If the cluster interconnect fails or the node panics, following events occur.
One of the nodes becomes LEFTCLUSTER state.
The I/O error occurs in the slices that configure the netmirror volume.
During this time, the following will be the state until the LEFTCLUSTER state is canceled.
I/O from the application to the netmirror volume can be continued.
However, during synchronization or resynchronization copying, I/O to the area where copying process is running is suspended and the copying process does not progress.
The slice where the I/O error has occurred will be in the following state:
It does not change to INVALID state but remains in ACTIVE, STOP or COPY state.
It becomes pre-detached state.
Pre-detached state is the state in which the message 44005 (WARNING: object.volume: detached status slice by an I/O error, class=class) is not output on any node after the warning message 22023 (WARNING: sfdsk: slice is pre-detached by an I/O error: ...) is output.
The pre-detached state cannot be checked in the sdxinfo command or the GDS Management View.
I/O from the application to the net mirror volume is not issued to the slice where the I/O error occurred, but is issued only to another slice.
Therefore, the latest data is not stored in the slice where the I/O error occurred.
When the LEFTCLUSTER state is canceled, the slice in which the I/O error occurred is detached and becomes the INVALID state.
Also, during synchronization or resynchronization copying, the suspended I/O returns after the copy destination slice is detached.
The LEFTCLUSTER state may be automatically canceled or not.
When the LEFTCLUSTER state is automatically canceled
When the LEFTCLUSTER state is automatically canceled by the shutdown facility (SF) of PRIMECLUSTER, the maximum value of the time from the occurrence of an abnormality until the slice is detached is as follows.
[Maximum value of the time until the slice is detached (default)]
PRIMERGY: 120 seconds
PRIMEQUEST 3000 Series Business Model: 100 seconds
PRIMEQUEST 3000 Series Enterprise Model: 140 seconds
PRIMEQUEST 2000 Series: 100 seconds
KVM guest: 160 seconds
VMware guest: 110 seconds
FJcloud-O environment: 270 seconds
AWS environment: 50 seconds
Azure environment: 110 seconds
NIFCLOUD environment: 70 seconds
RHOSP environment: 270 seconds
If [Timeout value of PRIMECLUSTER] below is changed from the default value, the maximum value of the time from the occurrence of an abnormality until the slice is detached can be calculated according to [Maximum value of the time until the slice is detached].
[Timeout value of PRIMECLUSTER]
Timeout detection time of heartbeat of CF (CLUSTER_TIMEOUT)
Timeout value of shutdown facility
[Maximum value of the time until the slice is detached]
<Maximum value of the time until the slice is detached (default) >
+ <The value of CLUSTER_TIMEOUT after removing the default value>
+ <Number of nodes> x <Timeout value of shutdown facility after removing the default value
"Timeout value of shutdown facility" is the total value of timeout values of all the shutdown agents.
When the LEFTCLUSTER state is not automatically canceled
The PRIMECLUSTER Shutdown Facility (SF) fails to shut down the node and the slice where the I/O error occurred is not detached without automatically canceling the LEFTCLUSTER state.
In this case, the LEFTCLUSTER state must be cleared manually.
For how to clear the LEFTCLUSTER state, see "7.16.4 Restoration from the LEFTCLUSTER State."
