Fix starting agent after idle shutdown (#609)

jenkinsci · Dec 20, 2024 · 7288787 · 7288787
1 parent 45bbb66
commit 7288787
Show file tree

Hide file tree

Showing 3 changed files with 23 additions and 11 deletions.
diff --git a/src/main/java/com/microsoft/azure/vmagent/AzureVMAgent.java b/src/main/java/com/microsoft/azure/vmagent/AzureVMAgent.java
@@ -403,6 +403,11 @@ private void setCleanUpReason(Localizable cleanUpReason) {
     public void clearCleanUpAction() {
         setCleanUpAction(CleanUpAction.DEFAULT);
         setCleanUpReason(null);
+
+        Computer computer = toComputer();
+        if (computer != null) {
+            computer.setTemporaryOfflineCause(null);
+        }
     }
 
     /**
@@ -424,7 +429,11 @@ public void setCleanUpAction(CleanUpAction action, Localizable reason) {
         AzureVMComputer computer = (AzureVMComputer) this.toComputer();
         if (computer != null) {
             // Set the machine temporarily offline machine with an offline reason.
-            computer.setTemporarilyOffline(true, OfflineCause.create(reason));
+            if (action == CleanUpAction.SHUTDOWN) {
+                computer.setTemporaryOfflineCause(new OfflineCause.IdleOfflineCause());
+            } else {
+                computer.setTemporaryOfflineCause(OfflineCause.create(reason));
+            }
         }
         setCleanUpAction(action);
         setCleanUpReason(reason);
@@ -570,7 +579,7 @@ public AzureVMCloud getCloud() {
 
     public synchronized void shutdown(Localizable reason) {
         if (isEligibleForReuse()) {
-            LOGGER.log(Level.INFO, "Agent {0} is always shut down", this.
+            LOGGER.log(Level.INFO, "Agent {0} is shut down", this.
                     getDisplayName());
             return;
         }
@@ -583,7 +592,7 @@ public synchronized void shutdown(Localizable reason) {
             return;
         }
         computer.setAcceptingTasks(false);
-        computer.disconnect(OfflineCause.create(reason));
+        computer.setTemporaryOfflineCause(new OfflineCause.IdleOfflineCause());
         LOGGER.log(Level.INFO, "Shutting down agent {0}", this.getDisplayName());
 
         AzureVMManagementServiceDelegate serviceDelegate = getServiceDelegate();
@@ -615,8 +624,7 @@ public synchronized void deprovision(Localizable reason) throws Exception {
 
         ComputerLauncher launcher = computer.getLauncher();
 
-        if ((launcher instanceof AzureVMAgentSSHLauncher)) {
-            AzureVMAgentSSHLauncher azureLauncher = (AzureVMAgentSSHLauncher) launcher;
+        if ((launcher instanceof AzureVMAgentSSHLauncher azureLauncher)) {
             PrintStream terminateStream = new LogTaskListener(LOGGER, Level.INFO).getLogger();
 
             final boolean isUnix = this.getOsType().equals(OperatingSystemTypes.LINUX);
@@ -649,7 +657,8 @@ public synchronized void deprovision(Localizable reason) throws Exception {
                 }
                 if (!skipTerminateScript
                         && azureLauncher.executeRemoteCommand(this, command, terminateStream, isUnix) != 0) {
-                    LOGGER.info("Terminate script is not null, " + "preparing to execute script remotely");
+                    LOGGER.info("Terminate script present for " + computer.getName()
+                    + " preparing to execute script remotely");
                     if (isUnix) {
                         azureLauncher.copyFileToRemote(
                                 this,

diff --git a/src/main/java/com/microsoft/azure/vmagent/AzureVMAgentCleanUpTask.java b/src/main/java/com/microsoft/azure/vmagent/AzureVMAgentCleanUpTask.java
@@ -484,15 +484,20 @@ private void cleanVMs() {
     private void cleanVMs(ExecutionEngine executionEngine) {
         LOGGER.log(getNormalLoggingLevel(), "Beginning");
         for (Computer computer : Jenkins.get().getComputers()) {
-            if (computer instanceof AzureVMComputer) {
-                AzureVMComputer azureComputer = (AzureVMComputer) computer;
+            if (computer instanceof AzureVMComputer azureComputer) {
                 final AzureVMAgent agentNode = azureComputer.getNode();
 
                 // If the machine is not offline, then don't do anything.
                 if (!azureComputer.isOffline()) {
                     continue;
                 }
 
+                if (agentNode == null) {
+                    LOGGER.log(getNormalLoggingLevel(), "Node {0} is missing, skipping",
+                            azureComputer.getDisplayName());
+                    continue;
+                }
+
                 // Even if offline, a machine that has been temporarily marked offline
                 // should stay (this could be for investigation).
                 if (azureComputer.isSetOfflineByUser()) {

diff --git a/src/main/java/com/microsoft/azure/vmagent/AzureVMCloud.java b/src/main/java/com/microsoft/azure/vmagent/AzureVMCloud.java
@@ -657,10 +657,8 @@ public Collection<PlannedNode> provision(CloudState cloudState, int workLoad) {
             if (numberOfAgents == 0) {
                 break;
             }
-            if (agentComputer instanceof AzureVMComputer && agentComputer.isOffline()) {
-                final AzureVMComputer azureComputer = (AzureVMComputer) agentComputer;
+            if (agentComputer instanceof AzureVMComputer azureComputer && agentComputer.isOffline()) {
                 final AzureVMAgent agentNode = azureComputer.getNode();
-
                 if (agentNode != null && isNodeEligibleForReuse(agentNode, template)) {
                     LOGGER.log(Level.FINE, "Agent computer eligible for reuse {0}", agentComputer.getName());