LupusNobilis
/
sogo
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

Monotone-Parent: 263b6e0123652c8010371f69903e3a605a0c7a96 

Monotone-Revision: 6c845252270eb03967ac431437dea26441d69532

Monotone-Author: wsourdeau@inverse.ca
Monotone-Date: 2009-12-09T21:22:18
Monotone-Branch: ca.inverse.sogo
maint-2.0.2
Wolfgang Sourdeau 2009-12-09 21:22:18 +00:00
parent 380f095815
commit 66eee3f043
1 changed files with 160 additions and 103 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

263
SOPE/sope-patchset-r1660.diff
View File

@ -4579,7 +4579,7 @@ Index: sope-appserver/NGObjWeb/WOWatchDogApplicationMain.m
#if defined(__CYGWIN32__) || defined(__MINGW32__)
int WOWatchDogApplicationMain
@@ -39,201 +60,845 @@
@@ -39,201 +60,894 @@
#include <time.h>
#include <string.h>
@ -4627,6 +4627,7 @@ Index: sope-appserver/NGObjWeb/WOWatchDogApplicationMain.m
+@interface WOWatchDogChild : NSObject <RunLoopEvents>
+{
+ int pid;
+ int SIGCHLDStatus;
+ int counter;
+ NGActiveSocket *controlSocket;
+ WOChildStatus status;
@ -4640,6 +4641,10 @@ Index: sope-appserver/NGObjWeb/WOWatchDogApplicationMain.m
+- (void) setPid: (int) newPid;
+- (int) pid;
+
+- (void) setSIGCHLDStatus: (int) newSIGCHLDStatus;
+- (int) SIGCHLDStatus;
+- (void) handleSIGCHLDStatus;
+
+- (void) setControlSocket: (NGActiveSocket *) newSocket;
+- (NGActiveSocket *) controlSocket;
+
@ -4666,6 +4671,9 @@ Index: sope-appserver/NGObjWeb/WOWatchDogApplicationMain.m
+
+ BOOL terminate;
+ BOOL willTerminate;
+ NSNumber *terminationSignal;
+ int pendingSIGCHLD;
+ int pendingSIGHUP;
+
+ NGPassiveSocket *listeningSocket;
+
@ -4704,6 +4712,7 @@ Index: sope-appserver/NGObjWeb/WOWatchDogApplicationMain.m
+ {
+ pid = -1;
+ controlSocket = nil;
+ SIGCHLDStatus = -1;
+ status = WOChildStatusDown;
+ counter = 0;
+ lastSpawn = nil;
@ -4731,16 +4740,7 @@ Index: sope-appserver/NGObjWeb/WOWatchDogApplicationMain.m
+
+- (void) dealloc
+{
+ NSRunLoop *runLoop;
+
+ if (controlSocket) {
+ runLoop = [NSRunLoop currentRunLoop];
+ [runLoop removeEvent: (void *) [controlSocket fileDescriptor]
+ type: ET_RDESC
+ forMode: NSDefaultRunLoopMode
+ all: YES];
+ [controlSocket release];
+ }
+ [self setControlSocket: nil];
+ [lastSpawn release];
+ [super dealloc];
+}
@ -4760,6 +4760,35 @@ Index: sope-appserver/NGObjWeb/WOWatchDogApplicationMain.m
+ return pid;
+}
+
+- (void) setSIGCHLDStatus: (int) newSIGCHLDStatus
+{
+ SIGCHLDStatus = newSIGCHLDStatus;
+}
+
+- (int) SIGCHLDStatus
+{
+ return SIGCHLDStatus;
+}
+
+- (void) handleSIGCHLDStatus
+{
+ int code;
+
+ [self logWithFormat: @"received SIGCHLD from pid %d", pid];
+ code = WEXITSTATUS (SIGCHLDStatus);
+ if (code != 0)
+ [self logWithFormat: @"child %d exited with code %i", pid, code];
+ if (WIFSIGNALED (SIGCHLDStatus))
+ [self logWithFormat: @" (terminated due to signal %i%@)",
+ WTERMSIG (SIGCHLDStatus),
+ WCOREDUMP (SIGCHLDStatus) ? @", coredump" : @""];
+ if (WIFSTOPPED (SIGCHLDStatus))
+ [self logWithFormat: @" (stopped due to signal %i)", WSTOPSIG (SIGCHLDStatus)];
+ SIGCHLDStatus = -1;
+ [self setStatus: WOChildStatusDown];
+ [self setControlSocket: nil];
+}
+
+- (void) setControlSocket: (NGActiveSocket *) newSocket
+{
+ NSRunLoop *runLoop;
@ -4831,6 +4860,7 @@ Index: sope-appserver/NGObjWeb/WOWatchDogApplicationMain.m
+ count: sizeof (WOChildMessage)] == NGStreamError) {
+ rc = NO;
+ [self errorWithFormat: @"FAILURE receiving status for child %d", pid];
+ [self errorWithFormat: @" socket: %@", controlSocket];
}
+ else {
+ rc = YES;
@ -4845,7 +4875,6 @@ Index: sope-appserver/NGObjWeb/WOWatchDogApplicationMain.m
+ // status = WOChildStatusDown;
+ // [watchDog declareChildDown: self];
+ // }
+ // NSLog (@"message read status (%d):", pid);
+ // [self logStatus];
+ }
+
@ -4870,7 +4899,7 @@ Index: sope-appserver/NGObjWeb/WOWatchDogApplicationMain.m
+- (void) _killKill
+{
+ if (status != WOChildStatusDown) {
+ [self warnWithFormat: @"sending KILL signal to child %d", pid];
+ [self warnWithFormat: @"sending KILL signal to pid %d", pid];
+ kill (pid, SIGKILL);
+ }
+}
@ -4878,7 +4907,7 @@ Index: sope-appserver/NGObjWeb/WOWatchDogApplicationMain.m
+- (void) _kill
+{
+ if (status != WOChildStatusDown) {
+ [self logWithFormat: @"sending terminate signal to child %d", pid];
+ [self logWithFormat: @"sending terminate signal to pid %d", pid];
+ status = WOChildStatusTerminating;
+ kill (pid, SIGTERM);
+ /* We hardcode a 5 minutes delay before ensuring that all children are
@ -4906,8 +4935,6 @@ Index: sope-appserver/NGObjWeb/WOWatchDogApplicationMain.m
+
+- (void) terminate
+{
+ WOChildMessage message;
+
+ if (status == WOChildStatusDown) {
+ [self logWithFormat: @"child is already down"];
+ } else {
@ -4916,23 +4943,22 @@ Index: sope-appserver/NGObjWeb/WOWatchDogApplicationMain.m
+ }
+}
+
+- (void) delayedTerminate
+{
+ // NSLog (@"delayed terminate on child %d (wait: %d)", pid,
+ // waitpid(pid, NULL, WNOHANG));
+ [NSTimer scheduledTimerWithTimeInterval: 0.1
+ target: self
+ selector: @selector (terminate)
+ userInfo: nil
+ repeats: NO];
+}
+
+- (void) receivedEvent: (void*)data
+ type: (RunLoopEventType)type
+ extra: (void*)extra
+ forMode: (NSString*)mode
+{
+ [self readMessage];
+ if ([controlSocket isAlive])
+ [self readMessage];
+ else {
+ /* This happens when a socket has been closed by the child but the child
+ has not terminated yet. */
+ [[NSRunLoop currentRunLoop] removeEvent: data
+ type: ET_RDESC
+ forMode: NSDefaultRunLoopMode
+ all: YES];
+ [self setControlSocket: nil];
+ }
+}
+
+@end
@ -4953,9 +4979,12 @@ Index: sope-appserver/NGObjWeb/WOWatchDogApplicationMain.m
+{
+ if ((self = [super init]))
+ {
+ listeningSocket = NULL;
+ listeningSocket = nil;
+ terminate = NO;
+ willTerminate = NO;
+ terminationSignal = nil;
+ pendingSIGCHLD = 0;
+ pendingSIGHUP = 0;
+
+ numberOfChildren = 0;
+ children = [[NSMutableArray alloc] initWithCapacity: 10];
@ -4986,6 +5015,7 @@ Index: sope-appserver/NGObjWeb/WOWatchDogApplicationMain.m
+- (void) dealloc
+{
+ [self _releaseListeningSocket];
+ [terminationSignal release];
+ [appName release];
+ [children release];
+ [super dealloc];
@ -5030,7 +5060,6 @@ Index: sope-appserver/NGObjWeb/WOWatchDogApplicationMain.m
+- (void) _cleanupSignalAndEventHandlers
+{
+ int count;
+ NGActiveSocket *controlSocket;
+ NSRunLoop *runLoop;
+
+ [[UnixSignalHandler sharedHandler] removeObserver: self];
@ -5042,12 +5071,13 @@ Index: sope-appserver/NGObjWeb/WOWatchDogApplicationMain.m
+ all: YES];
+
+ for (count = 0; count < numberOfChildren; count++) {
+ controlSocket = [[children objectAtIndex: count] controlSocket];
+ if (controlSocket)
+ [runLoop removeEvent: (void *) [controlSocket fileDescriptor]
+ type: ET_RDESC
+ forMode: NSDefaultRunLoopMode
+ all: YES];
+ [[children objectAtIndex: count] setControlSocket: nil];
+ // controlSocket = [[children objectAtIndex: count] controlSocket];
+ // if (controlSocket)
+ // [runLoop removeEvent: (void *) [controlSocket fileDescriptor]
+ // type: ET_RDESC
+ // forMode: NSDefaultRunLoopMode
+ // all: YES];
}
}
@ -5375,59 +5405,23 @@ Index: sope-appserver/NGObjWeb/WOWatchDogApplicationMain.m
+- (void) _handleSIGCHLD:(NSNumber *)_signal {
+ WOWatchDogChild *child;
+ pid_t childPid;
+ int status, code;
+ int status;
+
+ childPid = wait (&status);
+ [self logWithFormat: @"received SIGCHLD: %d", childPid];
+ if (childPid > -1) {
+ code = WEXITSTATUS(status);
+ if (code != 0)
+ [self logWithFormat: @"child %d exited with code %i", childPid, code];
+ if (WIFSIGNALED(status))
+ [self logWithFormat: @" (terminated due to signal %i%@)",
+ WTERMSIG(status),
+ WCOREDUMP(status) ? @", coredump" : @""];
+ if (WIFSTOPPED(status))
+ [self logWithFormat: @" (stopped due to signal %i)", WSTOPSIG(status)];
+ pendingSIGCHLD++;
+ child = [self _childWithPID: childPid];
+ if (child) {
+ [child setStatus: WOChildStatusDown];
+ [self declareChildDown: child];
+ [child setControlSocket: nil];
+ if (willTerminate && [downChildren count] == numberOfChildren) {
+ [self logWithFormat: @"all child exited"];
+ terminate = YES;
+ }
+ }
+ [child setSIGCHLDStatus: status];
+ }
+ else
+ [self errorWithFormat: @"no pid received"];
+}
+
+- (void) _handleTermination:(NSNumber *)_signal {
+ WOWatchDogChild *child;
+ int count, max;
+
+ if (!willTerminate) {
+ [self logWithFormat: @"Terminating with signal %@", _signal];
+ [self _releaseListeningSocket];
+ willTerminate = YES;
+ max = [children count];
+ for (count = 0; count < max; count++) {
+ child = [children objectAtIndex: count];
+ if ([child status] != WOChildStatusDown
+ && [child status] != WOChildStatusTerminating)
+ [child delayedTerminate];
+ }
+ }
+ if (!terminationSignal)
+ ASSIGN (terminationSignal, _signal);
+}
+
+- (void) _handleSIGHUP:(NSNumber *)_signal {
+ [NSTimer scheduledTimerWithTimeInterval: 1.0
+ target: self
+ selector: @selector (_ensureWorkersCount)
+ userInfo: nil
+ repeats: NO];
+ pendingSIGHUP++;
+}
+
+- (void) _setupSignals
@ -5478,21 +5472,63 @@ Index: sope-appserver/NGObjWeb/WOWatchDogApplicationMain.m
+ if (newNumberOfChildren < 1)
+ newNumberOfChildren = 1;
+ numberOfChildren = newNumberOfChildren;
+ [NSTimer scheduledTimerWithTimeInterval: 0.1
+ target: self
+ selector: @selector (_noop)
+ userInfo: nil
+ repeats: NO];
+}
+
+/* WOPidFile, WOPort */
+- (void) _processArguments
+- (void) _handlePostTerminationSignal
+{
+ NSArray *arguments;
+ NSProcessInfo *processInfo;
+ WOWatchDogChild *child;
+ int count;
+
+ [self logWithFormat: @"Terminating with signal %@", terminationSignal];
+ [self _releaseListeningSocket];
+ for (count = 0; count < numberOfChildren; count++) {
+ child = [children objectAtIndex: count];
+ if ([child status] != WOChildStatusDown
+ && [child status] != WOChildStatusTerminating)
+ [child terminate];
+ }
+ [terminationSignal release];
+ terminationSignal = nil;
+ if ([downChildren count] == numberOfChildren) {
+ [self logWithFormat: @"all children exited. We now terminate."];
+ terminate = YES;
+ }
+ else
+ willTerminate = YES;
+}
+
+- (void) _handlePostSIGCHLDStatus
+{
+ int status, count;
+ WOWatchDogChild *child;
+
+ for (count = 0; pendingSIGCHLD && count < numberOfChildren; count++) {
+ child = [children objectAtIndex: count];
+ status = [child SIGCHLDStatus];
+ if (status != -1) {
+ [child handleSIGCHLDStatus];
+ pendingSIGCHLD--;
+ [self declareChildDown: child];
+ if (willTerminate && [downChildren count] == numberOfChildren) {
+ [self logWithFormat: @"all children exited. We now terminate."];
+ terminate = YES;
+ }
+ }
+ }
+}
+
+- (void) _setupProcessName
+{
+ NSProcessInfo *processInfo;
+ NSString *name;
+
+ /* this does not seem to work */
+ processInfo = [NSProcessInfo processInfo];
+ arguments = [processInfo arguments];
+ name = [processInfo processName];
+ if (!name)
+ name = @"";
+ [processInfo setProcessName: [NSString stringWithFormat: @"%@: %@ watchdog",
+ name, appName]];
+}
+
+- (int) run: (NSString *) newAppName
@ -5507,11 +5543,11 @@ Index: sope-appserver/NGObjWeb/WOWatchDogApplicationMain.m
+ willTerminate = NO;
+
+ ASSIGN (appName, newAppName);
+ [self _setupProcessName];
+
+ argc = newArgC;
+ argv = newArgV;
+
+ [self _processArguments];
+
+ listening = NO;
+ retries = 0;
+ while (!listening && retries < 5) {
@ -5523,16 +5559,22 @@ Index: sope-appserver/NGObjWeb/WOWatchDogApplicationMain.m
+ }
+ }
+ if (listening) {
+ [self logWithFormat: @"watchdog process pid: %d", getpid ()];
+ [self _setupSignals];
+ [self _ensureWorkersCount];
+
+ // NSLog (@"ready to process requests");
+ runLoop = [NSRunLoop currentRunLoop];
+ terminate = NO;
+
+ while (!terminate) {
+ pool = [NSAutoreleasePool new];
+
+ while (pendingSIGHUP) {
+ [self logWithFormat: @"received SIGHUP"];
+ [self _ensureWorkersCount];
+ pendingSIGHUP--;
+ }
+
+ // [self logWithFormat: @"watchdog loop"];
+ NS_DURING {
+ terminate = [self _ensureChildren];
@ -5547,6 +5589,13 @@ Index: sope-appserver/NGObjWeb/WOWatchDogApplicationMain.m
+ @"an exception occured in runloop %@", localException];
+ }
+ NS_ENDHANDLER;
+
+ if (!terminate) {
+ if (terminationSignal)
+ [self _handlePostTerminationSignal];
+ while (pendingSIGCHLD)
+ [self _handlePostSIGCHLDStatus];
+ }
+ [pool release];
+ }
+
@ -5586,7 +5635,7 @@ Index: sope-appserver/NGObjWeb/WOWatchDogApplicationMain.m
#if LIB_FOUNDATION_LIBRARY || defined(GS_PASS_ARGUMENTS)
{
extern char **environ;
@@ -241,179 +906,67 @@
@@ -241,179 +955,67 @@
environment:(void*)environ];
}
#endif
@ -5820,7 +5869,7 @@ Index: sope-appserver/NGObjWeb/WOWatchDogApplicationMain.m
}
#endif
@@ -421,8 +974,8 @@
@@ -421,8 +1023,8 @@
@interface NSUserDefaults(ServerDefaults)
+ (id)hackInServerDefaults:(NSUserDefaults *)_ud
@ -5831,7 +5880,7 @@ Index: sope-appserver/NGObjWeb/WOWatchDogApplicationMain.m
@end
int WOWatchDogApplicationMainWithServerDefaults
@@ -437,7 +990,7 @@
@@ -437,7 +1039,7 @@
{
extern char **environ;
[NSProcessInfo initializeWithArguments:(void*)argv count:argc
@ -5840,7 +5889,7 @@ Index: sope-appserver/NGObjWeb/WOWatchDogApplicationMain.m
}
#endif
@@ -446,8 +999,8 @@
@@ -446,8 +1048,8 @@
ud = [NSUserDefaults standardUserDefaults];
sd = [defClass hackInServerDefaults:ud
@ -5900,7 +5949,21 @@ Index: sope-appserver/NGObjWeb/ChangeLog
===================================================================
--- sope-appserver/NGObjWeb/ChangeLog (revision 1660)
+++ sope-appserver/NGObjWeb/ChangeLog (working copy)
@@ -1,3 +1,71 @@
@@ -1,3 +1,85 @@
+2009-12-09 Wolfgang Sourdeau <wsourdeau@inverse.ca>
+
+ * WOWatchDogApplicationMain.m (_handleSIGCHLD:)
+ (_handleTermination:, _handleSIGHUP:): the actual handling is now
+ done elsewhere, in order to avoid messing with memory allocation
+ and risking a dead lock.
+ (-_handlePostTerminationSignal): we set "terminate" to YES if all
+ children are already down, in order to avoid another deadlock
+ where the process termination would stall waiting for SIGCHLD.
+ (-receivedEvent:type:extra:forMode:): check that the control
+ socket is still "alive" before reading from it. If not, we
+ unregister the filedescriptor passed as "data" from the runloop
+ listener.
+
+2009-12-07 Wolfgang Sourdeau <wsourdeau@inverse.ca>
+
+ * WOCoreApplication.m (+initialize): we invoke
@ -6870,7 +6933,7 @@ Index: sope-appserver/NGObjWeb/WOHttpAdaptor/WOHttpAdaptor.m
[(WORunLoop *)[WORunLoop currentRunLoop]
removeFileObject:self->socket forMode:NSDefaultRunLoopMode];
[[NSNotificationCenter defaultCenter] removeObserver:self];
@@ -603,52 +522,97 @@
@@ -603,52 +522,91 @@
return _connection;
}
@ -6928,19 +6991,13 @@ Index: sope-appserver/NGObjWeb/WOHttpAdaptor/WOHttpAdaptor.m
+ NSLog (@"renotifying watchdog");
+ if (message == WOChildMessageAccept) {
+ pool = [NSAutoreleasePool new];
+ // NSLog (@" accepting");
+ connection = [self _accept];
+ // NSLog (@" AAAAAAAAAAAAAA accepted. Handling request.");
+ if ([controlSocket safeWriteBytes: &message
+ count: sizeof (WOChildMessage)])
+ ;
+ // NSLog (@" 1 notified watchdog");
+ [self _handleConnection: connection];
+ message = WOChildMessageReady;
+ // NSLog (@" BBBBBBBBBBBBBB Request handled. Notify watchdog.");
+ [controlSocket safeWriteBytes: &message count: sizeof (WOChildMessage)];
+ // NSLog (@" 2 notified watchdog");
+ // NSLog (@" CCCCCCCCCCCCCC done.");
+ [pool release];
+ }
+ else if (message == WOChildMessageShutdown) {