3.1.2 MxFS-Linux Administration Guide

Chapter 2: Configure Export Groups 21

How the High-Availability Monitor Works

When configuring the high-availability monitor associated with an

Export Group, you should be aware of the actions taken by the monitor.

During each probe cycle, the monitor performs a series of checks:

1. The monitor first checks basic NFS Server health by issuing a NULL

RPC call to the NFS Server on the local node. If this call fails, the

Export Group is considered to be DOWN.

2. The monitor next checks the general health of the MxFS

high-availability service by checking for critical MxFS processes. If

any of these processes are not running, the Export Group is

considered to be DOWN. (The processes may not be running because

the node is still initializing or shutting down.)

3. The monitor then checks that each exported path in the Export Group

is available and is mounted on the PSFS filesystem. The monitor does

these checks in timed background threads. If any exported path

cannot be verified in one-half of the probe’s timeout value (or five

seconds, whichever is greater), the Export Group is considered to be

DOWN, as the verifying operations are simple and should reasonably

be completed in this time period.

Because the third check takes a linear amount of time based on the

number of exported paths in the Export Group, it is important to set the

probe timeout value to something larger than the default value of 15

seconds if the Export Group contains more than a few exported paths. If

the high-availability monitor cannot obtain reasonable responsiveness

from the system for exported paths, neither can NFS clients. This check

verifies the practical availability of an exported path as well as its literal

availability.

In the third check, the monitor only issues a warning if a path does not

exist (instead of considering the Export Group to be DOWN as it does in

the situations described above). This method avoids problems that can

arise if an exported path is deleted from a shared filesystem. Because the

purpose of the monitor is to determine whether the local node is healthy

enough to host the NFS service, rather than considering the entire Export

Group to be DOWN on a cluster-wide basis, the monitor will warn than it