bool dmtcp::DmtcpWorker::waitForStage2bCheckpoint()
{
waitForCoordinatorMsg ( "PEER_LOOKUP", DMT_DO_PEER_LOOKUP );
JTRACE ( "Looking up Socket Peers..." );
theTcpConnections.clear();
theCheckpointState->preCheckpointPeerLookup(theTcpConnections);
sendPeerLookupRequest(theTcpConnections);
JTRACE ( "Done Socket Peer Lookup" );
WorkerState::setCurrentState ( WorkerState::PEER_LOOKUP_COMPLETE );
{
dmtcp::DmtcpMessage msg;
msg.type = DMT_OK;
msg.state = WorkerState::currentState();
_coordinatorSocket << msg;
JTRACE ( "waiting for DRAIN/RESUME message" );
do {
msg.poison();
_coordinatorSocket >> msg;
msg.assertValid();
if ( msg.type == DMT_KILL_PEER ) {
JTRACE ( "Received KILL message from coordinator, exiting" );
_exit ( 0 );
}
JTRACE ( "received message" ) (msg.type );
if ( msg.type != DMT_UNKNOWN_PEER )
break;
JTRACE ("received DMT_UNKNOWN_PEER message") (msg.conId);
TcpConnection* con =
(TcpConnection*) &( ConnectionList::instance() [msg.conId] );
con->markExternal();
externalTcpConnections.push_back(msg.conId);
_waitingForExternalSocketsToClose = true;
} while ( msg.type == DMT_UNKNOWN_PEER );
JASSERT ( msg.type == DMT_DO_DRAIN || msg.type == DMT_DO_RESUME )
( msg.type );
ConnectionList& connections = ConnectionList::instance();
// Tcp Accept and Connect connection with PeerType UNKNOWN should be marked as INTERNAL
for ( ConnectionList::iterator i = connections.begin()
; i!= connections.end()
; ++i )
{
Connection* con = i->second;
if ( con->conType() == Connection::TCP ) {
TcpConnection* tcpCon = (TcpConnection *) con;
if ( (tcpCon->tcpType() == TcpConnection::TCP_ACCEPT ||
tcpCon->tcpType() == TcpConnection::TCP_CONNECT) &&
tcpCon->peerType() == TcpConnection::PEER_UNKNOWN )
tcpCon->markInternal();
}
}
if ( msg.type == DMT_DO_RESUME ) {
JTRACE ( "Peer Lookup not complete, skipping checkpointing \n\n\n\n\n");
return false;
}
JASSERT (msg.type == DMT_DO_DRAIN);
}
}
void dmtcp::ConnectionState::doReconnect(jalib::JSocket& coordinator,
jalib::JSocket& restoreListen)
{
_rewirer.addDataSocket(new jalib::JChunkReader(coordinator,
sizeof(DmtcpMessage)));
_rewirer.addListenSocket(restoreListen);
_rewirer.setCoordinatorFd(coordinator.sockfd());
handleDuplicateFilesInSeparateConnections();
ConnectionList& connections = ConnectionList::instance();
// Here we modify the restore algorithm by splitting it in two parts. In the
// first part we restore all the connection except the PTY_SLAVE types and in
// the second part we restore only PTY_SLAVE connections. This is done to
// make sure that by the time we are trying to restore a PTY_SLAVE
// connection, its corresponding PTY_MASTER connection has already been
// restored.
// UPDATE: We also restore the files for which the we didn't have the lock in
// second iteration along with PTY_SLAVEs
// Part 1: Restore all but Pseudo-terminal slaves and file connection which
// were not checkpointed
ConnectionList::iterator i;
for (i= connections.begin(); i != connections.end(); ++i) {
ConnectionIdentifier id = i->first;
Connection *con = i->second;
JASSERT(_conToFds[id].size() > 0)
.Text("stale connections should be gone by now");
if (con->subType() == FileConnection::FILE_PROCFS) {
continue;
}
if (con->conType() == Connection::TCP) {
TcpConnection *tcpCon =(TcpConnection *) con;
if (tcpCon->peerType() == TcpConnection::PEER_SOCKETPAIR) {
ConnectionIdentifier peerId = tcpCon->getSocketpairPeerId();
TcpConnection *peerCon =
(TcpConnection*) connections.getConnection(peerId);
if (peerCon != NULL) {
tcpCon->restoreSocketPair(_conToFds[id], peerCon, _conToFds[peerId]);
continue;
}
}
}
if (con->restoreInSecondIteration() == false) {
con->restore(_conToFds[id], &_rewirer);
}
}
// Part 2: Restore all Pseudo-terminal slaves and file connections that were
// not checkpointed.
for (i = connections.begin(); i != connections.end(); ++i) {
Connection *con = i->second;
JASSERT(_conToFds[i->first].size() > 0)
.Text("stale connections should be gone by now");
if (con->subType() == FileConnection::FILE_PROCFS) {
continue;
}
if (con->restoreInSecondIteration() == true) {
con->restore(_conToFds[i->first], &_rewirer);
}
}
_rewirer.doReconnect();
}