Revise migration state machine to use notifyCompletedWithMessage()

* Remove entry action for `Done` state and replace with appropriate transition action. * Add new FSM action `notifyCompletedWithInsufficientReplicas()`. * Add new guarded transitions to `Done` which invoke `notifyCompletedWithInsufficientReplicas()` when appropriate. * Update state machine diagram.
dCache · Dec 10, 2024 · 37acd5b · 37acd5b
1 parent 5fee409
commit 37acd5b
Show file tree

Hide file tree

Showing 4 changed files with 67 additions and 10 deletions.
diff --git a/modules/dcache/src/main/java/org/dcache/pool/migration/Task.java b/modules/dcache/src/main/java/org/dcache/pool/migration/Task.java
@@ -241,6 +241,10 @@ synchronized boolean needsMoreReplicas() {
         return _replicas.size() < _parameters.replicas;
     }
 
+    synchronized boolean moreReplicasPossible() {
+      return _parameters.waitForTargets || (_replicas.size() < _parameters.poolList.getPools().size());
+    }
+
     /**
      * FSM Action
      */
@@ -347,6 +351,16 @@ void notifyCompleted() {
               new FireAndForgetTask(() -> _callbackHandler.taskCompleted(Task.this)));
     }
 
+    /**
+     * FSM Action
+     */
+    void notifyCompletedWithInsufficientReplicas() {
+      _parameters.executor.execute(
+              new FireAndForgetTask(() -> _callbackHandler.taskCompletedWithNote(Task.this,
+                      String.format("File replicas truncated at %s due to pool availability (%s requested)",
+                                    _replicas.size(), _parameters.replicas))));
+    }
+
     /**
      * FSM Action
      */

diff --git a/modules/dcache/src/main/smc/org/dcache/pool/migration/Task.sm b/modules/dcache/src/main/smc/org/dcache/pool/migration/Task.sm
@@ -181,6 +181,7 @@ Entry
         messageArrived(message: PoolMigrationCopyFinishedMessage)
                 Done
                 {
+                        notifyCompleted();
                 }
 
 }
@@ -266,6 +267,7 @@ Entry
         messageArrived(message: PoolMigrationCopyFinishedMessage)
                 Done
                 {
+                        notifyCompleted();
                 }
 }
 
@@ -359,6 +361,12 @@ Exit
                 UpdatingExistingFile
                 {
                 }
+        messageArrived(message: PoolMigrationCopyFinishedMessage)
+                [ ctxt.needsMoreReplicas() && ! (ctxt.moreReplicasPossible() || ctxt.isMetaOnly()) ]
+                Done
+                {
+                        notifyCompletedWithInsufficientReplicas();
+                }
         messageArrived(message: PoolMigrationCopyFinishedMessage)
                 [ ctxt.needsMoreReplicas() && !ctxt.isMetaOnly() ]
                 InitiatingCopy
@@ -378,6 +386,7 @@ Exit
         messageArrived(message: PoolMigrationCopyFinishedMessage)
                 Done
                 {
+                        notifyCompleted();
                 }
         cancel
                 Cancelling
@@ -435,6 +444,12 @@ Entry
                 UpdatingExistingFile
                 {
                 }
+        messageArrived(message: PoolMigrationCopyFinishedMessage)
+                [ ctxt.needsMoreReplicas() && ! (ctxt.moreReplicasPossible() || ctxt.isMetaOnly()) ]
+                Done
+                {
+                        notifyCompletedWithInsufficientReplicas();
+                }
         messageArrived(message: PoolMigrationCopyFinishedMessage)
                 [ ctxt.needsMoreReplicas() && !ctxt.isMetaOnly() ]
                 InitiatingCopy
@@ -454,6 +469,7 @@ Entry
         messageArrived(message: PoolMigrationCopyFinishedMessage)
                 Done
                 {
+                        notifyCompleted();
                 }
         cancel
                 Cancelling
@@ -527,6 +543,12 @@ Exit
                 UpdatingExistingFile
                 {
                 }
+        messageArrived(message: PoolMigrationCopyFinishedMessage)
+                [ ctxt.needsMoreReplicas() && ! (ctxt.moreReplicasPossible() || ctxt.isMetaOnly()) ]
+                Done
+                {
+                        notifyCompletedWithInsufficientReplicas();
+                }
         messageArrived(message: PoolMigrationCopyFinishedMessage)
                 [ ctxt.needsMoreReplicas() && !ctxt.isMetaOnly() ]
                 InitiatingCopy
@@ -546,6 +568,7 @@ Exit
         messageArrived(message: PoolMigrationCopyFinishedMessage)
                 Done
                 {
+                        notifyCompleted();
                 }
         cancel
                 Cancelling
@@ -595,6 +618,12 @@ Exit
                 UpdatingExistingFile
                 {
                 }
+        messageArrived(message: PoolMigrationCopyFinishedMessage)
+                [ ctxt.needsMoreReplicas() && ! (ctxt.moreReplicasPossible() || ctxt.isMetaOnly()) ]
+                Done
+                {
+                        notifyCompletedWithInsufficientReplicas();
+                }
         messageArrived(message: PoolMigrationCopyFinishedMessage)
                 [ ctxt.needsMoreReplicas() && !ctxt.isMetaOnly() ]
                 InitiatingCopy
@@ -614,6 +643,7 @@ Exit
         messageArrived(message: PoolMigrationCopyFinishedMessage)
                 Done
                 {
+                        notifyCompleted();
                 }
 }
 
@@ -626,6 +656,7 @@ Entry
         move_success
                 Done
                 {
+                        notifyCompleted();
                 }
         move_failure(rc: Integer, cause: Object)
                 Failed
@@ -689,6 +720,7 @@ Exit
                 [ message.getReturnCode() == 0 ]
                 Done
                 {
+                        notifyCompleted();
                 }
         cancel_success
                 nil
@@ -731,7 +763,6 @@ Failed
 Done
 Entry
 {
-        notifyCompleted();
 }
 {
 }

diff --git a/modules/dcache/src/main/smc/org/dcache/pool/migration/Task_sm.dot b/modules/dcache/src/main/smc/org/dcache/pool/migration/Task_sm.dot
@@ -51,7 +51,7 @@ digraph Task {
             [label="{Failed|Default()/\l}"];
 
         "TASK::Done"
-            [label="{Done|Entry/\l&nbsp;&nbsp;&nbsp;notifyCompleted();\l|Default()/\l}"];
+            [label="{Done|Entry/\l|Default()/\l}"];
 
         "%start"
             [label="" shape=circle style=filled fillcolor=black width=0.25];
@@ -143,7 +143,7 @@ digraph Task {
             [label="messageArrived(message: PoolMigrationCopyFinishedMessage)\l\[ ctxt.getMustMovePins() \]/\l"];
 
         "TASK::UpdatingExistingFile" -> "TASK::Done"
-            [label="messageArrived(message: PoolMigrationCopyFinishedMessage)/\l"];
+            [label="messageArrived(message: PoolMigrationCopyFinishedMessage)/\lnotifyCompleted();\l"];
 
         "TASK::InitiatingCopy" -> "TASK::Copying"
             [label="copy_success()/\l"];
@@ -187,7 +187,7 @@ digraph Task {
             [label="messageArrived(message: PoolMigrationCopyFinishedMessage)\l\[ ctxt.getMustMovePins() \]/\l"];
 
         "TASK::InitiatingCopy" -> "TASK::Done"
-            [label="messageArrived(message: PoolMigrationCopyFinishedMessage)/\l"];
+            [label="messageArrived(message: PoolMigrationCopyFinishedMessage)/\lnotifyCompleted();\l"];
 
         "TASK::WaitingForCopyReplicaReply" -> "TASK::UpdatingExistingFile"
             [label="copy_success()\l\[ ctxt.hasMoreLocations() \]/\l"];
@@ -232,6 +232,9 @@ digraph Task {
         "TASK::Copying" -> "TASK::UpdatingExistingFile"
             [label="messageArrived(message: PoolMigrationCopyFinishedMessage)\l\[ ctxt.needsMoreReplicas() && ctxt.hasMoreLocations() \]/\l"];
 
+        "TASK::Copying" -> "TASK::Done"
+            [label="messageArrived(message: PoolMigrationCopyFinishedMessage)\l\[ ctxt.needsMoreReplicas() && ! (ctxt.moreReplicasPossible() || ctxt.isMetaOnly()) \]/\lnotifyCompletedWithInsufficientReplicas();\l"];
+
         "TASK::Copying" -> "TASK::InitiatingCopy"
             [label="messageArrived(message: PoolMigrationCopyFinishedMessage)\l\[ ctxt.needsMoreReplicas() && !ctxt.isMetaOnly() \]/\l"];
 
@@ -242,7 +245,7 @@ digraph Task {
             [label="messageArrived(message: PoolMigrationCopyFinishedMessage)\l\[ ctxt.getMustMovePins() \]/\l"];
 
         "TASK::Copying" -> "TASK::Done"
-            [label="messageArrived(message: PoolMigrationCopyFinishedMessage)/\l"];
+            [label="messageArrived(message: PoolMigrationCopyFinishedMessage)/\lnotifyCompleted();\l"];
 
         "TASK::Copying" -> "TASK::Cancelling"
             [label="cancel()/\l"];
@@ -272,6 +275,9 @@ digraph Task {
         "TASK::Pinging" -> "TASK::UpdatingExistingFile"
             [label="messageArrived(message: PoolMigrationCopyFinishedMessage)\l\[ ctxt.needsMoreReplicas() && ctxt.hasMoreLocations() \]/\l"];
 
+        "TASK::Pinging" -> "TASK::Done"
+            [label="messageArrived(message: PoolMigrationCopyFinishedMessage)\l\[ ctxt.needsMoreReplicas() && ! (ctxt.moreReplicasPossible() || ctxt.isMetaOnly()) \]/\lnotifyCompletedWithInsufficientReplicas();\l"];
+
         "TASK::Pinging" -> "TASK::InitiatingCopy"
             [label="messageArrived(message: PoolMigrationCopyFinishedMessage)\l\[ ctxt.needsMoreReplicas() && !ctxt.isMetaOnly() \]/\l"];
 
@@ -282,7 +288,7 @@ digraph Task {
             [label="messageArrived(message: PoolMigrationCopyFinishedMessage)\l\[ ctxt.getMustMovePins() \]/\l"];
 
         "TASK::Pinging" -> "TASK::Done"
-            [label="messageArrived(message: PoolMigrationCopyFinishedMessage)/\l"];
+            [label="messageArrived(message: PoolMigrationCopyFinishedMessage)/\lnotifyCompleted();\l"];
 
         "TASK::Pinging" -> "TASK::Cancelling"
             [label="cancel()/\l"];
@@ -314,6 +320,9 @@ digraph Task {
         "TASK::NoResponse" -> "TASK::UpdatingExistingFile"
             [label="messageArrived(message: PoolMigrationCopyFinishedMessage)\l\[ ctxt.needsMoreReplicas() && ctxt.hasMoreLocations() \]/\l"];
 
+        "TASK::NoResponse" -> "TASK::Done"
+            [label="messageArrived(message: PoolMigrationCopyFinishedMessage)\l\[ ctxt.needsMoreReplicas() && ! (ctxt.moreReplicasPossible() || ctxt.isMetaOnly()) \]/\lnotifyCompletedWithInsufficientReplicas();\l"];
+
         "TASK::NoResponse" -> "TASK::InitiatingCopy"
             [label="messageArrived(message: PoolMigrationCopyFinishedMessage)\l\[ ctxt.needsMoreReplicas() && !ctxt.isMetaOnly() \]/\l"];
 
@@ -324,7 +333,7 @@ digraph Task {
             [label="messageArrived(message: PoolMigrationCopyFinishedMessage)\l\[ ctxt.getMustMovePins() \]/\l"];
 
         "TASK::NoResponse" -> "TASK::Done"
-            [label="messageArrived(message: PoolMigrationCopyFinishedMessage)/\l"];
+            [label="messageArrived(message: PoolMigrationCopyFinishedMessage)/\lnotifyCompleted();\l"];
 
         "TASK::NoResponse" -> "TASK::Cancelling"
             [label="cancel()/\l"];
@@ -346,6 +355,9 @@ digraph Task {
         "TASK::WaitingForCopyFinishedMessage" -> "TASK::UpdatingExistingFile"
             [label="messageArrived(message: PoolMigrationCopyFinishedMessage)\l\[ ctxt.needsMoreReplicas() && ctxt.hasMoreLocations() \]/\l"];
 
+        "TASK::WaitingForCopyFinishedMessage" -> "TASK::Done"
+            [label="messageArrived(message: PoolMigrationCopyFinishedMessage)\l\[ ctxt.needsMoreReplicas() && ! (ctxt.moreReplicasPossible() || ctxt.isMetaOnly()) \]/\lnotifyCompletedWithInsufficientReplicas();\l"];
+
         "TASK::WaitingForCopyFinishedMessage" -> "TASK::InitiatingCopy"
             [label="messageArrived(message: PoolMigrationCopyFinishedMessage)\l\[ ctxt.needsMoreReplicas() && !ctxt.isMetaOnly() \]/\l"];
 
@@ -356,10 +368,10 @@ digraph Task {
             [label="messageArrived(message: PoolMigrationCopyFinishedMessage)\l\[ ctxt.getMustMovePins() \]/\l"];
 
         "TASK::WaitingForCopyFinishedMessage" -> "TASK::Done"
-            [label="messageArrived(message: PoolMigrationCopyFinishedMessage)/\l"];
+            [label="messageArrived(message: PoolMigrationCopyFinishedMessage)/\lnotifyCompleted();\l"];
 
         "TASK::MovingPin" -> "TASK::Done"
-            [label="move_success()/\l"];
+            [label="move_success()/\lnotifyCompleted();\l"];
 
         "TASK::MovingPin" -> "TASK::Failed"
             [label="move_failure(rc: Integer, cause: Object)/\lfail(rc, \"Pin manager failed (\" + cause + \")\");\l"];
@@ -386,7 +398,7 @@ digraph Task {
                                            ctxt.getCancelReason()));\l"];
 
         "TASK::Cancelling" -> "TASK::Done"
-            [label="messageArrived(message: PoolMigrationCopyFinishedMessage)\l\[ message.getReturnCode() == 0 \]/\l"];
+            [label="messageArrived(message: PoolMigrationCopyFinishedMessage)\l\[ message.getReturnCode() == 0 \]/\lnotifyCompleted();\l"];
 
         "TASK::Cancelling" -> "TASK::Failed"
             [label="cancel_noroute()/\lfail(SERVICE_UNAVAILABLE, String.format(\"Cancelling task (%s) failed (no route)\",

diff --git a/modules/dcache/src/main/smc/org/dcache/pool/migration/Task_sm.dot.png b/modules/dcache/src/main/smc/org/dcache/pool/migration/Task_sm.dot.png