White Papers

Server configuration
15 Dell EMC SC Series: Red Hat Enterprise Linux Best Practices | CML1031
2.4.4 iSCSI timeout
The pre-configured iSCSI settings, defined in /etc/iscsi/iscsi.conf, work with SC Series storage system when
the Linux system is also configured to use dm-multipath. It is a best practice to use dm-multipath for path
redundancy and failover. It is important to note the following timeout settings because they directly affect the
storage availability and failure recovery for a SAN storage system. When an SC Series storage system
controller fails, services and volumes fail over to the surviving controller and this failover might take up to 60
seconds. Therefore, ensure the iSCSI timeout settings are sufficient to accommodate this kind of event.
These preconfigured timeout settings listed below work well with SC Series storage systems.
• node.conn[0].timeo.noop_out_interval = 5
• node.conn[0].timeo.noop_out_timeout = 5
• node.session.timeo.replacement_timeout = 120
noop_out_interval controls how often to send iSCSI NOP-Out requests. noop_out_timeout controls the
number of seconds to wait for a NOP-Out. Together, they control how quickly the iSCSI layer detects network
problems. The replacement_timeout controls the number of seconds to wait for session re-establishment
before failing pending SCSI commands and commands worked on by the SCSI error handler.
When a network problem is detected (a NOP-Out times out), the iSCSI layer fails the running commands to
the SCSI layer. If dm-multipath is used, the SCSI layer fails them to the multipath layer to handle the
queueing and retry on other paths according to the policy in /etc/multipath.conf (see section 3.6). If dm-
multipath is not used, the SCSI layer retries the commands 5 times before failing them. The SCSI device
command timeout is defined in /dev/block/sdX/device/timeout (see section 2.2).
When the SCSI layer error handler kicks in (triggered by the timeout in /dev/block/sdX/device/timeout) and
is running, the commands do not fail immediately. Instead, they wait
node.session.timeo.replacement_timeout seconds before failing them. After the replacement_timeout
seconds, the commands are failed to the multipath layer. To verify the error handler is running, run iscsiadm
command and look for State: Recovery for any Host Number.
# iscsiadm -m session -P 3 |egrep "Host Number"
Host Number: X State: Recovery
The iSCSI configuration should be thoroughly tested in the environment against all perceivable failover
scenarios (such as switch, port, fabric, and controller) before deploying them into a production environment.
2.4.5 Filesystem on iSCSI volume
Since iSCSI is dependent on a running network, filesystems created on iSCSI volumes should be designated
as network-dependent in /etc/fstab. In other words, do not attempt to mount an iSCSI volume until the
network layer services have fully started. The following example demonstrates how to create this network
dependency to the iSCSI mount using the _netdev mount option in /etc/fstab.
LABEL=iscsiVOL /mnt/iscsi ext4 _netdev 0 0