static bool RegisterSelf(SocketEndpoint &masterEp)
{
StringBuffer slfStr;
StringBuffer masterStr;
LOG(MCdebugProgress, thorJob, "registering %s - master %s",slfEp.getUrlStr(slfStr).str(),masterEp.getUrlStr(masterStr).str());
try
{
SocketEndpoint ep = masterEp;
ep.port = getFixedPort(getMasterPortBase(), TPORT_mp);
Owned<INode> masterNode = createINode(ep);
CMessageBuffer msg;
if (!queryWorldCommunicator().recv(msg, masterNode, MPTAG_THORREGISTRATION))
return false;
PROGLOG("Initialization received");
unsigned vmajor, vminor;
msg.read(vmajor);
msg.read(vminor);
if (vmajor != THOR_VERSION_MAJOR || vminor != THOR_VERSION_MINOR)
{
replyError(TE_FailedToRegisterSlave, "Thor master/slave version mismatch");
return false;
}
Owned<IGroup> rawGroup = deserializeIGroup(msg);
globals->Release();
globals = createPTree(msg);
mergeCmdParams(globals); // cmd line
unsigned slavesPerNode = globals->getPropInt("@slavesPerNode", 1);
unsigned channelsPerSlave = globals->getPropInt("@channelsPerSlave", 1);
unsigned localThorPortInc = globals->getPropInt("@localThorPortInc", DEFAULT_SLAVEPORTINC);
unsigned slaveBasePort = globals->getPropInt("@slaveport", DEFAULT_THORSLAVEPORT);
setClusterGroup(masterNode, rawGroup, slavesPerNode, channelsPerSlave, slaveBasePort, localThorPortInc);
unsigned numStrands, blockSize;
if (globals->hasProp("Debug/@forceNumStrands"))
numStrands = globals->getPropInt("Debug/@forceNumStrands");
else
{
numStrands = defaultForceNumStrands;
globals->setPropInt("Debug/@forceNumStrands", defaultForceNumStrands);
}
if (globals->hasProp("Debug/@strandBlockSize"))
blockSize = globals->getPropInt("Debug/@strandBlockSize");
else
{
blockSize = defaultStrandBlockSize;
globals->setPropInt("Debug/@strandBlockSize", defaultStrandBlockSize);
}
PROGLOG("Strand defaults: numStrands=%u, blockSize=%u", numStrands, blockSize);
const char *_masterBuildTag = globals->queryProp("@masterBuildTag");
const char *masterBuildTag = _masterBuildTag?_masterBuildTag:"no build tag";
PROGLOG("Master build: %s", masterBuildTag);
if (!_masterBuildTag || 0 != strcmp(BUILD_TAG, _masterBuildTag))
{
StringBuffer errStr("Thor master/slave build mismatch, master = ");
errStr.append(masterBuildTag).append(", slave = ").append(BUILD_TAG);
ERRLOG("%s", errStr.str());
#ifndef _DEBUG
replyError(TE_FailedToRegisterSlave, errStr.str());
return false;
#endif
}
msg.read((unsigned &)masterSlaveMpTag);
msg.clear();
msg.setReplyTag(MPTAG_THORREGISTRATION);
if (!queryNodeComm().reply(msg))
return false;
PROGLOG("Registration confirmation sent");
if (!queryNodeComm().recv(msg, 0, MPTAG_THORREGISTRATION)) // when all registered
return false;
::masterNode = LINK(masterNode);
PROGLOG("verifying mp connection to rest of cluster");
if (!queryNodeComm().verifyAll())
ERRLOG("Failed to connect to all nodes");
else
PROGLOG("verified mp connection to rest of cluster");
LOG(MCdebugProgress, thorJob, "registered %s",slfStr.str());
}
catch (IException *e)
{
FLLOG(MCexception(e), thorJob, e,"slave registration error");
e->Release();
return false;
}
return true;
}
static bool RegisterSelf(SocketEndpoint &masterEp)
{
StringBuffer slfStr;
StringBuffer masterStr;
LOG(MCdebugProgress, thorJob, "registering %s - master %s",slfEp.getUrlStr(slfStr).toCharArray(),masterEp.getUrlStr(masterStr).toCharArray());
try
{
SocketEndpoint ep = masterEp;
ep.port = getFixedPort(getMasterPortBase(), TPORT_mp);
Owned<INode> masterNode = createINode(ep);
CMessageBuffer msg;
if (!queryWorldCommunicator().recv(msg, masterNode, MPTAG_THORREGISTRATION))
return false;
PROGLOG("Initialization received");
unsigned vmajor, vminor;
msg.read(vmajor);
msg.read(vminor);
if (vmajor != THOR_VERSION_MAJOR || vminor != THOR_VERSION_MINOR)
{
replyError("Thor master/slave version mismatch");
return false;
}
Owned<IGroup> group = deserializeIGroup(msg);
setClusterGroup(group);
SocketEndpoint myEp = queryMyNode()->endpoint();
rank_t groupPos = group->rank(queryMyNode());
if (RANK_NULL == groupPos)
{
replyError("Node not part of thorgroup");
return false;
}
if (globals->hasProp("@SLAVENUM") && (mySlaveNum != (unsigned)groupPos))
{
VStringBuffer errStr("Slave group rank[%d] does not match provided cmd line slaveNum[%d]", mySlaveNum, (unsigned)groupPos);
replyError(errStr.str());
return false;
}
globals->Release();
globals = createPTree(msg);
mergeCmdParams(globals); // cmd line
const char *_masterBuildTag = globals->queryProp("@masterBuildTag");
const char *masterBuildTag = _masterBuildTag?_masterBuildTag:"no build tag";
PROGLOG("Master build: %s", masterBuildTag);
#ifndef _DEBUG
if (!_masterBuildTag || 0 != strcmp(BUILD_TAG, _masterBuildTag))
{
StringBuffer errStr("Thor master/slave build mismatch, master = ");
replyError(errStr.append(masterBuildTag).append(", slave = ").append(BUILD_TAG).str());
return false;
}
#endif
msg.read((unsigned &)masterSlaveMpTag);
msg.clear();
msg.setReplyTag(MPTAG_THORREGISTRATION);
if (!queryClusterComm().reply(msg))
return false;
PROGLOG("Registration confirmation sent");
if (!queryClusterComm().recv(msg, 0, MPTAG_THORREGISTRATION)) // when all registered
return false;
PROGLOG("verifying mp connection to rest of cluster");
if (!queryClusterComm().verifyAll())
ERRLOG("Failed to connect to all nodes");
else
PROGLOG("verified mp connection to rest of cluster");
::masterNode = LINK(masterNode);
LOG(MCdebugProgress, thorJob, "registered %s",slfStr.toCharArray());
}
catch (IException *e)
{
FLLOG(MCexception(e), thorJob, e,"slave registration error");
e->Release();
return false;
}
return true;
}
int main( int argc, char *argv[] )
{
#if defined(WIN32) && defined(_DEBUG)
int tmpFlag = _CrtSetDbgFlag( _CRTDBG_REPORT_FLAG );
tmpFlag |= _CRTDBG_LEAK_CHECK_DF;
_CrtSetDbgFlag( tmpFlag );
#endif
InitModuleObjects();
addAbortHandler(ControlHandler);
EnableSEHtoExceptionMapping();
dummyProc();
#ifndef __64BIT__
// Restrict stack sizes on 32-bit systems
Thread::setDefaultStackSize(0x10000); // NB under windows requires linker setting (/stack:)
#endif
#ifdef _WIN32
Owned<CReleaseMutex> globalNamedMutex;
#endif
if (globals)
globals->Release();
{
Owned<IFile> iFile = createIFile("thor.xml");
globals = iFile->exists() ? createPTree(*iFile, ipt_caseInsensitive) : createPTree("Thor", ipt_caseInsensitive);
}
unsigned multiThorMemoryThreshold = 0;
Owned<IException> unregisterException;
try
{
if (argc==1)
{
usage();
return 1;
}
cmdArgs = argv+1;
mergeCmdParams(globals);
cmdArgs = argv+1;
const char *master = globals->queryProp("@MASTER");
if (!master)
usage();
const char *slave = globals->queryProp("@SLAVE");
if (slave)
{
slfEp.set(slave);
localHostToNIC(slfEp);
}
else
slfEp.setLocalHost(0);
mySlaveNum = globals->getPropInt("@SLAVENUM");
setMachinePortBase(slfEp.port);
slfEp.port = getMachinePortBase();
startSlaveLog();
setSlaveAffinity(globals->getPropInt("@SLAVEPROCESSNUM"));
startMPServer(getFixedPort(TPORT_mp));
#ifdef USE_MP_LOG
startLogMsgParentReceiver();
LOG(MCdebugProgress, thorJob, "MPServer started on port %d", getFixedPort(TPORT_mp));
#endif
SocketEndpoint masterEp(master);
localHostToNIC(masterEp);
setMasterPortBase(masterEp.port);
markNodeCentral(masterEp);
if (RegisterSelf(masterEp))
{
if (globals->getPropBool("Debug/@slaveDaliClient"))
enableThorSlaveAsDaliClient();
IDaFileSrvHook *daFileSrvHook = queryDaFileSrvHook();
if (daFileSrvHook) // probably always installed
daFileSrvHook->addFilters(globals->queryPropTree("NAS"), &slfEp);
StringBuffer thorPath;
globals->getProp("@thorPath", thorPath);
recursiveCreateDirectory(thorPath.str());
int err = _chdir(thorPath.str());
if (err)
{
IException *e = makeErrnoExceptionV(-1, "Failed to change dir to '%s'", thorPath.str());
FLLOG(MCexception(e), thorJob, e);
throw e;
}
// Initialization from globals
setIORetryCount(globals->getPropInt("Debug/@ioRetries")); // default == 0 == off
StringBuffer str;
if (globals->getProp("@externalProgDir", str.clear()))
_mkdir(str.str());
else
globals->setProp("@externalProgDir", thorPath);
const char * overrideBaseDirectory = globals->queryProp("@thorDataDirectory");
const char * overrideReplicateDirectory = globals->queryProp("@thorReplicateDirectory");
StringBuffer datadir;
StringBuffer repdir;
if (getConfigurationDirectory(globals->queryPropTree("Directories"),"data","thor",globals->queryProp("@name"),datadir))
overrideBaseDirectory = datadir.str();
if (getConfigurationDirectory(globals->queryPropTree("Directories"),"mirror","thor",globals->queryProp("@name"),repdir))
overrideReplicateDirectory = repdir.str();
if (overrideBaseDirectory&&*overrideBaseDirectory)
setBaseDirectory(overrideBaseDirectory, false);
if (overrideReplicateDirectory&&*overrideBaseDirectory)
setBaseDirectory(overrideReplicateDirectory, true);
StringBuffer tempDirStr;
if (getConfigurationDirectory(globals->queryPropTree("Directories"),"temp","thor",globals->queryProp("@name"), tempDirStr))
globals->setProp("@thorTempDirectory", tempDirStr.str());
else
tempDirStr.append(globals->queryProp("@thorTempDirectory"));
addPathSepChar(tempDirStr).append(getMachinePortBase());
logDiskSpace(); // Log before temp space is cleared
SetTempDir(tempDirStr.str(), "thtmp", true);
useMemoryMappedRead(globals->getPropBool("@useMemoryMappedRead"));
LOG(MCdebugProgress, thorJob, "ThorSlave Version LCR - %d.%d started",THOR_VERSION_MAJOR,THOR_VERSION_MINOR);
StringBuffer url;
LOG(MCdebugProgress, thorJob, "Slave %s - temporary dir set to : %s", slfEp.getUrlStr(url).str(), queryTempDir());
#ifdef _WIN32
ULARGE_INTEGER userfree;
ULARGE_INTEGER total;
ULARGE_INTEGER free;
if (GetDiskFreeSpaceEx("c:\\",&userfree,&total,&free)&&total.QuadPart) {
unsigned pc = (unsigned)(free.QuadPart*100/total.QuadPart);
LOG(MCdebugProgress, thorJob, "Total disk space = %" I64F "d k", total.QuadPart/1000);
LOG(MCdebugProgress, thorJob, "Free disk space = %" I64F "d k", free.QuadPart/1000);
LOG(MCdebugProgress, thorJob, "%d%% disk free\n",pc);
}
#endif
if (getConfigurationDirectory(globals->queryPropTree("Directories"),"query","thor",globals->queryProp("@name"),str.clear()))
globals->setProp("@query_so_dir", str.str());
else
globals->getProp("@query_so_dir", str.clear());
if (str.length())
{
if (globals->getPropBool("Debug/@dllsToSlaves", true))
{
StringBuffer uniqSoPath;
if (PATHSEPCHAR == str.charAt(str.length()-1))
uniqSoPath.append(str.length()-1, str.str());
else
uniqSoPath.append(str);
uniqSoPath.append("_").append(getMachinePortBase());
str.swapWith(uniqSoPath);
globals->setProp("@query_so_dir", str.str());
}
PROGLOG("Using querySo directory: %s", str.str());
recursiveCreateDirectory(str.str());
}
multiThorMemoryThreshold = globals->getPropInt("@multiThorMemoryThreshold")*0x100000;
if (multiThorMemoryThreshold) {
StringBuffer lgname;
if (!globals->getProp("@multiThorResourceGroup",lgname))
globals->getProp("@nodeGroup",lgname);
if (lgname.length()) {
Owned<ILargeMemLimitNotify> notify = createMultiThorResourceMutex(lgname.str());
setMultiThorMemoryNotify(multiThorMemoryThreshold,notify);
PROGLOG("Multi-Thor resource limit for %s set to %" I64F "d",lgname.str(),(__int64)multiThorMemoryThreshold);
}
else
multiThorMemoryThreshold = 0;
}
slaveMain(jobListenerStopped);
}
LOG(MCdebugProgress, thorJob, "ThorSlave terminated OK");
}
catch (IException *e)
{
if (!jobListenerStopped)
FLLOG(MCexception(e), thorJob, e,"ThorSlave");
unregisterException.setown(e);
}
ClearTempDirs();
if (multiThorMemoryThreshold)
setMultiThorMemoryNotify(0,NULL);
roxiemem::releaseRoxieHeap();
if (unregisterException.get())
UnregisterSelf(unregisterException);
if (globals->getPropBool("Debug/@slaveDaliClient"))
disableThorSlaveAsDaliClient();
#ifdef USE_MP_LOG
stopLogMsgReceivers();
#endif
stopMPServer();
::Release(globals);
releaseAtoms(); // don't know why we can't use a module_exit to destruct these...
ExitModuleObjects(); // not necessary, atexit will call, but good for leak checking
return 0;
}
int main( int argc, char *argv[] )
{
#if defined(WIN32) && defined(_DEBUG)
int tmpFlag = _CrtSetDbgFlag( _CRTDBG_REPORT_FLAG );
tmpFlag |= _CRTDBG_LEAK_CHECK_DF;
_CrtSetDbgFlag( tmpFlag );
#endif
InitModuleObjects();
addAbortHandler(ControlHandler);
EnableSEHtoExceptionMapping();
dummyProc();
#ifndef __64BIT__
Thread::setDefaultStackSize(0x10000); // NB under windows requires linker setting (/stack:)
#endif
#ifdef _WIN32
Owned<CReleaseMutex> globalNamedMutex;
#endif
if (globals)
globals->Release();
{
Owned<IFile> iFile = createIFile("thor.xml");
globals = iFile->exists() ? createPTree(*iFile, ipt_caseInsensitive) : createPTree("Thor", ipt_caseInsensitive);
}
unsigned multiThorMemoryThreshold = 0;
try {
if (argc==1)
{
usage();
return 1;
}
cmdArgs = argv+1;
mergeCmdParams(globals);
cmdArgs = argv+1;
const char *master = globals->queryProp("@MASTER");
if (!master)
usage();
const char *slave = globals->queryProp("@SLAVE");
if (slave)
{
slfEp.set(slave);
localHostToNIC(slfEp);
}
else
slfEp.setLocalHost(0);
if (globals->hasProp("@SLAVENUM"))
mySlaveNum = atoi(globals->queryProp("@SLAVENUM"));
else
mySlaveNum = slfEp.port; // shouldn't happen, provided by script
setMachinePortBase(slfEp.port);
slfEp.port = getMachinePortBase();
startSlaveLog();
startMPServer(getFixedPort(TPORT_mp));
#ifdef USE_MP_LOG
startLogMsgParentReceiver();
LOG(MCdebugProgress, thorJob, "MPServer started on port %d", getFixedPort(TPORT_mp));
#endif
SocketEndpoint masterEp(master);
localHostToNIC(masterEp);
setMasterPortBase(masterEp.port);
markNodeCentral(masterEp);
if (RegisterSelf(masterEp))
{
#define ISDALICLIENT // JCSMORE plugins *can* access dali - though I think we should probably prohibit somehow.
#ifdef ISDALICLIENT
const char *daliServers = globals->queryProp("@DALISERVERS");
if (!daliServers)
{
LOG(MCerror, thorJob, "No Dali server list specified\n");
return 1;
}
Owned<IGroup> serverGroup = createIGroup(daliServers, DALI_SERVER_PORT);
unsigned retry = 0;
loop {
try {
LOG(MCdebugProgress, thorJob, "calling initClientProcess");
initClientProcess(serverGroup,DCR_ThorSlave, getFixedPort(TPORT_mp));
break;
}
catch (IJSOCK_Exception *e) {
if ((e->errorCode()!=JSOCKERR_port_in_use))
throw;
FLLOG(MCexception(e), thorJob, e,"InitClientProcess");
if (retry++>10)
throw;
e->Release();
LOG(MCdebugProgress, thorJob, "Retrying");
Sleep(retry*2000);
}
}
setPasswordsFromSDS();
#endif
IDaFileSrvHook *daFileSrvHook = queryDaFileSrvHook();
if (daFileSrvHook) // probably always installed
daFileSrvHook->addSubnetFilters(globals->queryPropTree("NAS"), NULL);
StringBuffer thorPath;
globals->getProp("@thorPath", thorPath);
recursiveCreateDirectory(thorPath.str());
int err = _chdir(thorPath.str());
if (err)
{
IException *e = MakeErrnoException(-1, "Failed to change dir to '%s'",thorPath.str());
FLLOG(MCexception(e), thorJob, e);
throw e;
}
// Initialization from globals
setIORetryCount(globals->getPropInt("Debug/@ioRetries")); // default == 0 == off
StringBuffer str;
if (globals->getProp("@externalProgDir", str.clear()))
_mkdir(str.str());
else
globals->setProp("@externalProgDir", thorPath);
const char * overrideBaseDirectory = globals->queryProp("@thorDataDirectory");
const char * overrideReplicateDirectory = globals->queryProp("@thorReplicateDirectory");
StringBuffer datadir;
StringBuffer repdir;
if (getConfigurationDirectory(globals->queryPropTree("Directories"),"data","thor",globals->queryProp("@name"),datadir))
overrideBaseDirectory = datadir.str();
if (getConfigurationDirectory(globals->queryPropTree("Directories"),"mirror","thor",globals->queryProp("@name"),repdir))
overrideReplicateDirectory = repdir.str();
if (overrideBaseDirectory&&*overrideBaseDirectory)
setBaseDirectory(overrideBaseDirectory, false);
if (overrideReplicateDirectory&&*overrideBaseDirectory)
setBaseDirectory(overrideReplicateDirectory, true);
StringBuffer tempdirstr;
const char *tempdir = globals->queryProp("@thorTempDirectory");
if (getConfigurationDirectory(globals->queryPropTree("Directories"),"temp","thor",globals->queryProp("@name"),tempdirstr))
tempdir = tempdirstr.str();
SetTempDir(tempdir,true);
useMemoryMappedRead(globals->getPropBool("@useMemoryMappedRead"));
LOG(MCdebugProgress, thorJob, "ThorSlave Version LCR - %d.%d started",THOR_VERSION_MAJOR,THOR_VERSION_MINOR);
StringBuffer url;
LOG(MCdebugProgress, thorJob, "Slave %s - temporary dir set to : %s", slfEp.getUrlStr(url).toCharArray(), queryTempDir());
#ifdef _WIN32
ULARGE_INTEGER userfree;
ULARGE_INTEGER total;
ULARGE_INTEGER free;
if (GetDiskFreeSpaceEx("c:\\",&userfree,&total,&free)&&total.QuadPart) {
unsigned pc = (unsigned)(free.QuadPart*100/total.QuadPart);
LOG(MCdebugProgress, thorJob, "Total disk space = %"I64F"d k", total.QuadPart/1000);
LOG(MCdebugProgress, thorJob, "Free disk space = %"I64F"d k", free.QuadPart/1000);
LOG(MCdebugProgress, thorJob, "%d%% disk free\n",pc);
}
#endif
if (getConfigurationDirectory(globals->queryPropTree("Directories"),"query","thor",globals->queryProp("@name"),str.clear()))
globals->setProp("@query_so_dir", str.str());
else
globals->getProp("@query_so_dir", str.clear());
if (str.length())
{
if (globals->getPropBool("Debug/@dllsToSlaves", true))
{
StringBuffer uniqSoPath;
if (PATHSEPCHAR == str.charAt(str.length()-1))
uniqSoPath.append(str.length()-1, str.str());
else
uniqSoPath.append(str);
uniqSoPath.append("_").append(getMachinePortBase());
str.swapWith(uniqSoPath);
globals->setProp("@query_so_dir", str.str());
}
PROGLOG("Using querySo directory: %s", str.str());
recursiveCreateDirectory(str.str());
}
multiThorMemoryThreshold = globals->getPropInt("@multiThorMemoryThreshold")*0x100000;
if (multiThorMemoryThreshold) {
StringBuffer lgname;
if (!globals->getProp("@multiThorResourceGroup",lgname))
globals->getProp("@nodeGroup",lgname);
if (lgname.length()) {
Owned<ILargeMemLimitNotify> notify = createMultiThorResourceMutex(lgname.str());
setMultiThorMemoryNotify(multiThorMemoryThreshold,notify);
PROGLOG("Multi-Thor resource limit for %s set to %"I64F"d",lgname.str(),(__int64)multiThorMemoryThreshold);
}
else
multiThorMemoryThreshold = 0;
}
slaveMain();
}
LOG(MCdebugProgress, thorJob, "ThorSlave terminated OK");
}
catch (IException *e)
{
FLLOG(MCexception(e), thorJob, e,"ThorSlave");
e->Release();
}
catch (CATCHALL)
{
FLLOG(MCerror, thorJob, "ThorSlave exiting because of uncaught exception");
}
ClearTempDirs();
if (multiThorMemoryThreshold)
setMultiThorMemoryNotify(0,NULL);
roxiemem::releaseRoxieHeap();
#ifdef ISDALICLIENT
closeEnvironment();
closedownClientProcess(); // dali client closedown
#endif
#ifdef USE_MP_LOG
stopLogMsgReceivers();
#endif
stopMPServer();
::Release(globals);
releaseAtoms(); // don't know why we can't use a module_exit to destruct these...
return 0;
}