failing to answer to a discover declare a coordinator dead in the master after some time

1 files changed, 20 insertions, 9 deletions

diff --git a/src/master.c b/src/master.c
index 53b71b8..3cd2173 100644
--- a/src/master.c
+++ b/src/master.c
@@ -16,14 +16,18 @@ static int master_must_terminate = 0;
 
 coordinator_t coordinator[MAX_COORDINATORS];
 
-static void reregister_coordinator( const char *host )
+static void recompute_coordinator_states( void )
 {
-	int pos;
-	for( pos = 0; pos < MAX_COORDINATORS; pos++ ) {
-		if( coordinator[pos].used &&
-			strcmp( coordinator[pos].host, host ) == 0 ) {
-				coordinator[pos].used = false;
-			return;
+	time_t now;
+	time( &now );
+	for( int pos = 0; pos < MAX_COORDINATORS; pos++ ) {
+		coordinator_t *c = &coordinator[pos];
+		if( c->used ) {
+			double d = difftime( now, c->lastAlive );
+			fprintf( stderr, "%s %lld %lld %g\n", c->host, (long long)c->lastAlive, (long long)now, d );
+			if( difftime( now, c->lastAlive ) > MAX_COORDINATOR_AGE ) {
+				c->alive = false;
+			}
 		}
 	}
 }
@@ -48,8 +52,11 @@ static void register_coordinator( json_object *obj )
 	
 	int pos;
 	for( pos = 0; pos < MAX_COORDINATORS; pos++ ) {
-		if( coordinator[pos].used &&
-			strcmp( coordinator[pos].host, host ) == 0 ) {
+		coordinator_t *coord = &coordinator[pos];
+		if( coord->used &&
+			strcmp( coord->host, host ) == 0 ) {
+				time( &coord->lastAlive );
+				coord->alive = true;
 				fprintf( stderr, "master: already registered coordinator from host '%s' (%d)\n", host, pos );
 			return;
 		}
@@ -69,6 +76,8 @@ static void register_coordinator( json_object *obj )
 	coord->os = strdup( os );
 	coord->arch = strdup( arch );
 	coord->cpus = cpus;
+	coord->alive = true;
+	time( &coord->lastAlive );
 	coord->used = true;
 }
 
@@ -116,6 +125,8 @@ NEXT_DISCOVER:
 		
 		printf( "master idle: %d\n", master_must_terminate );
 		
+		recompute_coordinator_states( );
+		
 		char *answer = NULL;
 		bytes = nn_recv( master_sock, &answer, NN_MSG, 0 );
 		if( master_must_terminate ) continue;