Merge pull request #1636 from jimenez/checkpoint_failover

Enabling checkpoint failover in FrameworkInfo

cli/help.go

@@ -46,11 +46,12 @@ Options:
    {{end}}{{if (eq .Name "manage")}}{{printf "\t * swarm.overcommit=0.05\tovercommit to apply on resources"}}
                                     {{printf "\t * swarm.createretry=0\tcontainer create retry count after initial failure"}}
                                     {{printf "\t * mesos.address=\taddress to bind on [$SWARM_MESOS_ADDRESS]"}}
+                                    {{printf "\t * mesos.checkpointfailover=false\tcheckpointing allows a restarted slave to reconnect with old executors and recover status updates, at the cost of disk I/O [$SWARM_MESOS_CHECKPOINT_FAILOVER]"}}
                                     {{printf "\t * mesos.port=\tport to bind on [$SWARM_MESOS_PORT]"}}
                                     {{printf "\t * mesos.offertimeout=30s\ttimeout for offers [$SWARM_MESOS_OFFER_TIMEOUT]"}}
+                                    {{printf "\t * mesos.offerrefusetimeout=5s\tseconds to consider unused resources refused [$SWARM_MESOS_OFFER_REFUSE_TIMEOUT]"}}
                                     {{printf "\t * mesos.tasktimeout=5s\ttimeout for task creation [$SWARM_MESOS_TASK_TIMEOUT]"}}
-                                    {{printf "\t * mesos.user=\tframework user [$SWARM_MESOS_USER]"}}
+                                    {{printf "\t * mesos.user=\tframework user [$SWARM_MESOS_USER]"}}{{end}}{{ end }}
-                                    {{printf "\t * mesos.offerrefusetimeout=5s\tseconds to consider unused resources refused [$SWARM_MESOS_OFFER_REFUSE_TIMEOUT]"}}{{end}}{{ end }}
 `
 
 }

cluster/mesos/cluster.go

@@ -119,6 +119,10 @@ func NewCluster(scheduler *scheduler.Scheduler, TLSConfig *tls.Config, master st
 		driverConfig.BindingAddress = bindingAddress
 	}
 
+	if checkpointFailover, ok := options.Bool("mesos.checkpointfailover", "SWARM_MESOS_CHECKPOINT_FAILOVER"); ok {
+		driverConfig.Framework.Checkpoint = &checkpointFailover
+	}
+
 	if offerTimeout, ok := options.String("mesos.offertimeout", "SWARM_MESOS_OFFER_TIMEOUT"); ok {
 		d, err := time.ParseDuration(offerTimeout)
 		if err != nil {

cluster/options.go

@@ -58,3 +58,12 @@ func (do DriverOpts) IP(key, env string) (net.IP, bool) {
 	}
 	return nil, false
 }
+
+// Bool returns a boolean from the driver options
+func (do DriverOpts) Bool(key, env string) (bool, bool) {
+	if value, ok := do.String(key, env); ok {
+		b, _ := strconv.ParseBool(value)
+		return b, true
+	}
+	return false, false
+}

test/integration/mesos/mesos_helpers.bash

@@ -8,7 +8,7 @@ export SWARM_MESOS_USER=root
 MESOS_IMAGE=dockerswarm/mesos:0.25.0
 MESOS_MASTER_PORT=$(( ( RANDOM % 1000 )  + 10000 ))
 
-# Start mesos master and slave.
+# Start mesos master and agent.
 function start_mesos() {
 	local current=${#DOCKER_CONTAINERS[@]}
 	MESOS_MASTER=$(
@@ -19,7 +19,7 @@ function start_mesos() {
 	retry 10 1 eval "docker_host ps | grep 'mesos-master'"
 	for ((i=0; i < current; i++)); do
 	    local docker_port=$(echo ${HOSTS[$i]} | cut -d: -f2)
-	    MESOS_SLAVES[$i]=$(
+	    MESOS_AGENTS[$i]=$(
 		docker_host run --privileged -d --name mesos-slave-$i --volumes-from node-$i -v /sys/fs/cgroup:/sys/fs/cgroup --net=host -u root \
 		$MESOS_IMAGE mesos-slave --master=127.0.0.1:$MESOS_MASTER_PORT --containerizers=docker --attributes="docker_port:$docker_port" --hostname=127.0.0.1 --port=$(($MESOS_MASTER_PORT + (1 + $i))) --docker=/usr/local/bin/docker
 		       )
@@ -37,7 +37,7 @@ function start_mesos_zk() {
 	retry 10 1 eval "docker_host ps | grep 'mesos-master'"
 	for ((i=0; i < current; i++)); do
 	    local docker_port=$(echo ${HOSTS[$i]} | cut -d: -f2)
-	    MESOS_SLAVES[$i]=$(
+	    MESOS_AGENTS[$i]=$(
 		docker_host run --privileged -d --name mesos-slave-$i --volumes-from node-$i -v /sys/fs/cgroup:/sys/fs/cgroup --net=host -u root \
 		$MESOS_IMAGE mesos-slave --master=127.0.0.1:$MESOS_MASTER_PORT --containerizers=docker --attributes="docker_port:$docker_port" --hostname=127.0.0.1 --port=$(($MESOS_MASTER_PORT + (1 + $i))) --docker=/usr/local/bin/docker
 			)
@@ -45,11 +45,11 @@ function start_mesos_zk() {
 	done
 }
 
-# Stop mesos master and slave.
+# Stop mesos master and agent
 function stop_mesos() {
 	echo "Stopping $MESOS_MASTER"
 	docker_host rm -f -v $MESOS_MASTER > /dev/null;
-	for id in ${MESOS_SLAVES[@]}; do
+	for id in ${MESOS_AGENTS[@]}; do
 	    echo "Stopping $id"
 	    docker_host rm -f -v $id > /dev/null;
 	done