
<pre wrap="">Jun 25 07:41:37 [5137] sgw-02 attrd: info:
write_attribute:
Sent update 11 with 1 changes for last-failure-
WindowSentinelOne_res,
id=<n/a>, set=(null)
Jun 25 07:41:37 [5130] sgw-02 cib: info:
cib_process_request: Forwarding cib_modify operation for
section
status to all (origin=local/attrd/11)
Jun 25 07:41:37 [5130] sgw-02 cib: info:
cib_perform_op:
Diff: --- 0.4704.68 2
Jun 25 07:41:37 [5130] sgw-02 cib: info:
cib_perform_op:
Diff: +++ 0.4704.69 (null)
Jun 25 07:41:37 [5130] sgw-02 cib: info:
cib_perform_op:
+ /cib: @num_updates=69
Jun 25 07:41:37 [5130] sgw-02 cib: info:
cib_perform_op:
++
/cib/status/node_state[@id='2']/transient_attributes[@id='2']/insta
nc
e_attributes[@id='status-2']:
<nvpair id="status-2-fail-count-WindowSentinelOne_res"
name="fail-count-WindowSentinelOne_res" value="1"/>
Jun 25 07:41:37 [5130] sgw-02 cib: info:
cib_process_request: Completed cib_modify operation for section
status: OK (rc=0, origin=sgw-02/attrd/10, version=0.4704.69)
Jun 25 07:41:37 [5137] sgw-02 attrd: info:
attrd_cib_callback: Update 10 for fail-count-
WindowSentinelOne_res:
OK (0)
Jun 25 07:41:37 [5137] sgw-02 attrd: info:
attrd_cib_callback: Update 10 for
fail-count-WindowSentinelOne_res[sgw-02]=1: OK (0)
Jun 25 07:41:37 [5130] sgw-02 cib: info:
cib_perform_op:
Diff: --- 0.4704.69 2
Jun 25 07:41:37 [5130] sgw-02 cib: info:
cib_perform_op:
Diff: +++ 0.4704.70 (null)
Jun 25 07:41:37 [5130] sgw-02 cib: info:
cib_perform_op:
+ /cib: @num_updates=70
Jun 25 07:41:37 [5130] sgw-02 cib: info:
cib_perform_op:
++
/cib/status/node_state[@id='2']/transient_attributes[@id='2']/insta
nc
e_attributes[@id='status-2']:
<nvpair id="status-2-last-failure-WindowSentinelOne_res"
name="last-failure-WindowSentinelOne_res" value="1529912497"/>
Jun 25 07:41:37 [5130] sgw-02 cib: info:
cib_process_request: Completed cib_modify operation for section
status: OK (rc=0, origin=sgw-02/attrd/11, version=0.4704.70)
Jun 25 07:41:37 [5137] sgw-02 attrd: info:
attrd_cib_callback: Update 11 for last-failure-
WindowSentinelOne_res:
OK (0)
Jun 25 07:41:37 [5137] sgw-02 attrd: info:
attrd_cib_callback: Update 11 for
last-failure-WindowSentinelOne_res[sgw-02]=1529912497: OK (0)
Jun 25 07:41:42 [5130] sgw-02 cib: info:
cib_process_ping:
Reporting our current digest to sgw-01:
3e27415fcb003ef3373b47ffa6c5f358
for 0.4704.70 (0x7faac1729720 0)
Sincerely ,
Vaggelis Papastavros
</pre>
</blockquote>
</blockquote>
</blockquote>
<br>
</body>
</html>