void
startNewCoordinator(CoordinatorMode mode)
{
string host;
int port;
getCoordHostAndPort(mode, host, &port);
JASSERT(strcmp(host.c_str(), "localhost") == 0 ||
strcmp(host.c_str(), "127.0.0.1") == 0 ||
jalib::Filesystem::GetCurrentHostname() == host.c_str())
(host) (jalib::Filesystem::GetCurrentHostname())
.Text("Won't automatically start coordinator because DMTCP_HOST"
" is set to a remote host.");
// Create a socket and bind it to an unused port.
errno = 0;
jalib::JServerSocket coordinatorListenerSocket(jalib::JSockAddr::ANY,
port, 128);
JASSERT(coordinatorListenerSocket.isValid())
(coordinatorListenerSocket.port()) (JASSERT_ERRNO) (host) (port)
.Text("Failed to create socket to coordinator port."
"\nIf msg is \"Address already in use\","
" this may be an old coordinator."
"\nEither try again a few seconds or a minute later,"
"\nOr kill other coordinators on this host and port:"
"\n dmtcp_command ---coord-host XXX --coord-port XXX"
"\nOr specify --join-coordinator if joining existing computation.");
// Now dup the sockfd to
coordinatorListenerSocket.changeFd(PROTECTED_COORD_FD);
setCoordPort(coordinatorListenerSocket.port());
JTRACE("Starting a new coordinator automatically.")
(coordinatorListenerSocket.port());
if (fork() == 0) {
/* NOTE: This code assumes that dmtcp_launch (the current program)
* and dmtcp_coordinator are in the same directory. Namely,
* GetProgramDir() gets the dir of the current program (dmtcp_launch).
* Hence, if dmtcp_coordinator is in a different directory, then
* jalib::Filesystem::GetProgramDir() + "/dmtcp_coordinator"
* will not exist, and the child will fail.
*/
// We can't use Util::getPath() here since the SharedData has not been
// initialized yet.
string coordinator =
jalib::Filesystem::GetProgramDir() + "/dmtcp_coordinator";
char *modeStr = (char *)"--daemon";
char *args[] = {
(char *)coordinator.c_str(),
(char *)"--quiet",
/* If we wish to also suppress coordinator warnings, call --quiet twice */
(char *)"--exit-on-last",
modeStr,
NULL
};
execv(args[0], args);
JASSERT(false)(coordinator)(JASSERT_ERRNO).Text(
"exec(dmtcp_coordinator) failed");
} else {
int status;
_real_close(PROTECTED_COORD_FD);
JASSERT(wait(&status) > 0) (JASSERT_ERRNO);
}
}
void dmtcp::DmtcpCoordinatorAPI::startNewCoordinator(int modes, int isRestart)
{
int coordinatorStatus = -1;
//get location of coordinator
const char *coordinatorAddr = getenv ( ENV_VAR_NAME_HOST );
if(coordinatorAddr == NULL) coordinatorAddr = DEFAULT_HOST;
const char *coordinatorPortStr = getenv ( ENV_VAR_NAME_PORT );
dmtcp::string s = coordinatorAddr;
if(s != "localhost" && s != "127.0.0.1" &&
s != jalib::Filesystem::GetCurrentHostname()){
JASSERT(false)(s)(jalib::Filesystem::GetCurrentHostname())
.Text("Won't automatically start coordinator because DMTCP_HOST"
" is set to a remote host.");
_real_exit(DMTCP_FAIL_RC);
}
if ( modes & COORD_BATCH || modes & COORD_FORCE_NEW ) {
// Create a socket and bind it to an unused port.
jalib::JServerSocket coordinatorListenerSocket ( jalib::JSockAddr::ANY, 0 );
errno = 0;
JASSERT ( coordinatorListenerSocket.isValid() )
( coordinatorListenerSocket.port() ) ( JASSERT_ERRNO )
.Text ( "Failed to create listen socket."
"\nIf msg is \"Address already in use\", this may be an old coordinator."
"\nKill other coordinators and try again in a minute or so." );
// Now dup the sockfd to
coordinatorListenerSocket.changeFd(PROTECTED_COORD_FD);
dmtcp::string coordPort= jalib::XToString(coordinatorListenerSocket.port());
setenv ( ENV_VAR_NAME_PORT, coordPort.c_str(), 1 );
}
JTRACE("Starting a new coordinator automatically.") (coordinatorPortStr);
if(fork()==0){
dmtcp::string coordinator = jalib::Filesystem::FindHelperUtility("dmtcp_coordinator");
char *modeStr = (char *)"--background";
if ( modes & COORD_BATCH ) {
modeStr = (char *)"--batch";
}
char * args[] = {
(char*)coordinator.c_str(),
(char*)"--exit-on-last",
modeStr,
NULL
};
execv(args[0], args);
JASSERT(false)(coordinator)(JASSERT_ERRNO).Text("exec(dmtcp_coordinator) failed");
} else {
_real_close ( PROTECTED_COORD_FD );
}
errno = 0;
if ( modes & COORD_BATCH ) {
// FIXME: If running in batch Mode, we sleep here for 5 seconds to let
// the coordinator get started up. We need to fix this in future.
sleep(5);
} else {
JASSERT(wait(&coordinatorStatus)>0)(JASSERT_ERRNO);
JASSERT(WEXITSTATUS(coordinatorStatus) == 0)
.Text("Failed to start coordinator, port already in use. You may use a different port by running with \'-p 12345\'\n");
}
}
