/*
* You should free the string once you've used it
*/
char * getPortFromFile( const char * fmt, ... ) {
char fname[PATH_MAX];
char dirname[PATH_MAX];
char *retPort;
char *cerr;
va_list ap;
FILE * fp;
int done = 0;
int count = 0; /* Just used for the NFS sync - not really a count */
retPort = (char * ) calloc( MPI_MAX_PORT_NAME + 1, sizeof( char ) );
va_start( ap, fmt );
vsnprintf( fname, PATH_MAX, fmt, ap );
va_end( ap );
srand( getpid() );
while( !done ) {
count += rand();
fp = fopen( fname, "rt" );
if( fp != NULL ) {
cerr = fgets( retPort, MPI_MAX_PORT_NAME, fp );
fclose( fp );
/* ignore bogus tag - assume that the real tag must be longer than 8
* characters */
if( strlen( retPort ) >= 8 ) {
done = 1;
}
}
if ( !done ) {
int retcode;
safeSleep( 0.1 );
/* force NFS to update by creating and then deleting a subdirectory. Ouch. */
snprintf( dirname, PATH_MAX, "%s___%d", fname, count );
retcode = mkdir( dirname, 0777 );
if( retcode != 0 ) {
perror( "Calling mkdir" );
_exit( 9 );
} else {
rmdir( dirname );
}
}
}
return retPort;
}
static void *threadLooper(void *x)
{
int runTime = 0;
size_t lastStrokeSeen = sWatchdogStrokeCount;
/* allow to run for up to 2 minutes */
while (runTime < sWatchdogTimeout) {
safeSleep(5);
runTime += 5;
if (lastStrokeSeen == sWatchdogStrokeCount) {
msg("Watchdog not stroked for a while, printing stack: \n");
printStackTrace();
}
lastStrokeSeen = sWatchdogStrokeCount;
}
msg("Watchdog about to abort with a timeout after %d\n", runTime);
_exit(20);
return NULL;
}
