Unable to reactivate host after reboot due to failed Gluster probe
Hello, we have a strange behavior within an oVirt cluster. Version is 3.5.1, engine is running on EL6 machine and hosts are using EL7 as operating system. The cluster uses a GlusterFS backed storage domain amongst others. Three of four hosts are peers in the Gluster cluster (3 bricks, 3 replica). When all hosts are restarted (maybe due to power outage), engine can't activate them again, because Gluster probe fails. The message given in UI is: "Gluster command [gluster peer node-03] failed on server node-03." Checking Gluster peer and volume status on each host confirms that Gluster peers are known to each other and volume is up. node-03:~ $ gluster peer status Number of Peers: 2 Hostname: node-02 Uuid: 3fc36f55-d3a2-4efc-b2f0-31f83ed709d9 State: Peer in Cluster (Connected) Hostname: node-01 Uuid: 18027b35-971b-4b21-bb3d-df252b4dd525 State: Peer in Cluster (Connected) node-03:~ $ gluster volume status Status of volume: glusterfs-1 Gluster process Port Online Pid ------------------------------------------------------------------------------ Brick node-01:/export/glusterfs/brick 49152 Y 12409 Brick node-02:/export/glusterfs/brick 49153 Y 9978 Brick node-03:/export/glusterfs/brick 49152 Y 10001 Self-heal Daemon on localhost N/A Y 10003 Self-heal Daemon on node-01 N/A Y 11590 Self-heal Daemon on node-02 N/A Y 9988 Task Status of Volume glusterfs-1 ------------------------------------------------------------------------------ There are no active volume tasks Storage domain in oVirt UI is fine (active and green) and usable. But neither Gluster volume nor any brick is visible in UI. If I try the command which is shown in UI it returns: root@node-03:~ $ gluster peer probe node-03 peer probe: success. Probe on localhost not needed root@node-03:~ $ gluster --mode=script peer probe node-03 --xml <?xml version="1.0" encoding="UTF-8" standalone="yes"?> <cliOutput> <opRet>0</opRet> <opErrno>1</opErrno> <opErrstr>(null)</opErrstr> <output>Probe on localhost not needed</output> </cliOutput> Is this maybe just an engine side parsing error? -- Kind regards Jan Siml
Hello, when looking into engine.log, I can see, that "gluster probe" returned errno 107. But I can't figure out why: 2015-01-29 10:40:03,546 ERROR [org.ovirt.engine.core.bll.InitVdsOnUpCommand] (DefaultQuartzScheduler_Worker-59) [5977aac5] Could not peer probe the gluster server node-03. Error: VdcBLLException: org.ovirt.eng ine.core.vdsbroker.vdsbroker.VDSErrorException: VDSGenericException: VDSErrorException: Failed to AddGlusterServerVDS, error = Add host failed error: Probe returned with unknown errno 107 Just for the record: We use the /etc/hosts method because of missing possibility to choose the network interface for Gluster. The three Gluster peer hosts have modified /etc/hosts files with addresses binded to a different interface than the ovirtmgmt addresses. Example: root@node-03:~ $ cat /etc/hosts 192.168.200.195 node-01 192.168.200.196 node-02 192.168.200.198 node-03 The /etc/hosts file on engine host isn't modified. On 01/29/2015 10:39 AM, Jan Siml wrote:
Hello,
we have a strange behavior within an oVirt cluster. Version is 3.5.1, engine is running on EL6 machine and hosts are using EL7 as operating system. The cluster uses a GlusterFS backed storage domain amongst others. Three of four hosts are peers in the Gluster cluster (3 bricks, 3 replica).
When all hosts are restarted (maybe due to power outage), engine can't activate them again, because Gluster probe fails. The message given in UI is:
"Gluster command [gluster peer node-03] failed on server node-03."
Checking Gluster peer and volume status on each host confirms that Gluster peers are known to each other and volume is up.
node-03:~ $ gluster peer status Number of Peers: 2
Hostname: node-02 Uuid: 3fc36f55-d3a2-4efc-b2f0-31f83ed709d9 State: Peer in Cluster (Connected)
Hostname: node-01 Uuid: 18027b35-971b-4b21-bb3d-df252b4dd525 State: Peer in Cluster (Connected)
node-03:~ $ gluster volume status Status of volume: glusterfs-1 Gluster process Port Online Pid ------------------------------------------------------------------------------
Brick node-01:/export/glusterfs/brick 49152 Y 12409 Brick node-02:/export/glusterfs/brick 49153 Y 9978 Brick node-03:/export/glusterfs/brick 49152 Y 10001 Self-heal Daemon on localhost N/A Y 10003 Self-heal Daemon on node-01 N/A Y 11590 Self-heal Daemon on node-02 N/A Y 9988
Task Status of Volume glusterfs-1 ------------------------------------------------------------------------------
There are no active volume tasks
Storage domain in oVirt UI is fine (active and green) and usable. But neither Gluster volume nor any brick is visible in UI.
If I try the command which is shown in UI it returns:
root@node-03:~ $ gluster peer probe node-03 peer probe: success. Probe on localhost not needed
root@node-03:~ $ gluster --mode=script peer probe node-03 --xml <?xml version="1.0" encoding="UTF-8" standalone="yes"?> <cliOutput> <opRet>0</opRet> <opErrno>1</opErrno> <opErrstr>(null)</opErrstr> <output>Probe on localhost not needed</output> </cliOutput>
Is this maybe just an engine side parsing error?
-- Kind regards Jan Siml
Hello, finally I got the nodes online. What helps was probing the not needed peer node-04 (no brick) from one of the other cluster nodes. When the node becames a Gluster peer, I am able to activate any oVirt node which serves bricks. Therefore I assume, the error message which the UI returns comes from node-04: root@node-04:~ $ gluster peer probe node-01 peer probe: failed: Probe returned with unknown errno 107 root@node-03:~ $ gluster peer status Number of Peers: 2 Hostname: node-01 Uuid: 18027b35-971b-4b21-bb3d-df252b4dd525 State: Peer in Cluster (Connected) Hostname: node-02 Uuid: 3fc36f55-d3a2-4efc-b2f0-31f83ed709d9 State: Peer in Cluster (Connected) root@node-03:~ $ gluster peer probe node-04 peer probe: success. root@node-03:~ $ gluster peer status Number of Peers: 3 Hostname: node-01 Uuid: 18027b35-971b-4b21-bb3d-df252b4dd525 State: Peer in Cluster (Connected) Hostname: node-02 Uuid: 3fc36f55-d3a2-4efc-b2f0-31f83ed709d9 State: Peer in Cluster (Connected) Hostname: node-04 Uuid: 9cdefc68-d710-4346-93b1-76b5307e258b State: Peer in Cluster (Connected) This (oVirt's behavior) seems to be reproducible. On 01/29/2015 11:10 AM, Jan Siml wrote:
Hello,
when looking into engine.log, I can see, that "gluster probe" returned errno 107. But I can't figure out why:
2015-01-29 10:40:03,546 ERROR [org.ovirt.engine.core.bll.InitVdsOnUpCommand] (DefaultQuartzScheduler_Worker-59) [5977aac5] Could not peer probe the gluster server node-03. Error: VdcBLLException: org.ovirt.eng ine.core.vdsbroker.vdsbroker.VDSErrorException: VDSGenericException: VDSErrorException: Failed to AddGlusterServerVDS, error = Add host failed error: Probe returned with unknown errno 107
Just for the record: We use the /etc/hosts method because of missing possibility to choose the network interface for Gluster. The three Gluster peer hosts have modified /etc/hosts files with addresses binded to a different interface than the ovirtmgmt addresses.
Example:
root@node-03:~ $ cat /etc/hosts 192.168.200.195 node-01 192.168.200.196 node-02 192.168.200.198 node-03
The /etc/hosts file on engine host isn't modified.
On 01/29/2015 10:39 AM, Jan Siml wrote:
Hello,
we have a strange behavior within an oVirt cluster. Version is 3.5.1, engine is running on EL6 machine and hosts are using EL7 as operating system. The cluster uses a GlusterFS backed storage domain amongst others. Three of four hosts are peers in the Gluster cluster (3 bricks, 3 replica).
When all hosts are restarted (maybe due to power outage), engine can't activate them again, because Gluster probe fails. The message given in UI is:
"Gluster command [gluster peer node-03] failed on server node-03."
Checking Gluster peer and volume status on each host confirms that Gluster peers are known to each other and volume is up.
node-03:~ $ gluster peer status Number of Peers: 2
Hostname: node-02 Uuid: 3fc36f55-d3a2-4efc-b2f0-31f83ed709d9 State: Peer in Cluster (Connected)
Hostname: node-01 Uuid: 18027b35-971b-4b21-bb3d-df252b4dd525 State: Peer in Cluster (Connected)
node-03:~ $ gluster volume status Status of volume: glusterfs-1 Gluster process Port Online Pid ------------------------------------------------------------------------------
Brick node-01:/export/glusterfs/brick 49152 Y 12409 Brick node-02:/export/glusterfs/brick 49153 Y 9978 Brick node-03:/export/glusterfs/brick 49152 Y 10001 Self-heal Daemon on localhost N/A Y 10003 Self-heal Daemon on node-01 N/A Y 11590 Self-heal Daemon on node-02 N/A Y 9988
Task Status of Volume glusterfs-1 ------------------------------------------------------------------------------
There are no active volume tasks
Storage domain in oVirt UI is fine (active and green) and usable. But neither Gluster volume nor any brick is visible in UI.
If I try the command which is shown in UI it returns:
root@node-03:~ $ gluster peer probe node-03 peer probe: success. Probe on localhost not needed
root@node-03:~ $ gluster --mode=script peer probe node-03 --xml <?xml version="1.0" encoding="UTF-8" standalone="yes"?> <cliOutput> <opRet>0</opRet> <opErrno>1</opErrno> <opErrstr>(null)</opErrstr> <output>Probe on localhost not needed</output> </cliOutput>
Is this maybe just an engine side parsing error?
-- Kind regards Jan Siml
On 01/29/2015 04:26 PM, Jan Siml wrote:
Hello,
finally I got the nodes online. What helps was probing the not needed peer node-04 (no brick) from one of the other cluster nodes. When the node becames a Gluster peer, I am able to activate any oVirt node which serves bricks.
Therefore I assume, the error message which the UI returns comes from node-04:
Yes, this could be an issue as all other successful cases, the value for opErrno is retruned as 0 and opErrStr is blank. I feel this scenario is treated as an error engine side.
root@node-04:~ $ gluster peer probe node-01 peer probe: failed: Probe returned with unknown errno 107
root@node-03:~ $ gluster peer status Number of Peers: 2
Hostname: node-01 Uuid: 18027b35-971b-4b21-bb3d-df252b4dd525 State: Peer in Cluster (Connected)
Hostname: node-02 Uuid: 3fc36f55-d3a2-4efc-b2f0-31f83ed709d9 State: Peer in Cluster (Connected)
root@node-03:~ $ gluster peer probe node-04 peer probe: success.
root@node-03:~ $ gluster peer status Number of Peers: 3
Hostname: node-01 Uuid: 18027b35-971b-4b21-bb3d-df252b4dd525 State: Peer in Cluster (Connected)
Hostname: node-02 Uuid: 3fc36f55-d3a2-4efc-b2f0-31f83ed709d9 State: Peer in Cluster (Connected)
Hostname: node-04 Uuid: 9cdefc68-d710-4346-93b1-76b5307e258b State: Peer in Cluster (Connected)
This (oVirt's behavior) seems to be reproducible.
On 01/29/2015 11:10 AM, Jan Siml wrote:
Hello,
when looking into engine.log, I can see, that "gluster probe" returned errno 107. But I can't figure out why:
2015-01-29 10:40:03,546 ERROR [org.ovirt.engine.core.bll.InitVdsOnUpCommand] (DefaultQuartzScheduler_Worker-59) [5977aac5] Could not peer probe the gluster server node-03. Error: VdcBLLException: org.ovirt.eng ine.core.vdsbroker.vdsbroker.VDSErrorException: VDSGenericException: VDSErrorException: Failed to AddGlusterServerVDS, error = Add host failed error: Probe returned with unknown errno 107
Just for the record: We use the /etc/hosts method because of missing possibility to choose the network interface for Gluster. The three Gluster peer hosts have modified /etc/hosts files with addresses binded to a different interface than the ovirtmgmt addresses.
Example:
root@node-03:~ $ cat /etc/hosts 192.168.200.195 node-01 192.168.200.196 node-02 192.168.200.198 node-03
The /etc/hosts file on engine host isn't modified.
On 01/29/2015 10:39 AM, Jan Siml wrote:
Hello,
we have a strange behavior within an oVirt cluster. Version is 3.5.1, engine is running on EL6 machine and hosts are using EL7 as operating system. The cluster uses a GlusterFS backed storage domain amongst others. Three of four hosts are peers in the Gluster cluster (3 bricks, 3 replica).
When all hosts are restarted (maybe due to power outage), engine can't activate them again, because Gluster probe fails. The message given in UI is:
"Gluster command [gluster peer node-03] failed on server node-03."
Checking Gluster peer and volume status on each host confirms that Gluster peers are known to each other and volume is up.
node-03:~ $ gluster peer status Number of Peers: 2
Hostname: node-02 Uuid: 3fc36f55-d3a2-4efc-b2f0-31f83ed709d9 State: Peer in Cluster (Connected)
Hostname: node-01 Uuid: 18027b35-971b-4b21-bb3d-df252b4dd525 State: Peer in Cluster (Connected)
node-03:~ $ gluster volume status Status of volume: glusterfs-1 Gluster process Port Online Pid ------------------------------------------------------------------------------
Brick node-01:/export/glusterfs/brick 49152 Y 12409 Brick node-02:/export/glusterfs/brick 49153 Y 9978 Brick node-03:/export/glusterfs/brick 49152 Y 10001 Self-heal Daemon on localhost N/A Y 10003 Self-heal Daemon on node-01 N/A Y 11590 Self-heal Daemon on node-02 N/A Y 9988
Task Status of Volume glusterfs-1 ------------------------------------------------------------------------------
There are no active volume tasks
Storage domain in oVirt UI is fine (active and green) and usable. But neither Gluster volume nor any brick is visible in UI.
If I try the command which is shown in UI it returns:
root@node-03:~ $ gluster peer probe node-03 peer probe: success. Probe on localhost not needed
root@node-03:~ $ gluster --mode=script peer probe node-03 --xml <?xml version="1.0" encoding="UTF-8" standalone="yes"?> <cliOutput> <opRet>0</opRet> <opErrno>1</opErrno> <opErrstr>(null)</opErrstr> <output>Probe on localhost not needed</output> </cliOutput>
Is this maybe just an engine side parsing error?
participants (2)
-
Jan Siml -
Shubhendu Tripathi