Merge pull request 'ADD: agent count check.' (#43) from add-unique in…

…to master Reviewed-on: https://gitea:3000/AVENTER/mesos-compose/pulls/43
AVENTER-UG · Oct 18, 2023 · b74711f · b74711f
2 parents 4a7701c + c512f03
commit b74711f
Show file tree

Hide file tree

Showing 6 changed files with 20 additions and 16 deletions.
diff --git a/changelog.md b/changelog.md
@@ -32,7 +32,7 @@
 - FIX: Conflict between reconcile and heatbeat could end in a task restart loop
 - ADD: Parameter to configure the Mesos Task DiscoveryInfoName Delimiter `DISCOVERY_INFONAME_DELIMITER`. Default value is ".".
 - ADD: Parameter to configure the Mesos Task DiscoveryPortName Delimiter `DISCOVERY_PORTNAME_DELIMITER`. Default value is "_".
-- ADD: Constraint `unique` to run a only one instance of a task per node.
+- ADD: Constraint `unique` to run only one instance of a task per node.
 
 ## 0.4.2
 

diff --git a/docs/example/mesos-compose.yml b/docs/example/mesos-compose.yml
@@ -54,6 +54,7 @@ services:
           - "node.hostname==localhost"
           - "node.platform.os==linux"
           - "node.platform.arch==arm"
+          - "unique"
       replicas: 1
       resources:
         limits:

diff --git a/docs/example/test-http.yaml b/docs/example/test-http.yaml
@@ -18,9 +18,3 @@ services:
     restart: always
     deploy:
       replicas: 1
-
-networks:
-  default:
-    external: true
-    name: weave
-
diff --git a/mesos/mesos.go b/mesos/mesos.go
@@ -22,6 +22,7 @@ type Mesos struct {
 	Framework  *cfg.FrameworkConfig
 	IsSuppress bool
 	IsRevive   bool
+	CountAgent int
 }
 
 // Marshaler to serialize Protobuf Message to JSON
@@ -167,7 +168,7 @@ func (e *Mesos) Call(message *mesosproto.Call) error {
 		body, err := ioutil.ReadAll(res.Body)
 		if err != nil {
 			logrus.WithField("func", "mesos.Call").Error("Call Handling (could not read res.Body)")
-			return fmt.Errorf("Error %d", res.StatusCode)
+			return fmt.Errorf("error %d", res.StatusCode)
 		}
 
 		logrus.WithField("func", "mesos.Call").Error("Call Handling: ", string(body))
@@ -276,7 +277,7 @@ func (e *Mesos) GetAgentInfo(agentID string) cfg.MesosSlaves {
 	res, err := client.Do(req)
 
 	if err != nil {
-		logrus.WithField("func", "getAgentInfo").Error("Could not connect to agent: ", err.Error())
+		logrus.WithField("func", "mesos.getAgentInfo").Error("Could not connect to master: ", err.Error())
 		return cfg.MesosSlaves{}
 	}
 
@@ -295,6 +296,9 @@ func (e *Mesos) GetAgentInfo(agentID string) cfg.MesosSlaves {
 			return cfg.MesosSlaves{}
 		}
 
+		// save how many agents the cluster has
+		e.CountAgent = len(agent.Slaves)
+
 		// get the used agent info
 		for _, a := range agent.Slaves {
 			if a.ID == agentID {

diff --git a/scheduler/handle_offers.go b/scheduler/handle_offers.go
@@ -118,15 +118,18 @@ func (e *Scheduler) getOffer(offers *mesosproto.Event_Offers, cmd cfg.Command) (
 
 		if e.getLabelValue("__mc_placement", cmd) == "unique" {
 			if e.alreadyRunningOnHostname(cmd, offer) {
+				logrus.WithField("func", "scheduler.getOffer").Debug("UNIQUE: Already running on node: ", offer.GetHostname())
 				continue
 			}
 		}
 
 		if !e.isAttributeMachted("__mc_placement_node_platform_os", "os", cmd, offer) {
+			logrus.WithField("func", "scheduler.getOffer").Debug("OS: Does not match Attribute")
 			continue
 		}
 
 		if !e.isAttributeMachted("__mc_placement_node_platform_arch", "arch", cmd, offer) {
+			logrus.WithField("func", "scheduler.getOffer").Debug("OS: Does not match Attribute")
 			continue
 		}
 
@@ -164,7 +167,7 @@ func (e *Scheduler) alreadyRunningOnHostname(cmd cfg.Command, offer mesosproto.O
 			continue
 		}
 
-		if task.MesosAgent.Hostname == cmd.MesosAgent.Hostname {
+		if task.MesosAgent.Hostname == cmd.MesosAgent.Hostname && task.TaskID != cmd.TaskID {
 			return true
 		}
 	}

diff --git a/scheduler/heartbeat.go b/scheduler/heartbeat.go
@@ -46,12 +46,14 @@ func (e *Scheduler) Heartbeat() {
 
 		// there are lesser instances are running as it should be
 		if e.Redis.CountRedisKey(task.TaskName+":*", "__KILL") < task.Instances {
-			logrus.WithField("func", "scheduler.CheckState").Info("Scale up Mesos Task: ", task.TaskName)
-			e.Mesos.Revive()
-			task.State = ""
-			task.TaskID = e.API.IncreaseTaskCount(task.TaskID)
-			e.Redis.SaveTaskRedis(task)
-			continue
+			if e.getLabelValue("__mc_placement", task) != "unique" && e.Mesos.CountAgent >= task.Instances {
+				logrus.WithField("func", "scheduler.CheckState").Info("Scale up Mesos Task: ", task.TaskName)
+				e.Mesos.Revive()
+				task.State = ""
+				task.TaskID = e.API.IncreaseTaskCount(task.TaskID)
+				e.Redis.SaveTaskRedis(task)
+				continue
+			}
 		}
 
 		// there are more instances are running as it should be