3.5.1 Matrix Server Administration Guide

Chapter 18: SAN Maintenance 299

Increase the Membership Partition Timeout

Under heavy I/O load, I/O timeouts can occur on membership partition

accesses. The I/O timeouts are reported as ʺSCSI error : <...> return code =

50000ʺ in the file /var/log/messages. The I/O timeouts can cause problems

such as the following:

• Excessive path switching.

• Filesystems appearing to be hung when a node crashes. Large

numbers of I/O timeouts can extend the time it takes to fence the node,

and filesystem operations cannot resume until the node is fenced.

If your site is experiencing the above problems due to I/O timeouts, you

may want to increase the I/O timeout parameter for accessing

membership partitions. You will need to set the timeout on each node

currently in the matrix and on any nodes added to the matrix.

Before setting the timeout, be sure to stop Matrix Server.

To increase the timeout, edit the file /etc/opt/polyserve/mxinit.conf. Locate

the following line in the file:

# sanpulse_start_options = { "--mxinit" };

You will need to add the parameter "-o sdmp_io_timeout=<millisec>"

to the start options. Also remove the comment character (#) from the

beginning of the line:

sanpulse_start_options = { "--mxinit","-o sdmp_io_timeout=

<millisec>" };

<millisec> is the number of milliseconds to be used as the I/O timeout for

accessing membership partitions. The default value is 5,000ms (5

seconds). Be sure to increase the timeout value in small increments, such

as 5,000ms. If the timeout value is too large, more time may be needed for

filesystem recovery after a node failure. The following example sets the

timeout to 10,000ms (10 seconds):

sanpulse_start_options = { "--mxinit","-o sdmp_io_timeout=10000" };

After adding the parameter, you can restart Matrix Server.