int LogMsgLogReceiverThread::run()
{
while(!done)
{
try
{
if(queryWorldCommunicator().recv(in, childNode, MPTAG_JLOG_CHILD_TO_PARENT))
{
msgBuffer.deserialize(in, true);
if(isListener)
listener->report(msgBuffer);
else
queryLogMsgManager()->report(msgBuffer);
}
}
catch(IException * e)
{
done = true;
CMessageBuffer out;
out.append('D').append(cid);
try
{
queryWorldCommunicator().send(out, queryMyNode(), MPTAG_JLOG_CONNECT_TO_PARENT, MP_ASYNC_SEND);
}
catch(IException * ee)
{
ee->Release();
}
e->Release();
}
}
return 0;
}
CSDSServerStatus::CSDSServerStatus(const char *servername)
{
conn = querySDS().connect("Status/Servers/Server", myProcessSession(), RTM_CREATE_ADD | RTM_LOCK_READ | RTM_DELETE_ON_DISCONNECT, 5*60*1000);
if (conn) {
IPropertyTree &root = *conn->queryRoot();
root.setProp("@name",servername);
StringBuffer node;
queryMyNode()->endpoint().getIpText(node);
root.setProp("@node",node.str());
root.setPropInt("@mpport",queryMyNode()->endpoint().port);
CDateTime dt;
dt.setNow();
StringBuffer str;
root.setProp("@started",dt.getString(str).str());
conn->commit();
}
}
static void replyError(unsigned errorCode, const char *errorMsg)
{
SocketEndpoint myEp = queryMyNode()->endpoint();
StringBuffer str("Node '");
myEp.getUrlStr(str);
str.append("' exception: ").append(errorMsg);
Owned<IException> e = MakeStringException(errorCode, "%s", str.str());
CMessageBuffer msg;
serializeException(e, msg);
queryWorldCommunicator().send(msg, 0, MPTAG_THORREGISTRATION);
}
CGraphProgressHandlerBase() : threaded("CGraphProgressHandler")
{
self = queryMyNode()->endpoint();
stopped = true;
progressEnabled = globals->getPropBool("@watchdogProgressEnabled");
stopped = false;
#ifdef _WIN32
threaded.adjustPriority(+1); // it is critical that watchdog packets get through.
#endif
threaded.init(this);
}
CGraphProgressHandler() : threaded("CGraphProgressHandler")
{
self = queryMyNode()->endpoint();
stopped = true;
StringBuffer ipStr;
queryClusterGroup().queryNode(0).endpoint().getIpText(ipStr);
sock.setown(ISocket::udp_connect(getFixedPort(getMasterPortBase(), TPORT_watchdog),ipStr.str()));
progressEnabled = globals->getPropBool("@watchdogProgressEnabled");
sendData(); // send initial data
stopped = false;
#ifdef _WIN32
threaded.adjustPriority(+1); // it is critical that watchdog packets get through.
#endif
threaded.init(this);
}
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;
}