Hello,

We have a replica set of 3 mongo server running on 3 hosts under debian 6x linux.
They were running with mongodb 2.0.2 and we are trying a migration to mongodb 2.2.1.

On test cluster everything went fine.
On prod cluster, we modified the init.d script some time ago to correctly handle NUMA options.

In mongodb 2.2.1, the numa is directly handled in the script delivered by the package with the following lines;

1. Handle NUMA access to CPUs (SERVER-3574)
2. This verifies the existence of numactl as well as testing that the command works
   NUMACTL_ARGS="--interleave=all"
   if which numactl >/dev/null 2>/dev/null && numactl $NUMACTL_ARGS ls / >/dev/null 2>/dev/null
   then
   NUMACTL="numactl $NUMACTL_ARGS"
   else
   NUMACTL=""
   fi
   ...
   start_server() {
3. Start the process using the wrapper
   start-stop-daemon --background --start --quiet --pidfile $PIDFILE \
   --make-pidfile --chuid $DAEMONUSER \
   --exec $NUMACTL $DAEMON – $DAEMON_OPTS
   errcode=$?
   return $errcode
   }

Unfortunately, this does not work, I get the error: "unrecognized option '--interleave=all'"

After some investigation it seems that the '--' is badly placed and the init.d script should instead contain the following lines:

1. Handle NUMA access to CPUs (SERVER-3574)
2. This verifies the existence of numactl as well as testing that the command works
   NUMACTL_ARGS="--interleave=all"
   if which numactl >/dev/null 2>/dev/null && numactl $NUMACTL_ARGS ls / >/dev/null 2>/dev/null
   then
3. NUMACTL="numactl $NUMACTL_ARGS"
   NUMACTL="/usr/bin/numactl – $NUMACTL_ARGS"
   else
   NUMACTL=""
   fi
   ...
   start_server() {
4. Start the process using the wrapper
   start-stop-daemon --background --start --quiet --pidfile $PIDFILE \
   --make-pidfile --chuid $DAEMONUSER \
   --exec $NUMACTL $DAEMON $DAEMON_OPTS
   errcode=$?
   return $errcode
   }

Could you please fix it in later mongodb releases/patches?

Regards