fix timeout issues

distributed_lock.go

@@ -1,21 +1,5 @@
 package main
 
-import (
-	"sync"
-	"time"
-)
-
 type distributedLock struct {
 	nodes []*distributedLockNode
 }
-
-func (lock *distributedLock) runHeartbeats(
-	period time.Duration,
-	canceler *sync.Cond,
-) {
-	for _, node := range lock.nodes {
-		if node.connection != nil {
-			go heartbeat(period, node, canceler)
-		}
-	}
-}

distributed_lock_node.go

@@ -97,7 +97,8 @@ func (node *distributedLockNode) lock(
 	if err != nil {
 		return hierr.Errorf(
 			err,
-			`can't read line from lock process`,
+			`%s can't read lock status line from lock process`,
+			node,
 		)
 	}
 

lock.go

@@ -24,6 +24,7 @@ func acquireDistributedLock(
 	addresses []address,
 	noLockFail bool,
 	noConnFail bool,
+	heartbeat func(*distributedLockNode),
 ) (*distributedLock, error) {
 	var (
 		cluster = &distributedLock{}
@@ -71,6 +72,8 @@ func acquireDistributedLock(
 							errors <- err
 							return
 						}
+					} else {
+						go heartbeat(node)
 					}
 				}
 

main.go

@@ -647,12 +647,27 @@ func connectAndLock(
 		}
 	}
 
+	heartbeatMillisecondsBase, err := strconv.Atoi(sendTimeout)
+	if err != nil {
+		return nil, hierr.Errorf(
+			err,
+			`can't use --send-timeout as heartbeat timeout`,
+		)
+	}
+
+	heartbeatMilliseconds := time.Duration(
+		float64(heartbeatMillisecondsBase)*heartbeatTimeoutCoefficient,
+	) * time.Millisecond
+
 	cluster, err := acquireDistributedLock(
 		lockFile,
 		runners,
 		addresses,
 		noLockFail,
 		noConnFail,
+		func(node *distributedLockNode) {
+			heartbeat(heartbeatMilliseconds, node, canceler)
+		},
 	)
 	if err != nil {
 		return nil, hierr.Errorf(
@@ -663,21 +678,6 @@ func connectAndLock(
 
 	debugf(`global lock acquired on %d nodes`, len(cluster.nodes))
 
-	heartbeatMilliseconds, err := strconv.Atoi(sendTimeout)
-	if err != nil {
-		return nil, hierr.Errorf(
-			err,
-			`can't use --send-timeout as heartbeat timeout`,
-		)
-	}
-
-	cluster.runHeartbeats(
-		time.Duration(
-			float64(heartbeatMilliseconds)*heartbeatTimeoutCoefficient,
-		)*time.Millisecond,
-		canceler,
-	)
-
 	return cluster, nil
 }