Sun Cluster Data Services Developer's Guide for Solaris OS

Defining a Fault Monitor

The sample application implements a basic fault monitor to monitor the reliability of the DNS resource (in.named). The fault monitor consists of:

dns_probe, a user-defined program that uses nslookup to verify that the DNS resource controlled by the sample data service is running. If DNS is not running, this method attempts to restart it locally, or depending on the number of restart attempts, requests that the RGM relocate the data service to a different node.
dns_monitor_start, a callback method that launches dns_probe. The RGM automatically calls dns_monitor_start after the sample data service is brought online if monitoring is enabled.
dns_monitor_stop, a callback method that stops dns_probe. The RGM automatically calls dns_monitor_stop before bringing the sample data service offline.
dns_monitor_check, a callback method that calls the Validate method to verify that the configuration directory is available when the PROBE program fails the data service over to a new node.

Probe Program

The dns_probe program implements a continuously running process that verifies the DNS resource controlled by the sample data service is running. The dns_probe is launched by the dns_monitor_start method, which is automatically invoked by the RGM after the sample data service is brought online. The data service is stopped by the dns_monitor_stop method, which then RGM invokes before bringing the sample data service offline.

This section describes the major pieces of the PROBE method for the sample application. It does not describe functionality common to all callback methods, such as the parse_args() function and obtaining the syslog facility, which are described in Providing Common Functionality to All Methods.

For the complete listing of the PROBE method, see PROBE Program.

Probe Overview

The probe runs in an infinite loop. It uses nslookup to verify that the proper DNS resource is running. If DNS is running, the probe sleeps for a prescribed interval (set by the Thorough_probe_interval system-defined property) and then checks again. If DNS is not running, this program attempts to restart it locally, or depending on the number of restart attempts, requests that the RGM relocate the data service to a different node.

Obtaining Property Values

This program needs the values of the following properties:

Thorough_probe_interval – To set the period during which the probe sleeps
Probe_timeout – to enforce the time-out value of the probe on the nslookup command that does the probing
Network_resources_used – To obtain the IP address on which DNS is running
Retry_count and Retry_interval – To determine the number of restart attempts and the period over which to count them
RT_basedir – To obtain the directory containing the PROBE program and the gettime utility

The scha_resource_get() function obtains the values of these properties and stores them in shell variables, as follows.

PROBE_INTERVAL=`scha_resource_get -O THOROUGH_PROBE_INTERVAL \
-R $RESOURCE_NAME -G $RESOURCEGROUP_NAME`

probe_timeout_info=`scha_resource_get -O Extension -R $RESOURCE_NAME
\
-G $RESOURCEGROUP_NAME Probe_timeout` 
PROBE_TIMEOUT=`echo $probe_timeout_info | awk '{print $2}'`

DNS_HOST=`scha_resource_get -O NETWORK_RESOURCES_USED -R $RESOURCE_NAME
\
-G $RESOURCEGROUP_NAME`

RETRY_COUNT=`scha_resource_get -O RETRY_COUNT -R $RESOURCE_NAME
-G\
 $RESOURCEGROUP_NAME`

RETRY_INTERVAL=`scha_resource_get -O RETRY_INTERVAL -R $RESOURCE_NAME
-G\
 $RESOURCEGROUP_NAME`

RT_BASEDIR=`scha_resource_get -O RT_BASEDIR -R $RESOURCE_NAME -G\
 $RESOURCEGROUP_NAME`

Note –

For system-defined properties, such as Thorough_probe_interval, scha_resource_get() returns the value only. For extension properties, such as Probe_timeout, scha_resource_get() returns the type and value. Use the awk command to obtain the value only.

Checking the Reliability of the Service

The probe itself is an infinite while loop of nslookup commands. Before the while loop, a temporary file is set up to hold the nslookup replies. The probefail and retries variables are initialized to 0.

# Set up a temporary file for the nslookup replies.
DNSPROBEFILE=/tmp/.$RESOURCE_NAME.probe
probefail=0
retries=0

The while loop itself:

Sets the sleep interval for the probe
Uses hatimerun to launch nslookup passing the Probe_timeout value and identifying the target host
Sets the probefail variable based on the success or failure of the nslookup return code
If probefail is set to 1 (failure), verifies that the reply to nslookup came from the sample data service and not some other DNS server

Here is the while loop code.

while :
do
   # The interval at which the probe needs to run is specified in the
   # property THOROUGH_PROBE_INTERVAL. Therefore, set the probe to sleep
   # for a duration of THOROUGH_PROBE_INTERVAL.
   sleep $PROBE_INTERVAL

   # Run an nslookup command of the IP address on which DNS is serving.
   hatimerun -t $PROBE_TIMEOUT /usr/sbin/nslookup $DNS_HOST $DNS_HOST \
          > $DNSPROBEFILE 2>&1

      retcode=$?
      if [ $retcode -ne 0 ]; then
            probefail=1
      fi

   # Make sure that the reply to nslookup comes from the HA-DNS
   # server and not from another nameserver mentioned in the 
   # /etc/resolv.conf file.
   if [ $probefail -eq 0 ]; then
# Get the name of the server that replied to the nslookup query.
   SERVER=` awk ' $1=="Server:" { print $2 }' \
   $DNSPROBEFILE | awk -F. ' { print $1 } ' `
   if [ -z "$SERVER" ]; then
      probefail=1
      else
         if [ $SERVER != $DNS_HOST ]; then
            probefail=1
         fi
   fi
fi

Evaluating Restart Versus Failover

If the probefail variable is something other than 0 (success), it means the nslookup command timed out or that the reply came from a server other than the sample service's DNS. In either case, the DNS server is not functioning as expected and the fault monitor calls the decide_restart_or_failover() function to determine whether to restart the data service locally or request that the RGM relocate the data service to a different node. If the probefail variable is 0, then a message is generated that the probe was successful.

   if [ $probefail -ne 0 ]; then
         decide_restart_or_failover
   else
         logger -p ${SYSLOG_FACILITY}.err\
         -t [$SYSLOG_TAG]\
         "${ARGV0} Probe for resource HA-DNS successful"
   fi

The decide_restart_or_failover() function uses a time window (Retry_interval) and a failure count (Retry_count) to determine whether to restart DNS locally or request that the RGM relocate the data service to a different node. It implements the following conditional code (see the code listing for decide_restart_or_failover() in PROBE Program).

If this is the first failure, restart the data service. Log an error message and bump the counter in the retries variable.
If this is not the first failure, but the window has been exceeded, restart the data service. Log an error message, reset the counter, and slide the window.
If the time is still within the window and the retry counter has been exceeded, then fail over to another node. If the fail over does not succeed, log an error and exit the probe program with status 1 (failure).
If time is still within the window but the retry counter has not been exceeded, restart the data service. Log an error message and bump the counter in the retries variable.

If the number of restarts reaches the limit during the time interval, the function requests that the RGM relocate the data service to a different node. If the number of restarts is under the limit, or the interval has been exceeded so the count begins again, the function attempts to restart DNS on the same node. Note the following about this function:

The gettime utility is used to track the time between restarts. This is a C program residing in the (RT_basedir) directory.
The Retry_count and Retry_interval system-defined resource properties determine the number of restart attempts and the interval over which to count. These properties default to 2 attempts in a period of 5 minutes (300 seconds) in the RTR file, though the cluster administrator could change them.
The restart_service() function is called to attempt to restart the data service on the same node. See the next section, Restarting the Data Service, for information about this function.
The scha_control() API function, with the GIVEOVER option, brings the resource group containing the sample data service offline and back online on a different node.

Restarting the Data Service

The restart_service() function is called by decide_restart_or_failover() to attempt to restart the data service on the same node. This function does the following.

It determines if the data service is still registered under PMF. If the service is still registered, the function:
- Obtains the Stop method name and the Stop_timeout value for the data service.
- Uses hatimerun to launch the Stop method for the data service, passing the Stop_timeout value.
- (If the data service is successfully stopped) obtains the Start method name and the Start_timeout value for the data service.
- Uses hatimerun to launch the Start method for the data service, passing the Start_timeout value.
If the data service is no longer registered under PMF, the implication is that the data service has exceeded the maximum number of allowable retries under PMF, so the scha_control() function is called with the GIVEOVER option to fail the data service over to a different node.

function restart_service
{

        # To restart the data service, first verify that the 
        # data service itself is still registered under PMF.
        pmfadm -q $PMF_TAG
        if [[ $? -eq 0 ]]; then
                # Since the TAG for the data service is still registered under
                # PMF, first stop the data service and start it back up again.

                # Obtain the Stop method name and the STOP_TIMEOUT value for
                # this resource.
                STOP_TIMEOUT=`scha_resource_get -O STOP_TIMEOUT \
                        -R $RESOURCE_NAME -G $RESOURCEGROUP_NAMÈ
                STOP_METHOD=`scha_resource_get -O STOP \
                        -R $RESOURCE_NAME -G $RESOURCEGROUP_NAMÈ
                hatimerun -t $STOP_TIMEOUT $RT_BASEDIR/$STOP_METHOD \
                        -R $RESOURCE_NAME -G $RESOURCEGROUP_NAME \
                        -T $RESOURCETYPE_NAME

                if [[ $? -ne 0 ]]; then
                        logger-p ${SYSLOG_FACILITY}.err -t [$SYSLOG_TAG] \
                                “${ARGV0} Stop method failed.”
                        return 1
                fi

                # Obtain the START method name and the START_TIMEOUT value for
                # this resource.
                START_TIMEOUT=`scha_resource_get -O START_TIMEOUT \
                        -R $RESOURCE_NAME -G $RESOURCEGROUP_NAMÈ
                START_METHOD=`scha_resource_get -O START \
                        -R $RESOURCE_NAME -G $RESOURCEGROUP_NAMÈ
                hatimerun -t $START_TIMEOUT $RT_BASEDIR/$START_METHOD \
                        -R $RESOURCE_NAME -G $RESOURCEGROUP_NAME \
                        -T $RESOURCETYPE_NAME

                if [[ $? -ne 0 ]]; then
                        logger-p ${SYSLOG_FACILITY}.err -t [$SYSLOG_TAG] \
                                “${ARGV0} Start method failed.”
                        return 1
                fi


        else
                # The absence of the TAG for the dataservice 
                # implies that the data service has already
                # exceeded the maximum retries allowed under PMF. 
                # Therefore, do not attempt to restart the
                # data service again, but try to failover
                # to another node in the cluster.
                scha_control -O GIVEOVER -G $RESOURCEGROUP_NAME \
                        -R $RESOURCE_NAME
        fi

        return 0
}

Probe Exit Status

The sample data service's PROBE program exits with failure if attempts to restart locally have failed and the attempt to fail over to a different node has failed as well. It logs the message, “Failover attempt failed”.

`Monitor_start` Method

The RGM calls the Monitor_start method to launch the dns_probe method after the sample data service is brought online.

This section describes the major pieces of the Monitor_start method for the sample application. This section does not describe functionality common to all callback methods, such as the parse_args() function and obtaining the syslog facility, which are described in Providing Common Functionality to All Methods.

For the complete listing of the Monitor_start method, see Monitor_start Method.

`Monitor_start` Overview

This method uses the PMF (pmfadm) to launch the probe.

Starting the Probe

The Monitor_start method obtains the value of the RT_basedir property to construct the full path name for the PROBE program. This method launches the probe using the infinite retries option of pmfadm (-n -1, -t -1), which means if the probe fails to start, PMF tries to start it an infinite number of times over an infinite period of time.

# Find where the probe program resides by obtaining the value of the
# RT_BASEDIR property of the resource.
RT_BASEDIR=`scha_resource_get -O RT_BASEDIR -R $RESOURCE_NAME -G \
$RESOURCEGROUP_NAME`

# Start the probe for the data service under PMF. Use the infinite retries 
# option to start the probe. Pass the resource name, type, and group to the 
# probe program. 
pmfadm -c $RESOURCE_NAME.monitor -n -1 -t -1 \
   $RT_BASEDIR/dns_probe -R $RESOURCE_NAME -G $RESOURCEGROUP_NAME \
   -T $RESOURCETYPE_NAME

`Monitor_stop` Method

The RGM calls the Monitor_stop method to stop execution of dns_probe when the sample data service is brought offline.

This section describes the major pieces of the Monitor_stop method for the sample application. This section does not describe functionality common to all callback methods, such as the parse_args() function and obtaining the syslog facility, which are described in Providing Common Functionality to All Methods.

For the complete listing of the Monitor_stop method, see Monitor_stop Method.

`Monitor_stop` Overview

This method uses the PMF (pmfadm) to see if the probe is running, and if so, to stop it.

Stopping the Monitor

The Monitor_stop method uses pmfadm -q to see if the probe is running, and if so, uses pmfadm -s to stop it. If the probe is already stopped, the method exits successfully anyway, which guarantees the idempotency of the method.

# See if the monitor is running, and if so, kill it. 
if pmfadm -q $PMF_TAG; then 
   pmfadm -s $PMF_TAG KILL
   if [ $? -ne 0 ]; then 
         logger -p ${SYSLOG_FACILITY}.err \
            -t [$SYSLOG_TAG] \
            "${ARGV0} Could not stop monitor for resource " \
            $RESOURCE_NAME
           exit 1
   else
         # could successfully stop the monitor. Log a message.
         logger -p ${SYSLOG_FACILITY}.err \
            -t [$SYSLOG_TAG] \
            "${ARGV0} Monitor for resource " $RESOURCE_NAME \
            " successfully stopped"
   fi
fi
exit 0

Caution –

Be certain to use the KILL signal with pmfadm to stop the probe and not a maskable signal such as TERM. Otherwise the Monitor_stop method can hang indefinitely and eventually time out. The reason for this problem is that the PROBE method calls scha_control() when it is necessary to restart or fail over the data service. When scha_control() calls Monitor_stop as part of the process of bringing the data service offline, if Monitor_stop uses a maskable signal, it hangs waiting for scha_control() to complete and scha_control() hangs waiting for Monitor_stop to complete.

`Monitor_stop` Exit Status

The Monitor_stop method logs an error message if it cannot stop the PROBE method. The RGM puts the sample data service into MONITOR_FAILED state on the primary node, which can panic the node.

Monitor_stop should not exit before the probe has been stopped.

`Monitor_check` Method

The RGM calls the Monitor_check method whenever the PROBE method attempts to fail the resource group containing the data service over to a new node.

This section describes the major pieces of the Monitor_check method for the sample application. This section does not describe functionality common to all callback methods, such as the parse_args() function and obtaining the syslog facility, which are described in Providing Common Functionality to All Methods.

For the complete listing of the Monitor_check method, see Monitor_check Method.

The Monitor_check method must be implemented so that it does not conflict with other methods running concurrently.

The Monitor_check method calls the Validate method to verify that the DNS configuration directory is available on the new node. The Confdir extension property points to the DNS configuration directory. Therefore Monitor_check obtains the path and name for the Validate method and the value of Confdir. It passes this value to Validate, as shown in the following listing.

# Obtain the full path for the Validate method from
# the RT_BASEDIR property of the resource type.
RT_BASEDIR=`scha_resource_get -O RT_BASEDIR -R $RESOURCE_NAME \
   -G $RESOURCEGROUP_NAMÈ

# Obtain the name of the Validate method for this resource.
VALIDATE_METHOD=`scha_resource_get -O VALIDATE \
   -R $RESOURCE_NAME -G $RESOURCEGROUP_NAMÈ

# Obtain the value of the Confdir property in order to start the
# data service. Use the resource name and the resource group entered to
# obtain the Confdir value set at the time of adding the resource.
config_info=`scha_resource_get -O Extension -R $RESOURCE_NAME \
 -G $RESOURCEGROUP_NAME Confdir`

# scha_resource_get returns the type as well as the value for extension
# properties. Use awk to get only the value of the extension property.
CONFIG_DIR=`echo $config_info | awk `{print $2}'`

# Call the validate method so that the dataservice can be failed over 
# successfully to the new node.
$RT_BASEDIR/$VALIDATE_METHOD -R $RESOURCE_NAME -G $RESOURCEGROUP_NAME \
   -T $RESOURCETYPE_NAME -x Confdir=$CONFIG_DIR

See the Validate Method to see how the sample application verifies the suitability of a node for hosting the data service.