[dolphinscheduler] branch dev updated: [fix][worker][bug] master/worker crash when registry recover from SUSPENDED to RECONNECTED (#13328)

[wenjun]

This is an automated email from the ASF dual-hosted git repository.

wenjun pushed a commit to branch dev
in repository https://gitbox.apache.org/repos/asf/dolphinscheduler.git


The following commit(s) were added to refs/heads/dev by this push:
     new 3b980cb06a [fix][worker][bug] master/worker crash when registry 
recover from SUSPENDED to RECONNECTED (#13328)
3b980cb06a is described below

commit 3b980cb06a22fc1a2e917547c98bb8a4a307f90e
Author: hokie-chan <88499013+hokie-c...@users.noreply.github.com>
AuthorDate: Tue Jan 3 19:24:11 2023 +0800

    [fix][worker][bug] master/worker crash when registry recover from SUSPENDED 
to RECONNECTED (#13328)
---
 .../master/registry/MasterWaitingStrategy.java     | 28 ++++++++++++----------
 .../worker/registry/WorkerWaitingStrategy.java     | 28 ++++++++++++----------
 2 files changed, 32 insertions(+), 24 deletions(-)

diff --git 
a/dolphinscheduler-master/src/main/java/org/apache/dolphinscheduler/server/master/registry/MasterWaitingStrategy.java
 
b/dolphinscheduler-master/src/main/java/org/apache/dolphinscheduler/server/master/registry/MasterWaitingStrategy.java
index a007e221b8..4556500db6 100644
--- 
a/dolphinscheduler-master/src/main/java/org/apache/dolphinscheduler/server/master/registry/MasterWaitingStrategy.java
+++ 
b/dolphinscheduler-master/src/main/java/org/apache/dolphinscheduler/server/master/registry/MasterWaitingStrategy.java
@@ -93,18 +93,22 @@ public class MasterWaitingStrategy implements 
MasterConnectStrategy {
 
     @Override
     public void reconnect() {
-        try {
-            ServerLifeCycleManager.recoverFromWaiting();
-            reStartMasterResource();
-            // reopen the resource
-            logger.info("Recover from waiting success, the current server 
status is {}",
-                    ServerLifeCycleManager.getServerStatus());
-        } catch (Exception e) {
-            String errorMessage =
-                    String.format("Recover from waiting failed, the current 
server status is %s, will stop the server",
-                            ServerLifeCycleManager.getServerStatus());
-            logger.error(errorMessage, e);
-            registryClient.getStoppable().stop(errorMessage);
+        if (ServerLifeCycleManager.isRunning()) {
+            logger.info("no need to reconnect, as the current server status is 
running");
+        } else {
+            try {
+                ServerLifeCycleManager.recoverFromWaiting();
+                reStartMasterResource();
+                logger.info("Recover from waiting success, the current server 
status is {}",
+                        ServerLifeCycleManager.getServerStatus());
+            } catch (Exception e) {
+                String errorMessage =
+                        String.format(
+                                "Recover from waiting failed, the current 
server status is %s, will stop the server",
+                                ServerLifeCycleManager.getServerStatus());
+                logger.error(errorMessage, e);
+                registryClient.getStoppable().stop(errorMessage);
+            }
         }
     }
 
diff --git 
a/dolphinscheduler-worker/src/main/java/org/apache/dolphinscheduler/server/worker/registry/WorkerWaitingStrategy.java
 
b/dolphinscheduler-worker/src/main/java/org/apache/dolphinscheduler/server/worker/registry/WorkerWaitingStrategy.java
index 1549385fe9..24a609b74f 100644
--- 
a/dolphinscheduler-worker/src/main/java/org/apache/dolphinscheduler/server/worker/registry/WorkerWaitingStrategy.java
+++ 
b/dolphinscheduler-worker/src/main/java/org/apache/dolphinscheduler/server/worker/registry/WorkerWaitingStrategy.java
@@ -93,19 +93,23 @@ public class WorkerWaitingStrategy implements 
WorkerConnectStrategy {
 
     @Override
     public void reconnect() {
-        try {
-            ServerLifeCycleManager.recoverFromWaiting();
-            reStartWorkerResource();
-            logger.info("Recover from waiting success, the current server 
status is {}",
-                    ServerLifeCycleManager.getServerStatus());
-        } catch (Exception e) {
-            String errorMessage =
-                    String.format("Recover from waiting failed, the current 
server status is %s, will stop the server",
-                            ServerLifeCycleManager.getServerStatus());
-            logger.error(errorMessage, e);
-            registryClient.getStoppable().stop(errorMessage);
+        if (ServerLifeCycleManager.isRunning()) {
+            logger.info("no need to reconnect, as the current server status is 
running");
+        } else {
+            try {
+                ServerLifeCycleManager.recoverFromWaiting();
+                reStartWorkerResource();
+                logger.info("Recover from waiting success, the current server 
status is {}",
+                        ServerLifeCycleManager.getServerStatus());
+            } catch (Exception e) {
+                String errorMessage =
+                        String.format(
+                                "Recover from waiting failed, the current 
server status is %s, will stop the server",
+                                ServerLifeCycleManager.getServerStatus());
+                logger.error(errorMessage, e);
+                registryClient.getStoppable().stop(errorMessage);
+            }
         }
-
     }
 
     @Override