int main(int argc, char* argv[])
{
enableMemLeakChecking(true);
InitModuleObjects();
EnableSEHtoExceptionMapping();
if (argc<2) {
usage(argv[0]);
return -1;
}
SocketEndpoint dalieps(argv[1],DALI_SERVER_PORT); // endpoint of dali server
Owned<IGroup> group = createIGroup(1,&dalieps);
try {
initClientProcess(group, DCR_Other); // I will add a DCR_Orbit at some point
try {
doStuff();
}
catch (IException *e) {
pexception(argv[0],e);
e->Release();
}
closedownClientProcess();
}
catch (IException *e) {
pexception(argv[0],e);
e->Release();
}
releaseAtoms();
return 0;
}
void enableThorSlaveAsDaliClient()
{
#ifdef ISDALICLIENT
PROGLOG("Slave activated as a Dali client");
const char *daliServers = globals->queryProp("@DALISERVERS");
if (!daliServers)
throw MakeStringException(0, "No Dali server list specified");
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
}
DaliClient(const char* daliserver): serverGroup(createIGroup(daliserver, DALI_SERVER_PORT))
{
if (!serverGroup)
throw MakeStringException(0, "Could not instantiate IGroup");
if (!initClientProcess(serverGroup,DCR_Util))
throw MakeStringException(0, "Could not initializing client process");
setPasswordsFromSDS();
}
bool updateDaliEnv(IPropertyTree *env, bool forceGroupUpdate, const char *daliIp)
{
Owned<IPropertyTreeIterator> dalis = env->getElements("Software/DaliServerProcess/Instance");
if (!dalis||!dalis->first()) {
fprintf(stderr,"Could not find DaliServerProcess\n");
return false;
}
SocketEndpoint daliep;
loop {
const char *ps = dalis->get().queryProp("@port");
unsigned port = ps?atoi(ps):0;
if (!port)
port = DALI_SERVER_PORT;
daliep.set(dalis->get().queryProp("@netAddress"),port);
if (daliIp && *daliIp) {
SocketEndpoint testep;
testep.set(daliIp,DALI_SERVER_PORT);
if (testep.equals(daliep))
break;
daliep.set(NULL,0);
}
if (!dalis->next())
break;
if (!daliep.isNull()) {
fprintf(stderr,"Ambiguous DaliServerProcess instance\n");
return false;
}
}
if (daliep.isNull()) {
fprintf(stderr,"Could not find DaliServerProcess instance\n");
return false;
}
SocketEndpointArray epa;
epa.append(daliep);
Owned<IGroup> group = createIGroup(epa);
bool ret = true;
initClientProcess(group, DCR_Util);
StringBuffer response;
if (querySDS().updateEnvironment(env, forceGroupUpdate, response))
{
StringBuffer tmp;
PROGLOG("Environment and node groups updated in dali at %s",daliep.getUrlStr(tmp).str());
}
else
ret = false;
if (response.length())
WARNLOG("%s", response.str());
closedownClientProcess();
return ret;
}
void testMultiCluster()
{
Owned<IGroup> grp1 = createIGroup("192.168.51.1-5");
Owned<IGroup> grp2 = createIGroup("192.168.16.1-5");
Owned<IGroup> grp3 = createIGroup("192.168.53.1-5");
queryNamedGroupStore().add("testgrp1",grp1);
queryNamedGroupStore().add("testgrp2",grp2);
queryNamedGroupStore().add("testgrp3",grp3);
Owned<IFileDescriptor> fdesc = createFileDescriptor();
fdesc->setDefaultDir("/c$/thordata/test");
fdesc->setPartMask("testfile1._$P$_of_$N$");
fdesc->setNumParts(5);
ClusterPartDiskMapSpec mapping;
fdesc->addCluster(grp1,mapping);
fdesc->addCluster(grp2,mapping);
fdesc->addCluster(grp3,mapping);
queryDistributedFileDirectory().removeEntry("test::testfile1",UNKNOWN_USER);
Owned<IDistributedFile> file = queryDistributedFileDirectory().createNew(fdesc);
queryDistributedFileDirectory().removeEntry("test::testfile1",UNKNOWN_USER);
file->attach("test::testfile1",UNKNOWN_USER);
StringBuffer name;
unsigned i;
for (i=0;i<file->numClusters();i++)
PROGLOG("cluster[%d] = %s",i,file->getClusterName(i,name.clear()).str());
file.clear();
file.setown(queryDistributedFileDirectory().lookup("test::testfile1",UNKNOWN_USER));
for (i=0;i<file->numClusters();i++)
PROGLOG("cluster[%d] = %s",i,file->getClusterName(i,name.clear()).str());
file.clear();
file.setown(queryDistributedFileDirectory().lookup("test::testfile1@testgrp1",UNKNOWN_USER));
for (i=0;i<file->numClusters();i++)
PROGLOG("cluster[%d] = %s",i,file->getClusterName(i,name.clear()).str());
file.clear();
queryDistributedFileDirectory().removePhysical("test::testfile1@testgrp2",UNKNOWN_USER);
file.setown(queryDistributedFileDirectory().lookup("test::testfile1",UNKNOWN_USER));
for (i=0;i<file->numClusters();i++)
PROGLOG("cluster[%d] = %s",i,file->getClusterName(i,name.clear()).str());
}
void CEspConfig::initDali(const char *servers)
{
if (servers!=NULL && *servers!=0 && !daliClientActive())
{
DBGLOG("Initializing DALI client [servers = %s]", servers);
useDali=true;
// Create server group
Owned<IGroup> serverGroup = createIGroup(servers, DALI_SERVER_PORT);
if (!serverGroup)
throw MakeStringException(0, "Could not instantiate dali IGroup");
// Initialize client process
if (!initClientProcess(serverGroup, DCR_EspServer))
throw MakeStringException(0, "Could not initialize dali client");
setPasswordsFromSDS();
serverstatus = new CSDSServerStatus("ESPserver");
}
}
void processMessage(CMessageBuffer &mb)
{
ICoven &coven=queryCoven();
SessionId id;
int fn;
mb.read(fn);
switch (fn) {
case MSR_REGISTER_PROCESS_SESSION: {
acceptConnections.wait();
acceptConnections.signal();
Owned<INode> node(deserializeINode(mb));
Owned<INode> servernode(deserializeINode(mb)); // hopefully me, but not if forwarded
int role=0;
if (mb.length()-mb.getPos()>=sizeof(role)) { // a capability block present
mb.read(role);
if (!manager.authorizeConnection(role,false)) {
SocketEndpoint sender = mb.getSender();
mb.clear();
coven.reply(mb);
MilliSleep(100+getRandom()%1000); // Causes client to 'work' for a short time.
Owned<INode> node = createINode(sender);
coven.disconnect(node);
break;
}
#ifdef _DEBUG
StringBuffer eps;
PROGLOG("Connection to %s authorized",mb.getSender().getUrlStr(eps).str());
#endif
}
IGroup *covengrp;
id = manager.registerClientProcess(node.get(),covengrp,(DaliClientRole)role);
mb.clear().append(id);
if (covengrp->rank(servernode)==RANK_NULL) { // must have been redirected
covengrp->Release(); // no good, so just use one we know about (may use something more sophisticated later)
INode *na = servernode.get();
covengrp = createIGroup(1, &na);
}
covengrp->serialize(mb);
covengrp->Release();
coven.reply(mb);
}
break;
case MSR_SECONDARY_REGISTER_PROCESS_SESSION: {
mb.read(id);
Owned<INode> node (deserializeINode(mb));
int role;
mb.read(role);
manager.addProcessSession(id,node.get(),(DaliClientRole)role);
mb.clear();
coven.reply(mb);
}
break;
case MSR_REGISTER_SESSION: {
SecurityToken tok;
SessionId parentid;
mb.read(tok).read(parentid);
SessionId id = manager.registerSession(tok,parentid);
mb.clear().append(id);
coven.reply(mb);
}
break;
case MSR_SECONDARY_REGISTER_SESSION: {
mb.read(id);
manager.addSession(id);
mb.clear();
coven.reply(mb);
}
break;
case MSR_LOOKUP_PROCESS_SESSION: {
// looks up from node or from id
Owned<INode> node (deserializeINode(mb));
if (node->endpoint().isNull()&&(mb.length()-mb.getPos()>=sizeof(id))) {
mb.read(id);
INode *n = manager.getProcessSessionNode(id);
if (n)
node.setown(n);
node->serialize(mb.clear());
}
else {
id = manager.lookupProcessSession(node.get());
mb.clear().append(id);
}
coven.reply(mb);
}
break;
case MSR_STOP_SESSION: {
SessionId sessid;
bool failed;
mb.read(sessid).read(failed);
manager.stopSession(sessid,failed);
mb.clear();
coven.reply(mb);
}
break;
case MSR_LOOKUP_LDAP_PERMISSIONS: {
StringAttr key;
StringAttr obj;
Owned<IUserDescriptor> udesc=createUserDescriptor();
StringAttr username;
StringAttr passwordenc;
mb.read(key).read(obj);
udesc->deserialize(mb);
#ifndef _NO_DALIUSER_STACKTRACE
//following debug code to be removed
StringBuffer sb;
udesc->getUserName(sb);
if (0==sb.length())
{
DBGLOG("UNEXPECTED USER (NULL) in dasess.cpp CSessionRequestServer::processMessage() line %d", __LINE__);
}
#endif
unsigned auditflags = 0;
if (mb.length()-mb.getPos()>=sizeof(auditflags))
mb.read(auditflags);
int err = 0;
int ret=manager.getPermissionsLDAP(key,obj,udesc,auditflags,&err);
mb.clear().append(ret);
if (err)
mb.append(err);
coven.reply(mb);
}
break;
}
}
int main(int argc, char* argv[])
{
InitModuleObjects();
int exitCode = 1;
try
{
if (argc<2)
{
printf("usage: dalistop <server_ip:port> [/nowait]\n");
printf("eg: dalistop . -- stop dali server running locally\n");
printf(" dalistop eq0001016 -- stop dali server running remotely\n");
}
else
{
SocketEndpoint ep;
ep.set(argv[1],DALI_SERVER_PORT);
bool nowait = false;
if (argc>=3)
nowait = stricmp(argv[2],"/nowait")==0;
printf("Stopping Dali Server on %s\n",argv[1]);
startMPServer(0);
Owned<IGroup> group = createIGroup(1,&ep);
Owned<ICommunicator> comm = createCommunicator(group);
CMessageBuffer mb;
int fn=-1;
mb.append(fn);
if (comm->verifyConnection(0,2000))
{
comm->send(mb,0,MPTAG_DALI_COVEN_REQUEST,MP_ASYNC_SEND);
if (nowait)
{
Sleep(1000);
exitCode = 0;
}
else
{
// verifyConnection() has a min conn timeout of 10s
// use recv() instead to check for socket closed ...
try
{
while (!comm->recv(mb,0,MPTAG_DALI_COVEN_REQUEST,nullptr,5000))
{
printf("Waiting for Dali Server to stop....\n");
}
exitCode = 0;
}
catch (IMP_Exception *e)
{
if (e->errorCode() == MPERR_link_closed)
exitCode = 0;
e->Release();
}
}
}
else
fprintf(stderr, "Dali not responding\n");
stopMPServer();
}
}
catch (IException *e)
{
pexception("Exception",e);
stopMPServer();
}
releaseAtoms();
return exitCode;
}
int main(int argc, const char *argv[])
{
InitModuleObjects();
EnableSEHtoExceptionMapping();
NoQuickEditSection xxx;
Owned<IFile> file = createIFile("dfuserver.xml");
if (file->exists())
globals.setown(createPTreeFromXMLFile("dfuserver.xml", ipt_caseInsensitive));
else
globals.setown(readOldIni());
for (unsigned i=1;i<(unsigned)argc;i++) {
const char *arg = argv[i];
StringBuffer prop("@");
StringBuffer val;
while (*arg && *arg != '=')
prop.append(*arg++);
if (*arg) {
arg++;
while (isspace(*arg))
arg++;
val.append(arg);
prop.clip();
val.clip();
if (prop.length()>1)
globals->setProp(prop.str(), val.str());
}
}
StringBuffer daliServer;
StringBuffer queue;
if (!globals->getProp("@DALISERVERS", daliServer)||!globals->getProp("@QUEUE", queue)) {
usage();
globals.clear();
releaseAtoms();
return 1;
}
Owned<IFile> sentinelFile;
bool stop = globals->getPropInt("@STOP",0)!=0;
if (!stop) {
sentinelFile.setown(createSentinelTarget());
removeSentinelFile(sentinelFile);
StringBuffer logname;
StringBuffer logdir;
if (!getConfigurationDirectory(globals->queryPropTree("Directories"),"log","dfuserver",globals->queryProp("@name"),logdir))
globals->getProp("@LOG_DIR", logdir);
if (logdir.length() && recursiveCreateDirectory(logdir.str()))
logname.append(logdir);
else
appendCurrentDirectory(logname, true);
if (logname.length() && logname.charAt(logname.length()-1) != PATHSEPCHAR)
logname.append(PATHSEPCHAR);
logname.append("dfuserver");
StringBuffer aliasLogName(logname);
aliasLogName.append(".log");
fileMsgHandler = getRollingFileLogMsgHandler(logname.str(), ".log", MSGFIELD_STANDARD, false, true, NULL, aliasLogName.str());
queryLogMsgManager()->addMonitorOwn(fileMsgHandler, getCategoryLogMsgFilter(MSGAUD_all, MSGCLS_all, 1000));
}
StringBuffer ftslogdir;
if (getConfigurationDirectory(globals->queryPropTree("Directories"),"log","ftslave",globals->queryProp("@name"),ftslogdir)) // NB instance deliberately dfuserver's
setFtSlaveLogDir(ftslogdir.str());
setRemoteSpawnSSH(
globals->queryProp("SSH/@SSHidentityfile"),
globals->queryProp("SSH/@SSHusername"),
globals->queryProp("SSH/@SSHpassword"),
globals->getPropInt("SSH/@SSHtimeout",0),
globals->getPropInt("SSH/@SSHretries",3),
"run_");
bool enableSNMP = globals->getPropInt("@enableSNMP")!=0;
CSDSServerStatus *serverstatus=NULL;
Owned<IReplicateServer> replserver;
try {
Owned<IGroup> serverGroup = createIGroup(daliServer.str(),DALI_SERVER_PORT);
initClientProcess(serverGroup, DCR_DfuServer, 0, NULL, NULL, stop?(1000*30):MP_WAIT_FOREVER);
setPasswordsFromSDS();
if(!stop)
{
if (globals->getPropBool("@enableSysLog",true))
UseSysLogForOperatorMessages();
serverstatus = new CSDSServerStatus("DFUserver");
setDaliServixSocketCaching(true); // speeds up lixux operations
startLogMsgParentReceiver(); // for auditing
connectLogMsgManagerToDali();
engine.setown(createDFUengine());
addAbortHandler(exitDFUserver);
}
const char *q = queue.str();
loop {
StringBuffer subq;
const char *comma = strchr(q,',');
if (comma)
subq.append(comma-q,q);
else
subq.append(q);
if (stop) {
stopDFUserver(subq.str());
}
else {
StringBuffer mask;
mask.appendf("Queue[@name=\"%s\"][1]",subq.str());
IPropertyTree *t=serverstatus->queryProperties()->queryPropTree(mask.str());
if (t)
t->setPropInt("@num",t->getPropInt("@num",0)+1);
else {
t = createPTree();
t->setProp("@name",subq.str());
t->setPropInt("@num",1);
serverstatus->queryProperties()->addPropTree("Queue",t);
}
serverstatus->commitProperties();
engine->setDefaultTransferBufferSize((size32_t)globals->getPropInt("@transferBufferSize"));
engine->startListener(subq.str(),serverstatus);
}
if (!comma)
break;
q = comma+1;
if (!*q)
break;
}
q = globals->queryProp("@MONITORQUEUE");
if (q&&*q) {
if (stop) {
stopDFUserver(q);
}
else {
IPropertyTree *t=serverstatus->queryProperties()->addPropTree("MonitorQueue",createPTree());
t->setProp("@name",q);
engine->startMonitor(q,serverstatus,globals->getPropInt("@MONITORINTERVAL",60)*1000);
}
}
q = globals->queryProp("@REPLICATEQUEUE");
if (q&&*q) {
if (stop) {
// TBD?
}
else {
replserver.setown(createReplicateServer(q));
replserver->runServer();
}
}
if (!stop) {
serverstatus->commitProperties();
writeSentinelFile(sentinelFile);
engine->joinListeners();
if (replserver.get())
replserver->stopServer();
LOG(MCprogress, unknownJob, "Exiting");
}
}
catch(IException *e){
EXCLOG(e, "DFU Server Exception: ");
e->Release();
}
catch (const char *s) {
WARNLOG("DFU: %s",s);
}
delete serverstatus;
if (stop)
Sleep(2000); // give time to stop
engine.clear();
globals.clear();
closeEnvironment();
closedownClientProcess();
UseSysLogForOperatorMessages(false);
setDaliServixSocketCaching(false);
releaseAtoms();
return 0;
}
static void test2()
{
const size32_t recsize = 17;
printf("Test DFS\n");
StringBuffer s;
unsigned i;
unsigned n;
unsigned t;
queryNamedGroupStore().remove("daregress_group");
queryDistributedFileDirectory().removeEntry("daregress::superfile1");
SocketEndpointArray epa;
for (n=0;n<400;n++) {
s.clear().append("192.168.").append(n/256).append('.').append(n%256);
SocketEndpoint ep(s.str());
epa.append(ep);
}
Owned<IGroup> group = createIGroup(epa);
queryNamedGroupStore().add("daregress_group",group,true);
if (!queryNamedGroupStore().find(group,s.clear()))
ERROR("Created logical group not found");
if (stricmp(s.str(),"daregress_group")!=0)
ERROR("Created logical group found with wrong name");
group.setown(queryNamedGroupStore().lookup("daregress_group"));
if (!group)
ERROR("named group lookup failed");
printf("Named group created - 400 nodes\n");
for (i=0;i<100;i++) {
Owned<IPropertyTree> pp = createPTree("Part");
Owned<IFileDescriptor>fdesc = createFileDescriptor();
fdesc->setDefaultDir("c:\\thordata\\regress");
n = 9;
for (unsigned k=0;k<400;k++) {
s.clear().append("192.168.").append(n/256).append('.').append(n%256);
Owned<INode> node = createINode(s.str());
pp->setPropInt64("@size",(n*777+i)*recsize);
s.clear().append("daregress_test").append(i).append("._").append(n+1).append("_of_400");
fdesc->setPart(n,node,s.str(),pp);
n = (n+9)%400;
}
fdesc->queryProperties().setPropInt("@recordSize",17);
s.clear().append("daregress::test").append(i);
queryDistributedFileDirectory().removeEntry(s.str());
StringBuffer cname;
Owned<IDistributedFile> dfile = queryDistributedFileDirectory().createNew(fdesc);
if (stricmp(dfile->getClusterName(0,cname),"daregress_group")!=0)
ERROR1("Cluster name wrong %d",i);
s.clear().append("daregress::test").append(i);
dfile->attach(s.str());
}
printf("DFile create done - 100 files\n");
unsigned samples = 5;
t = 33;
for (i=0;i<100;i++) {
s.clear().append("daregress::test").append(t);
if (!queryDistributedFileDirectory().exists(s.str()))
ERROR1("Could not find %s",s.str());
Owned<IDistributedFile> dfile = queryDistributedFileDirectory().lookup(s.str());
if (!dfile) {
ERROR1("Could not find %s",s.str());
continue;
}
offset_t totsz = 0;
n = 11;
for (unsigned k=0;k<400;k++) {
Owned<IDistributedFilePart> part = dfile->getPart(n);
if (!part) {
ERROR2("part not found %d %d",t,n);
continue;
}
s.clear().append("192.168.").append(n/256).append('.').append(n%256);
Owned<INode> node = createINode(s.str());
if (!node->equals(part->queryNode()))
ERROR2("part node mismatch %d, %d",t,n);
if (part->getFileSize(false,false)!=(n*777+t)*recsize)
ERROR4("size node mismatch %d, %d, %d, %d",t,n,(unsigned)part->getFileSize(false,false),(n*777+t)*recsize);
s.clear().append("daregress_test").append(t).append("._").append(n+1).append("_of_400");
/* ** TBD
if (stricmp(s.str(),part->queryPartName())!=0)
ERROR4("part name mismatch %d, %d '%s' '%s'",t,n,s.str(),part->queryPartName());
*/
totsz += (n*777+t)*recsize;
if ((samples>0)&&(i+n+t==k)) {
samples--;
RemoteFilename rfn;
part->getFilename(rfn,samples%2);
StringBuffer fn;
rfn.getRemotePath(fn);
printf("SAMPLE: %d,%d %s\n",t,n,fn.str());
}
n = (n+11)%400;
}
if (totsz!=dfile->getFileSize(false,false))
ERROR1("total size mismatch %d",t);
t = (t+33)%100;
}
printf("DFile lookup done - 100 files\n");
// check iteration
__int64 crctot = 0;
unsigned np = 0;
unsigned totrows = 0;
Owned<IDistributedFileIterator> fiter = queryDistributedFileDirectory().getIterator("daregress::*",false);
Owned<IDistributedFilePartIterator> piter;
ForEach(*fiter) {
piter.setown(fiter->query().getIterator());
ForEach(*piter) {
RemoteFilename rfn;
StringBuffer s;
piter->query().getFilename(rfn,0);
rfn.getRemotePath(s);
piter->query().getFilename(rfn,1);
rfn.getRemotePath(s);
crctot += crc32(s.str(),s.length(),0);
np++;
totrows += (unsigned)(piter->query().getFileSize(false,false)/fiter->query().queryProperties().getPropInt("@recordSize",-1));
}
}
piter.clear();
fiter.clear();
printf("DFile iterate done - %d parts, %d rows, CRC sum %"I64F"d\n",np,totrows,crctot);
Owned<IDistributedSuperFile> sfile;
sfile.setown(queryDistributedFileDirectory().createSuperFile("daregress::superfile1",true));
for (i = 0;i<100;i++) {
s.clear().append("daregress::test").append(i);
sfile->addSubFile(s.str());
}
sfile.clear();
sfile.setown(queryDistributedFileDirectory().lookupSuperFile("daregress::superfile1"));
if (!sfile) {
ERROR("Could not find added superfile");
return;
}
__int64 savcrc = crctot;
crctot = 0;
np = 0;
totrows = 0;
size32_t srs = (size32_t)sfile->queryProperties().getPropInt("@recordSize",-1);
if (srs!=17)
ERROR1("Superfile does not match subfile row size %d",srs);
piter.setown(sfile->getIterator());
ForEach(*piter) {
RemoteFilename rfn;
StringBuffer s;
piter->query().getFilename(rfn,0);
rfn.getRemotePath(s);
piter->query().getFilename(rfn,1);
rfn.getRemotePath(s);
crctot += crc32(s.str(),s.length(),0);
np++;
totrows += (unsigned)(piter->query().getFileSize(false,false)/srs);
}
piter.clear();
printf("Superfile iterate done - %d parts, %d rows, CRC sum %"I64F"d\n",np,totrows,crctot);
if (crctot!=savcrc)
ERROR("SuperFile does not match sub files");
unsigned tr = (unsigned)(sfile->getFileSize(false,false)/srs);
if (totrows!=tr)
ERROR1("Superfile size does not match part sum %d",tr);
sfile->detach();
sfile.clear();
sfile.setown(queryDistributedFileDirectory().lookupSuperFile("daregress::superfile1"));
if (sfile)
ERROR("Superfile deletion failed");
t = 37;
for (i=0;i<100;i++) {
s.clear().append("daregress::test").append(t);
if (i%1) {
Owned<IDistributedFile> dfile = queryDistributedFileDirectory().lookup(s.str());
if (!dfile)
ERROR1("Could not find %s",s.str());
dfile->detach();
}
else
queryDistributedFileDirectory().removeEntry(s.str());
t = (t+37)%100;
}
printf("DFile removal complete\n");
t = 39;
for (i=0;i<100;i++) {
if (queryDistributedFileDirectory().exists(s.str()))
ERROR1("Found %s after deletion",s.str());
Owned<IDistributedFile> dfile = queryDistributedFileDirectory().lookup(s.str());
if (dfile)
ERROR1("Found %s after deletion",s.str());
t = (t+39)%100;
}
printf("DFile removal check complete\n");
queryNamedGroupStore().remove("daregress_group");
if (queryNamedGroupStore().lookup("daregress_group"))
ERROR("Named group not removed");
}
int CEclAgentExecutionServer::run()
{
Owned<IFile> sentinelFile = createSentinelTarget();
removeSentinelFile(sentinelFile);
try
{
Owned<IGroup> serverGroup = createIGroup(daliServers, DALI_SERVER_PORT);
initClientProcess(serverGroup, DCR_EclServer);
getAgentQueueNames(queueNames, agentName);
queue.setown(createJobQueue(queueNames.str()));
queue->connect();
}
catch (IException *e)
{
EXCLOG(e, "Server queue create/connect: ");
e->Release();
return -1;
}
catch(...)
{
ERRLOG("Terminating unexpectedly");
}
writeSentinelFile(sentinelFile);
try
{
while (started)
{
PROGLOG("AgentExec: Waiting on queue(s) '%s'", queueNames.str());
Owned<IJobQueueItem> item = queue->dequeue(WAIT_FOREVER);
if (item.get())
{
rebuildLogfileName();//rebuild in case date rollover
StringAttr wuid;
wuid.set(item->queryWUID());
PROGLOG("AgentExec: Dequeued workunit request '%s'", wuid.get());
try
{
executeWorkunit(wuid);
}
catch(IException *e)
{
EXCLOG(e, "CEclAgentExecutionServer::run: ");
}
catch(...)
{
ERRLOG("Unexpected exception in CEclAgentExecutionServer::run caught");
}
}
else
{
ERRLOG("Unexpected dequeue of bogus job queue item, exiting agentexec");
removeSentinelFile(sentinelFile);//no reason to restart
assert(!started);
break;
}
}
}
catch (IException *e)
{
EXCLOG(e, "Server Exception: ");
e->Release();
PROGLOG("Exiting");
}
try
{
queue->disconnect();
}
catch (IException *e)
{
EXCLOG(e, "Server queue disconnect: ");
e->Release();
}
PROGLOG("Exiting agentexec\n");
return 1;
}
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;
}