Linux – HP EVA P6000 Storage User Manual
Page 166
Failure effect
Fault stimulus
to a different served path. When the single-path node
crashes, only the processes executing on that node fail.
In either case, no data is lost or corrupted.
I/O is suspended or process is terminated across this HBA
until switch is back online. No data is lost or corrupted.
The operating system will report the volume in a Mount
Verify state until the MVTIMEOUT limit is exceeded, when
Switch failure (SAN switch disabled)
it then marks the volume as Mount Verify Timeout. No data
is lost or corrupted.
I/O fails over to the surviving controller. No data is lost or
corrupted.
Controller failure
I/O is suspended or process is terminated across this HBA
until EVA is back online. No data is lost or corrupted.
The operating system will report the volume in a Mount
Verify state until the MVTIMEOUT limit is exceeded, when
it then marks the volume as Mount Verify Timeout.
Controller restart
If the LUN is not shared, I/O is suspended or process is
terminated across this HBA until path is restored.
If running OpenVMS 7.3-1 and the LUN is shared, another
cluster node having direct access will take over serving the
device, resulting in no loss of service.
Server path failure
In either case, no data is lost or corrupted.
The operating system will report the volume in a Mount
Verify state until the MVTIMEOUT limit is exceeded, when
it then marks the volume as Mount Verify Timeout.
I/O is suspended or process is terminated across this HBA
until path is restored. No data is lost or corrupted.
The operating system will report the volume in a Mount
Verify state until the MVTIMEOUT limit is exceeded, when
it then marks the volume as Mount Verify Timeout.
Storage path failure
Linux
Failure effect
Fault stimulus
OS reboots, automatically checks disks. HSV disks must be manually
checked unless auto mounted by the system.
Server failure (host power-cycled)
Short: I/O suspended, possible data loss.
Long: I/O halts with I/O errors, data loss. HBA driver must be reloaded
before failed drives can be recovered, fsck should be run on any failed
drives before remounting.
Switch failure (SAN switch disabled)
Short term: I/O suspended, possible data loss.
Long term: I/O halts with I/O errors, data loss. Cannot reload driver,
need to reboot system, fsck should be run on any failed disks before
remounting.
Controller failure
Short term: I/O suspended, possible data loss.
Long term: I/O halts with I/O errors, data loss. Cannot reload driver,
need to reboot system, fsck should be run on any failed disks before
remounting.
Controller restart
166 Single path implementation