attempt to make restart of a process more reliable in the case where it stopped

responding to the Command server for a while.

ChangeLog

@@ -1,3 +1,17 @@
+2020-04-02 Richard Frith-Macdonald <rfm@gnu.org>
+
+	* EcCommand.m:
+	* EcLogger.m:
+	* EcProcess.h:
+	* EcProcess.m:
+	Remove obsolete methods for registration/unregistration.  Change code
+	to refrain from registering with Command server if in the process of
+	shutting down, and to refrain from attempting an unregistration if
+	not already registered.  The hope is that this will prevent a race
+	condition where loss of connection to the Command server can cause a
+	shutdown and re-register during the shutdown fooling the Command
+	server into thinking the process is still running.
+
 2020-03-23 Richard Frith-Macdonald <rfm@gnu.org>
 
 	* EcProcess.h:

EcCommand.m

@@ -549,8 +549,6 @@ static NSMutableDictionary	*launchInfo = nil;
 - (void) quitAll: (NSDate*)by;
 - (void) reLaunch: (NSTimer*)t;
 - (void) requestConfigFor: (id<CmdConfig>)c;
-- (NSData*) registerClient: (id<CmdClient>)c
-		      name: (NSString*)n;
 - (NSData*) registerClient: (id<CmdClient>)c
 		      name: (NSString*)n
 		 transient: (BOOL)t;
@@ -563,7 +561,6 @@ static NSMutableDictionary	*launchInfo = nil;
 - (void) _tryLaunch: (NSTimer*)t;
 - (void) tryLaunchSoon;
 - (void) unregisterByObject: (id)obj;
-- (void) unregisterByName: (NSString*)n;
 - (void) unregisterClient: (EcClientI*)o gracefully: (BOOL)clean;
 - (void) update;
 - (void) updateConfig: (NSData*)data;
@@ -2947,12 +2944,6 @@ NSLog(@"Problem %@", localException);
     }
 }
 
-- (NSData*) registerClient: (id<CmdClient>)c
-		      name: (NSString*)n
-{
-  return [self registerClient: c name: n transient: NO];
-}
-
 - (NSData*) registerClient: (id<CmdClient>)c
 		      name: (NSString*)n
 		 transient: (BOOL)t
@@ -3032,6 +3023,10 @@ NSLog(@"Problem %@", localException);
     }
   else
     {
+      /* Rejecting means the client is not registered (and therefore should
+       * not be told to quit when the objct is deallocated.
+       */
+      [obj setUnregistered: YES];
       RELEASE(obj);
       m = [NSString stringWithFormat: 
 	@"%@ rejected new server with name '%@' on %@\n",
@@ -4024,14 +4019,10 @@ NSLog(@"Problem %@", localException);
 {
   EcClientI	*o = [self findIn: clients byObject: obj];
 
-  [self unregisterClient: o gracefully: YES];
+  if (o != nil)
-}
+    {
-
+      [self unregisterClient: o gracefully: YES];
-- (void) unregisterByName: (NSString*)n
+    }
-{
-  EcClientI	*o = [self findIn: clients byName: n];
-
-  [self unregisterClient: o gracefully: YES];
 }
 
 - (void) update

EcLogger.m

@@ -147,7 +147,7 @@ static NSArray          *modes;
   id<CmdLogger>	server;
 
   server = (id<CmdLogger>)[EcProc server: name];
-  [server registerClient: self name: cmdLogName()];
+  [server registerClient: self name: cmdLogName() transient: NO];
 }
 
 /* Should only be called on main thread, but doesn't matter.

EcProcess.h

@@ -157,9 +157,9 @@ typedef enum    {
 		      type: (EcLogType)t
 		       for: (id)o;
 - (bycopy NSData*) registerClient: (id)c
-			     name: (NSString*)n;
+			     name: (NSString*)n
+			transient: (BOOL)t;
 - (void) unregisterByObject: (id)obj;
-- (void) unregisterByName: (NSString*)n;
 @end
 
 @protocol	Console <NSObject>

EcProcess.m

@@ -241,6 +241,7 @@ static BOOL             configInProgress = NO;
 static BOOL		cmdFlagDaemon = NO;
 static BOOL		cmdFlagTesting = NO;
 static BOOL		cmdIsRunning = NO;
+static BOOL		cmdIsRegistered = NO;
 static BOOL		cmdKeepStderr = NO;
 static BOOL		cmdKillDebug = NO;
 static NSString		*cmdBase = nil;
@@ -1966,10 +1967,12 @@ findMode(NSDictionary* d, NSString* s)
       [connection invalidate];
     }
   DESTROY(cmdServer);
+  cmdIsRegistered = NO;
 }
 
 - (void) _connectionRegistered
 {
+  cmdIsRegistered = YES;
   [alarmDestination domanage: nil];
 }
 
@@ -3411,6 +3414,29 @@ NSLog(@"Ignored attempt to set timer interval to %g ... using 10.0", interval);
 		    rootProxyForConnectionWithRegisteredName: name
 							host: host
                                              usingNameServer: ns];
+		  [proxy setProtocolForProxy: @protocol(Command)];
+		  if (nil == proxy)
+		    {
+		      NSLog(@"Unable to connect to Command server");
+		    }
+		  else
+		    {
+		      NSConnection	*connection;
+
+		      ASSIGN(cmdServer, proxy);
+		      connection = [cmdServer connectionForProxy];
+		      [connection enableMultipleThreads];
+		      if (nil == alarmDestination)
+			{
+			  alarmDestination = [EcAlarmDestination new];
+			}
+		      [[self ecAlarmDestination] setDestination: cmdServer];
+		      [[NSNotificationCenter defaultCenter]
+			addObserver: self
+			   selector: @selector(cmdConnectionBecameInvalid:)
+			       name: NSConnectionDidDieNotification
+			     object: connection];
+		    }
 		}
 	      NS_HANDLER
 		{
@@ -3420,12 +3446,15 @@ NSLog(@"Ignored attempt to set timer interval to %g ... using 10.0", interval);
 		}
 	      NS_ENDHANDLER
 
-	      if (proxy != nil)
+	      /* We only register and fetch new configuration information
+	       * if we are NOT in the process of shutting down: a process
+	       * which is shutting down should only use the Command server
+	       * for logging/alerting purposes.
+	       */
+	      if (proxy != nil && 0.0 == beganQuitting)
 		{
 		  NSMutableDictionary	*r = nil;
 		  
-		  [proxy setProtocolForProxy: @protocol(Command)];
-
 		  NS_DURING
 		    {
 		      NSData	*d;
@@ -3466,26 +3495,11 @@ NSLog(@"Ignored attempt to set timer interval to %g ... using 10.0", interval);
 		    }
 		  else if ([r objectForKey: @"back-off"] != nil)
 		    {
-		      NSLog(@"Unable to connect to Command server ..."
+		      NSLog(@"Unable to register with Command server ..."
 			@" back-off");
 		    }
 		  else
 		    {
-		      NSConnection	*connection;
-
-		      cmdServer = [proxy retain];
-		      connection = [cmdServer connectionForProxy];
-		      [connection enableMultipleThreads];
-		      if (nil == alarmDestination)
-			{
-			  alarmDestination = [EcAlarmDestination new];
-			}
-		      [[self ecAlarmDestination] setDestination: cmdServer];
-		      [[NSNotificationCenter defaultCenter]
-			addObserver: self
-			   selector: @selector(cmdConnectionBecameInvalid:)
-			       name: NSConnectionDidDieNotification
-			     object: connection];
 		      [self _update: r];
 
                       /* If we just connected to the command server,
@@ -3499,10 +3513,6 @@ NSLog(@"Ignored attempt to set timer interval to %g ... using 10.0", interval);
                         }
 		    }
 		}
-	      else
-		{
-		  NSLog(@"Unable to connect to Command server ... not found");
-		}
 	    }
 	  connecting = NO;
 	  if (nil == cmdLast && nil == cmdServer && YES == [self cmdIsClient])
@@ -3532,7 +3542,11 @@ NSLog(@"Ignored attempt to set timer interval to %g ... using 10.0", interval);
     {
       NS_DURING
 	{
-	  [cmdServer unregisterByObject: self];
+	  if (cmdIsRegistered)
+	    {
+	      cmdIsRegistered = NO;
+	      [cmdServer unregisterByObject: self];
+	    }
 	}
       NS_HANDLER
 	{