bool
DCCollector::finishUpdate( DCCollector *self, Sock* sock, ClassAd* ad1, ClassAd* ad2 )
{
// This is a static function so that we can call it from a
// nonblocking startCommand() callback without worrying about
// longevity of the DCCollector instance.
sock->encode();
if( ad1 && ! putClassAd(sock, *ad1) ) {
if(self) {
self->newError( CA_COMMUNICATION_ERROR,
"Failed to send ClassAd #1 to collector" );
}
return false;
}
if( ad2 && ! putClassAd(sock, *ad2) ) {
if(self) {
self->newError( CA_COMMUNICATION_ERROR,
"Failed to send ClassAd #2 to collector" );
return false;
}
}
if( ! sock->end_of_message() ) {
if(self) {
self->newError( CA_COMMUNICATION_ERROR,
"Failed to send EOM to collector" );
}
return false;
}
return true;
}
bool
TransferQueueRequest::SendGoAhead(XFER_QUEUE_ENUM go_ahead,char const *reason) {
ASSERT( m_sock );
m_sock->encode();
ClassAd msg;
msg.Assign(ATTR_RESULT,(int)go_ahead);
if( reason ) {
msg.Assign(ATTR_ERROR_STRING,reason);
}
// how often should transfer processes send a report of I/O activity
// 0 means never
int report_interval = param_integer("TRANSFER_IO_REPORT_INTERVAL",10,0);
msg.Assign(ATTR_REPORT_INTERVAL,report_interval);
if(!putClassAd( m_sock, msg ) || !m_sock->end_of_message()) {
dprintf(D_ALWAYS,
"TransferQueueRequest: failed to send GoAhead to %s\n",
Description() );
return false;
}
m_gave_go_ahead = true;
m_time_go_ahead = time(NULL);
return true;
}
bool
ClaimStartdMsg::writeMsg( DCMessenger * /*messenger*/, Sock *sock ) {
// save startd fqu for hole punching
m_startd_fqu = sock->getFullyQualifiedUser();
m_startd_ip_addr = sock->peer_ip_str();
// Insert an attribute in the request ad to inform the
// startd that this schedd is capable of understanding
// the newer protocol where the claim response may send
// over any leftover resources from a partitionable slot.
m_job_ad.Assign("_condor_SEND_LEFTOVERS",
param_boolean("CLAIM_PARTITIONABLE_LEFTOVERS",true));
// Insert an attribute in the request ad to inform the
// startd that this schedd is capable of understanding
// the newer protocol where the claim response may send
// over the ad and claim id of the partner of a paired slot.
m_job_ad.Assign("_condor_SEND_PAIRED_SLOT",
param_boolean("CLAIM_PAIRED_SLOT",true));
if( !sock->put_secret( m_claim_id.c_str() ) ||
!putClassAd( sock, m_job_ad ) ||
!sock->put( m_scheduler_addr.c_str() ) ||
!sock->put( m_alive_interval ) ||
!this->putExtraClaims(sock))
{
dprintf(failureDebugLevel(),
"Couldn't encode request claim to startd %s\n",
description() );
sockFailed( sock );
return false;
}
// end_of_message() is done by caller
return true;
}
void
CCBServer::ForwardRequestToTarget( CCBServerRequest *request, CCBTarget *target )
{
Sock *sock = target->getSock();
ClassAd msg;
msg.Assign( ATTR_COMMAND, CCB_REQUEST );
msg.Assign( ATTR_MY_ADDRESS, request->getReturnAddr() );
msg.Assign( ATTR_CLAIM_ID, request->getConnectID() );
// for easier debugging
msg.Assign( ATTR_NAME, request->getSock()->peer_description() );
MyString reqid_str;
CCBIDToString( request->getRequestID(), reqid_str);
msg.Assign( ATTR_REQUEST_ID, reqid_str );
sock->encode();
if( !putClassAd( sock, msg ) || !sock->end_of_message() ) {
dprintf(D_ALWAYS,
"CCB: failed to forward request id %lu from %s to target "
"daemon %s with ccbid %lu\n",
request->getRequestID(),
request->getSock()->peer_description(),
target->getSock()->peer_description(),
target->getCCBID());
RequestFinished( request, false, "failed to forward request to target" );
return;
}
// Now wait for target to respond (HandleRequestResultsMsg).
// We will get the response next time we poll the socket.
// To get a faster response, we _could_ register the socket
// now, if it has not already been registered.
}
bool
ClassAdMsg::writeMsg( DCMessenger * /*messenger*/, Sock *sock )
{
if( !putClassAd( sock, m_msg ) ) {
sockFailed( sock );
return false;
}
return true;
}
bool
DCStartd::drainJobs(int how_fast,bool resume_on_completion,char const *check_expr,char const *start_expr,std::string &request_id)
{
std::string error_msg;
ClassAd request_ad;
Sock *sock = startCommand( DRAIN_JOBS, Sock::reli_sock, 20 );
if( !sock ) {
formatstr(error_msg,"Failed to start DRAIN_JOBS command to %s",name());
newError(CA_FAILURE,error_msg.c_str());
return false;
}
request_ad.Assign(ATTR_HOW_FAST,how_fast);
request_ad.Assign(ATTR_RESUME_ON_COMPLETION,resume_on_completion);
if( check_expr ) {
request_ad.AssignExpr(ATTR_CHECK_EXPR,check_expr);
}
if( start_expr ) {
request_ad.AssignExpr(ATTR_START_EXPR,start_expr);
}
if( !putClassAd(sock, request_ad) || !sock->end_of_message() ) {
formatstr(error_msg,"Failed to compose DRAIN_JOBS request to %s",name());
newError(CA_FAILURE,error_msg.c_str());
delete sock;
return false;
}
sock->decode();
ClassAd response_ad;
if( !getClassAd(sock, response_ad) || !sock->end_of_message() ) {
formatstr(error_msg,"Failed to get response to DRAIN_JOBS request to %s",name());
newError(CA_FAILURE,error_msg.c_str());
delete sock;
return false;
}
response_ad.LookupString(ATTR_REQUEST_ID,request_id);
bool result = false;
int error_code = 0;
response_ad.LookupBool(ATTR_RESULT,result);
if( !result ) {
std::string remote_error_msg;
response_ad.LookupString(ATTR_ERROR_STRING,remote_error_msg);
response_ad.LookupInteger(ATTR_ERROR_CODE,error_code);
formatstr(error_msg,
"Received failure from %s in response to DRAIN_JOBS request: error code %d: %s",
name(),error_code,remote_error_msg.c_str());
newError(CA_FAILURE,error_msg.c_str());
delete sock;
return false;
}
delete sock;
return true;
}
// The transferd only needs a small amount of the information from this
// transfer request seriazed to it from the schedd. Basically the m_ip
// classad which represents the header information for this treq, and
// the job ads.
int
TransferRequest::put(Stream *sock)
{
ClassAd *ad = NULL;
sock->encode();
// shove the internal header classad across
putClassAd(sock, *m_ip);
sock->end_of_message();
// now dump all of the jobads through
m_todo_ads.Rewind();
while(m_todo_ads.Next(ad))
{
putClassAd(sock, *ad);
sock->end_of_message();
}
return TRUE;
}
bool
DCStarter::createJobOwnerSecSession(int timeout,char const *job_claim_id,char const *starter_sec_session,char const *session_info,MyString &owner_claim_id,MyString &error_msg,MyString &starter_version,MyString &starter_addr)
{
ReliSock sock;
if (IsDebugLevel(D_COMMAND)) {
dprintf (D_COMMAND, "DCStarter::createJobOwnerSecSession(%s,...) making connection to %s\n",
getCommandStringSafe(CREATE_JOB_OWNER_SEC_SESSION), _addr ? _addr : "NULL");
}
if( !connectSock(&sock, timeout, NULL) ) {
error_msg = "Failed to connect to starter";
return false;
}
if( !startCommand(CREATE_JOB_OWNER_SEC_SESSION, &sock,timeout,NULL,NULL,false,starter_sec_session) ) {
error_msg = "Failed to send CREATE_JOB_OWNER_SEC_SESSION to starter";
return false;
}
ClassAd input;
input.Assign(ATTR_CLAIM_ID,job_claim_id);
input.Assign(ATTR_SESSION_INFO,session_info);
sock.encode();
if( !putClassAd(&sock, input) || !sock.end_of_message() ) {
error_msg = "Failed to compose CREATE_JOB_OWNER_SEC_SESSION to starter";
return false;
}
sock.decode();
ClassAd reply;
if( !getClassAd(&sock, reply) || !sock.end_of_message() ) {
error_msg = "Failed to get response to CREATE_JOB_OWNER_SEC_SESSION from starter";
return false;
}
bool success = false;
reply.LookupBool(ATTR_RESULT,success);
if( !success ) {
reply.LookupString(ATTR_ERROR_STRING,error_msg);
return false;
}
reply.LookupString(ATTR_CLAIM_ID,owner_claim_id);
reply.LookupString(ATTR_VERSION,starter_version);
// get the full starter address from the starter in case it contains
// extra CCB info that we don't already know about
reply.LookupString(ATTR_STARTER_IP_ADDR,starter_addr);
return true;
}
bool
SwapClaimsMsg::writeMsg( DCMessenger * /*messenger*/, Sock *sock ) {
if ( ! sock->put_secret(m_claim_id.c_str()) || ! putClassAd(sock, m_opts))
{
dprintf(failureDebugLevel(),
"Couldn't encode claim swap request to startd %s\n",
description() );
sockFailed(sock);
return false;
}
// end_of_message() is done by caller
return true;
}
bool
DCStartd::cancelDrainJobs(char const *request_id)
{
std::string error_msg;
ClassAd request_ad;
Sock *sock = startCommand( CANCEL_DRAIN_JOBS, Sock::reli_sock, 20 );
if( !sock ) {
formatstr(error_msg,"Failed to start CANCEL_DRAIN_JOBS command to %s",name());
newError(CA_FAILURE,error_msg.c_str());
return false;
}
if( request_id ) {
request_ad.Assign(ATTR_REQUEST_ID,request_id);
}
if( !putClassAd(sock, request_ad) || !sock->end_of_message() ) {
formatstr(error_msg,"Failed to compose CANCEL_DRAIN_JOBS request to %s",name());
newError(CA_FAILURE,error_msg.c_str());
return false;
}
sock->decode();
ClassAd response_ad;
if( !getClassAd(sock, response_ad) || !sock->end_of_message() ) {
formatstr(error_msg,"Failed to get response to CANCEL_DRAIN_JOBS request to %s",name());
newError(CA_FAILURE,error_msg.c_str());
delete sock;
return false;
}
bool result = false;
int error_code = 0;
response_ad.LookupBool(ATTR_RESULT,result);
if( !result ) {
std::string remote_error_msg;
response_ad.LookupString(ATTR_ERROR_STRING,remote_error_msg);
response_ad.LookupInteger(ATTR_ERROR_CODE,error_code);
formatstr(error_msg,
"Received failure from %s in response to CANCEL_DRAIN_JOBS request: error code %d: %s",
name(),error_code,remote_error_msg.c_str());
newError(CA_FAILURE,error_msg.c_str());
delete sock;
return false;
}
delete sock;
return true;
}
bool
CCBListener::WriteMsgToCCB(ClassAd &msg)
{
if( !m_sock || m_waiting_for_connect ) {
return false;
}
m_sock->encode();
if( !putClassAd( m_sock, msg ) || !m_sock->end_of_message() ) {
Disconnected();
return false;
}
return true;
}
bool
ScheddNegotiate::sendJobInfo(Sock *sock)
{
// The Negotiator wants us to send it a job.
sock->encode();
if( m_current_job_id.cluster == -1 && !nextJob() ) {
if( !sock->snd_int(NO_MORE_JOBS,TRUE) ) {
dprintf( D_ALWAYS, "Can't send NO_MORE_JOBS to mgr\n" );
return false;
}
m_negotiation_finished = true;
return true;
}
if( !sock->put(JOB_INFO) ) {
dprintf( D_ALWAYS, "Can't send JOB_INFO to mgr\n" );
return false;
}
// request match diagnostics
m_current_job_ad.Assign(ATTR_WANT_MATCH_DIAGNOSTICS, true);
// Send the ad to the negotiator
if( !putClassAd(sock, m_current_job_ad) ) {
dprintf( D_ALWAYS,
"Can't send job ad to mgr\n" );
sock->end_of_message();
return false;
}
if( !sock->end_of_message() ) {
dprintf( D_ALWAYS,
"Can't send job eom to mgr\n" );
return false;
}
m_current_resources_delivered = 0;
m_current_resources_requested = 1;
m_current_job_ad.LookupInteger(ATTR_RESOURCE_REQUEST_COUNT,m_current_resources_requested);
dprintf( D_FULLDEBUG,
"Sent job %d.%d (autocluster=%d resources_requested=%d) to the negotiator\n",
m_current_job_id.cluster, m_current_job_id.proc,
m_current_auto_cluster_id, m_current_resources_requested );
return true;
}
int
CCBListener::ReverseConnected(Stream *stream)
{
Sock *sock = (Sock *)stream;
ClassAd *msg_ad = (ClassAd *)daemonCore->GetDataPtr();
ASSERT( msg_ad );
if( sock ) {
daemonCore->Cancel_Socket( sock );
}
if( !sock || !sock->is_connected() ) {
ReportReverseConnectResult(msg_ad,false,"failed to connect");
}
else {
// The reverse-connect protocol is designed to look like a
// raw cedar command, in case the thing we are connecting
// to is a cedar command socket.
sock->encode();
int cmd = CCB_REVERSE_CONNECT;
if( !sock->put(cmd) ||
!putClassAd( sock, *msg_ad ) ||
!sock->end_of_message() )
{
ReportReverseConnectResult(msg_ad,false,"failure writing reverse connect command");
}
else {
((ReliSock*)sock)->isClient(false);
daemonCore->HandleReqAsync(sock);
sock = NULL; // daemonCore took ownership of sock
ReportReverseConnectResult(msg_ad,true);
}
}
delete msg_ad;
if( sock ) {
delete sock;
}
decRefCount(); // we incremented ref count when setting up callback
return KEEP_STREAM;
}
void
CCBServer::RequestReply( Sock *sock, bool success, char const *error_msg, CCBID request_cid, CCBID target_cid )
{
if( success && sock->readReady() ) {
// the client must have disconnected (which is expected if
// the client has already received the reversed connection)
return;
}
ClassAd msg;
msg.Assign( ATTR_RESULT, success );
msg.Assign( ATTR_ERROR_STRING, error_msg );
sock->encode();
if( !putClassAd( sock, msg ) || !sock->end_of_message() ) {
// Would like to be completely quiet if success and the
// client has disconnected, since this is normal; however,
// the above write operations will generate noise when
// they fail, so at least in FULLDEBUG, we explain what's
// going on. Note that most of the time, we should not get
// here for successful requests, because we either observe
// the client disconnect earlier, or the above check on
// the socket catches it. Why bother sending a reply on
// success at all? Because if the client has not yet
// seen the reverse connect and we just disconnect without
// telling it the request was successful, then it will
// think something has gone wrong.
dprintf(success ? D_FULLDEBUG : D_ALWAYS,
"CCB: failed to send result (%s) for request id %lu "
"from %s requesting a reversed connection to target daemon "
"with ccbid %lu: %s %s\n",
success ? "request succeeded" : "request failed",
request_cid,
sock->peer_description(),
target_cid,
error_msg,
success ? "(since the request was successful, it is expected "
"that the client may disconnect before receiving "
"results)" : "" );
}
}
void
CCBServer::SendHeartbeatResponse( CCBTarget *target )
{
Sock *sock = target->getSock();
ClassAd msg;
msg.Assign( ATTR_COMMAND, ALIVE );
sock->encode();
if( !putClassAd( sock, msg ) || !sock->end_of_message() ) {
dprintf(D_ALWAYS,
"CCB: failed to send heartbeat to target "
"daemon %s with ccbid %lu\n",
target->getSock()->peer_description(),
target->getCCBID());
RemoveTarget( target );
return;
}
dprintf(D_FULLDEBUG,"CCB: sent heartbeat to target %s\n",
sock->peer_description());
}
int
SendSpoolFileIfNeeded( ClassAd& ad )
{
int rval = -1;
CurrentSysCall = CONDOR_SendSpoolFileIfNeeded;
qmgmt_sock->encode();
neg_on_error( qmgmt_sock->code(CurrentSysCall) );
neg_on_error( putClassAd(qmgmt_sock, ad) );
neg_on_error( qmgmt_sock->end_of_message() );
qmgmt_sock->decode();
neg_on_error( qmgmt_sock->code(rval) );
if( rval < 0 ) {
neg_on_error( qmgmt_sock->code(terrno) );
neg_on_error( qmgmt_sock->end_of_message() );
errno = terrno;
return rval;
}
neg_on_error( qmgmt_sock->end_of_message() );
return rval;
}
// I'm going to ask the schedd for where I can put the files for the jobs I've
// specified. The schedd is going to respond with A) a message telling me it
// has the answer right away, or B) an answer telling me I have to wait
// an unknown length of time for the schedd to schedule me a place to put it.
bool
DCSchedd::requestSandboxLocation(ClassAd *reqad, ClassAd *respad,
CondorError * errstack)
{
ReliSock rsock;
int will_block;
ClassAd status_ad;
rsock.timeout(20); // years of research... :)
if( ! rsock.connect(_addr) ) {
dprintf( D_ALWAYS, "DCSchedd::requestSandboxLocation(): "
"Failed to connect to schedd (%s)\n", _addr );
return false;
}
if( ! startCommand(REQUEST_SANDBOX_LOCATION, (Sock*)&rsock, 0,
errstack) ) {
dprintf( D_ALWAYS, "DCSchedd::requestSandboxLocation(): "
"Failed to send command (REQUEST_SANDBOX_LOCATION) "
"to schedd (%s)\n", _addr );
return false;
}
// First, if we're not already authenticated, force that now.
if (!forceAuthentication( &rsock, errstack )) {
dprintf( D_ALWAYS, "DCSchedd: authentication failure: %s\n",
errstack->getFullText().c_str() );
return false;
}
rsock.encode();
///////////////////////////////////////////////////////////////////////
// Send my sandbox location request packet to the schedd.
///////////////////////////////////////////////////////////////////////
// This request ad will either contain:
// ATTR_TREQ_PEER_VERSION
// ATTR_TREQ_HAS_CONSTRAINT
// ATTR_TREQ_JOBID_LIST
// ATTR_TREQ_FTP
//
// OR
//
// ATTR_TREQ_DIRECTION
// ATTR_TREQ_PEER_VERSION
// ATTR_TREQ_HAS_CONSTRAINT
// ATTR_TREQ_CONSTRAINT
// ATTR_TREQ_FTP
dprintf(D_ALWAYS, "Sending request ad.\n");
if (putClassAd(&rsock, *reqad) != 1) {
dprintf(D_ALWAYS,"DCSchedd:requestSandboxLocation(): "
"Can't send reqad to the schedd\n");
return false;
}
rsock.end_of_message();
rsock.decode();
///////////////////////////////////////////////////////////////////////
// Read back a response ad which will tell me which jobs the schedd
// said I could modify and whether or not I'm am going to have to block
// before getting the payload of the transferd location/capability ad.
///////////////////////////////////////////////////////////////////////
// This status ad will contain
// ATTR_TREQ_INVALID_REQUEST (set to true)
// ATTR_TREQ_INVALID_REASON
//
// OR
// ATTR_TREQ_INVALID_REQUEST (set to false)
// ATTR_TREQ_JOBID_ALLOW_LIST
// ATTR_TREQ_JOBID_DENY_LIST
// ATTR_TREQ_WILL_BLOCK
dprintf(D_ALWAYS, "Receiving status ad.\n");
if (getClassAd(&rsock, status_ad) == false) {
dprintf(D_ALWAYS, "Schedd closed connection to me. Aborting sandbox "
"submission.\n");
return false;
}
rsock.end_of_message();
status_ad.LookupInteger(ATTR_TREQ_WILL_BLOCK, will_block);
dprintf(D_ALWAYS, "Client will %s\n", will_block==1?"block":"not block");
if (will_block == 1) {
// set to 20 minutes.
rsock.timeout(60*20);
}
///////////////////////////////////////////////////////////////////////
// Read back the payload ad from the schedd about the transferd location
// and capability string I can use for the fileset I wish to transfer.
///////////////////////////////////////////////////////////////////////
// read back the response ad from the schedd which contains a
// td sinful string, and a capability. These represent my ability to
// read/write a certain fileset somewhere.
// This response ad from the schedd will contain:
//
// ATTR_TREQ_INVALID_REQUEST (set to true)
// ATTR_TREQ_INVALID_REASON
//
// OR
//
// ATTR_TREQ_INVALID_REQUEST (set to false)
// ATTR_TREQ_CAPABILITY
// ATTR_TREQ_TD_SINFUL
// ATTR_TREQ_JOBID_ALLOW_LIST
dprintf(D_ALWAYS, "Receiving response ad.\n");
if (getClassAd(&rsock, *respad) != true) {
dprintf(D_ALWAYS,"DCSchedd:requestSandboxLocation(): "
"Can't receive respond ad from the schedd\n");
return false;
}
rsock.end_of_message();
return true;
}
int
DCStartd::activateClaim( ClassAd* job_ad, int starter_version,
ReliSock** claim_sock_ptr )
{
int reply;
dprintf( D_FULLDEBUG, "Entering DCStartd::activateClaim()\n" );
setCmdStr( "activateClaim" );
if( claim_sock_ptr ) {
// our caller wants a pointer to the socket we used to
// successfully activate the claim. right now, set it to
// NULL to signify error, and if everything works out,
// we'll give them a pointer to the real object.
*claim_sock_ptr = NULL;
}
if( ! claim_id ) {
newError( CA_INVALID_REQUEST,
"DCStartd::activateClaim: called with NULL claim_id, failing" );
return CONDOR_ERROR;
}
// if this claim is associated with a security session
ClaimIdParser cidp(claim_id);
char const *sec_session = cidp.secSessionId();
Sock* tmp;
tmp = startCommand( ACTIVATE_CLAIM, Stream::reli_sock, 20, NULL, NULL, false, sec_session );
if( ! tmp ) {
newError( CA_COMMUNICATION_ERROR,
"DCStartd::activateClaim: Failed to send command ACTIVATE_CLAIM to the startd" );
return CONDOR_ERROR;
}
if( ! tmp->put_secret(claim_id) ) {
newError( CA_COMMUNICATION_ERROR,
"DCStartd::activateClaim: Failed to send ClaimId to the startd" );
delete tmp;
return CONDOR_ERROR;
}
if( ! tmp->code(starter_version) ) {
newError( CA_COMMUNICATION_ERROR,
"DCStartd::activateClaim: Failed to send starter_version to the startd" );
delete tmp;
return CONDOR_ERROR;
}
if( ! putClassAd(tmp, *job_ad) ) {
newError( CA_COMMUNICATION_ERROR,
"DCStartd::activateClaim: Failed to send job ClassAd to the startd" );
delete tmp;
return CONDOR_ERROR;
}
if( ! tmp->end_of_message() ) {
newError( CA_COMMUNICATION_ERROR,
"DCStartd::activateClaim: Failed to send EOM to the startd" );
delete tmp;
return CONDOR_ERROR;
}
// Now, try to get the reply
tmp->decode();
if( !tmp->code(reply) || !tmp->end_of_message()) {
std::string err = "DCStartd::activateClaim: ";
err += "Failed to receive reply from ";
err += _addr ? _addr : "NULL";
newError( CA_COMMUNICATION_ERROR, err.c_str() );
delete tmp;
return CONDOR_ERROR;
}
dprintf( D_FULLDEBUG, "DCStartd::activateClaim: "
"successfully sent command, reply is: %d\n", reply );
if( reply == OK && claim_sock_ptr ) {
*claim_sock_ptr = (ReliSock*)tmp;
} else {
// in any other case, we're going to leak this ReliSock
// object if we don't delete it here...
delete tmp;
}
return reply;
}
bool
DCTransferQueue::RequestTransferQueueSlot(bool downloading,filesize_t sandbox_size,char const *fname,char const *jobid,char const *queue_user,int timeout,MyString &error_desc)
{
ASSERT(fname);
ASSERT(jobid);
if( GoAheadAlways( downloading ) ) {
m_xfer_downloading = downloading;
m_xfer_fname = fname;
m_xfer_jobid = jobid;
return true;
}
CheckTransferQueueSlot();
if( m_xfer_queue_sock ) {
// A request has already been made.
// Currently, this is a no-op, because any upload/download slot
// is as good as any other. In the future, there may be
// different queues for different paths.
ASSERT( m_xfer_downloading == downloading );
m_xfer_fname = fname;
m_xfer_jobid = jobid;
return true;
}
time_t started = time(NULL);
CondorError errstack;
// Our caller has to finish this operation in the specified
// amount of time or risk not responding to the file transfer
// peer in time, so ignore the timeout multiplier and set the
// timeout exactly as specified.
m_xfer_queue_sock = reliSock( timeout, 0, &errstack, false, true );
if( !m_xfer_queue_sock ) {
formatstr(m_xfer_rejected_reason,
"Failed to connect to transfer queue manager for job %s (%s): %s.",
jobid, fname, errstack.getFullText().c_str() );
error_desc = m_xfer_rejected_reason;
dprintf(D_ALWAYS,"%s\n",m_xfer_rejected_reason.c_str());
return false;
}
if( timeout ) {
timeout -= time(NULL)-started;
if( timeout <= 0 ) {
timeout = 1;
}
}
if (IsDebugLevel(D_COMMAND)) {
int cmd = TRANSFER_QUEUE_REQUEST;
dprintf (D_COMMAND, "DCTransferQueue::RequestTransferQueueSlot(%s,...) making connection to %s\n", getCommandStringSafe(cmd), _addr ? _addr : "NULL");
}
bool connected = startCommand(
TRANSFER_QUEUE_REQUEST, m_xfer_queue_sock, timeout, &errstack );
if( !connected )
{
delete m_xfer_queue_sock;
m_xfer_queue_sock = NULL;
formatstr(m_xfer_rejected_reason,
"Failed to initiate transfer queue request for job %s (%s): %s.",
jobid, fname, errstack.getFullText().c_str() );
error_desc = m_xfer_rejected_reason;
dprintf(D_ALWAYS,"%s\n",m_xfer_rejected_reason.c_str());
return false;
}
m_xfer_downloading = downloading;
m_xfer_fname = fname;
m_xfer_jobid = jobid;
ClassAd msg;
msg.Assign(ATTR_DOWNLOADING,downloading);
msg.Assign(ATTR_FILE_NAME,fname);
msg.Assign(ATTR_JOB_ID,jobid);
msg.Assign(ATTR_USER,queue_user);
msg.Assign(ATTR_SANDBOX_SIZE,sandbox_size);
m_xfer_queue_sock->encode();
if( !putClassAd(m_xfer_queue_sock, msg) || !m_xfer_queue_sock->end_of_message() )
{
formatstr(m_xfer_rejected_reason,
"Failed to write transfer request to %s for job %s "
"(initial file %s).",
m_xfer_queue_sock->peer_description(),
m_xfer_jobid.c_str(), m_xfer_fname.c_str());
error_desc = m_xfer_rejected_reason;
dprintf(D_ALWAYS,"%s\n",m_xfer_rejected_reason.c_str());
return false;
}
m_xfer_queue_sock->decode();
// Request has been initiated. Now sender should call
// PollForTransferQueueSlot() to get response.
m_xfer_queue_pending = true;
return true;
}
int
part_send_job(
int test_starter,
char *host,
int &reason,
char *capability,
char * /*schedd*/,
PROC *proc,
int &sd1,
int &sd2,
char **name)
{
int reply;
ReliSock *sock = NULL;
StartdRec stRec;
PORTS ports;
bool done = false;
int retry_delay = 3;
int num_retries = 0;
// make sure we have the job classad
InitJobAd(proc->id.cluster, proc->id.proc);
while( !done ) {
Daemon startd(DT_STARTD, host, NULL);
if (!(sock = (ReliSock*)startd.startCommand ( ACTIVATE_CLAIM, Stream::reli_sock, 90))) {
dprintf( D_ALWAYS, "startCommand(ACTIVATE_CLAIM) to startd failed\n");
goto returnfailure;
}
// Send the capability
ClaimIdParser idp( capability );
dprintf(D_FULLDEBUG, "send capability %s\n", idp.publicClaimId() );
if( !sock->put_secret(capability) ) {
dprintf( D_ALWAYS, "sock->put(\"%s\") failed\n",idp.publicClaimId());
goto returnfailure;
}
// Send the starter number
if( test_starter ) {
dprintf( D_ALWAYS, "Requesting Alternate Starter %d\n", test_starter );
} else {
dprintf( D_ALWAYS, "Requesting Primary Starter\n" );
}
if( !sock->code(test_starter) ) {
dprintf( D_ALWAYS, "sock->code(%d) failed\n", test_starter );
goto returnfailure;
}
// Send the job info
if( !putClassAd(sock, *JobAd) ) {
dprintf( D_ALWAYS, "failed to send job ad\n" );
goto returnfailure;
}
if( !sock->end_of_message() ) {
dprintf( D_ALWAYS, "failed to send message to startd\n" );
goto returnfailure;
}
// We're done sending. Now, get the reply.
sock->decode();
if( !sock->code(reply) || !sock->end_of_message() ) {
dprintf( D_ALWAYS, "failed to receive reply from startd\n" );
goto returnfailure;
}
switch( reply ) {
case OK:
dprintf( D_ALWAYS, "Shadow: Request to run a job was ACCEPTED\n" );
done = true;
break;
case NOT_OK:
dprintf( D_ALWAYS, "Shadow: Request to run a job was REFUSED\n");
goto returnfailure;
break;
case CONDOR_TRY_AGAIN:
num_retries++;
dprintf( D_ALWAYS,
"Shadow: Request to run a job was TEMPORARILY REFUSED\n" );
if( num_retries > 20 ) {
dprintf( D_ALWAYS, "Shadow: Too many retries, giving up.\n" );
goto returnfailure;
}
delete sock;
dprintf( D_ALWAYS,
"Shadow: will try again in %d seconds\n", retry_delay );
sleep( retry_delay );
break;
default:
dprintf(D_ALWAYS,"Unknown reply from startd for command ACTIVATE_CLAIM\n");
dprintf(D_ALWAYS,"Shadow: Request to run a job was REFUSED\n");
goto returnfailure;
break;
}
}
/* start flock : dhruba */
sock->decode();
memset( &stRec, '\0', sizeof(stRec) );
if( !sock->code(stRec) || !sock->end_of_message() ) {
dprintf(D_ALWAYS, "Can't read reply from startd.\n");
goto returnfailure;
}
ports = stRec.ports;
if( stRec.ip_addr ) {
host = stRec.server_name;
if(name) {
*name = strdup(stRec.server_name);
}
dprintf(D_FULLDEBUG,
"host = %s inet_addr = 0x%x port1 = %d port2 = %d\n",
host, stRec.ip_addr,ports.port1, ports.port2
);
} else {
dprintf(D_FULLDEBUG,
"host = %s port1 = %d port2 = %d\n",
host, ports.port1, ports.port2
);
}
if( ports.port1 == 0 ) {
dprintf( D_ALWAYS, "Shadow: Request to run a job on %s was REFUSED\n",
host );
goto returnfailure;
}
/* end flock ; dhruba */
// We don't use the server_name in the StartdRec, because our
// DNS query may fail or may give us the wrong IP address
// (either because it's stale or because we're talking to a
// machine with multiple network interfaces). Sadly, we can't
// use the ip_addr either, because the startd doesn't send it in
// the correct byte ordering on little-endian machines. So, we
// grab the IP address from the ReliSock, since we konw the
// startd always uses the same IP address for all of its
// communication.
char sinfulstring[SINFUL_STRING_BUF_SIZE];
generate_sinful(sinfulstring, SINFUL_STRING_BUF_SIZE, sock->peer_ip_str(), ports.port1);
if( (sd1 = do_connect(sinfulstring, (char *)0, (u_short)ports.port1)) < 0 ) {
dprintf( D_ALWAYS, "failed to connect to scheduler on %s\n", sinfulstring );
goto returnfailure;
}
generate_sinful(sinfulstring, SINFUL_STRING_BUF_SIZE, sock->peer_ip_str(), ports.port2);
if( (sd2 = do_connect(sinfulstring, (char *)0, (u_short)ports.port2)) < 0 ) {
dprintf( D_ALWAYS, "failed to connect to scheduler on %s\n", sinfulstring );
close(sd1);
goto returnfailure;
}
delete sock;
sock = NULL;
if ( stRec.server_name ) {
free( stRec.server_name );
}
return 0;
returnfailure:
reason = JOB_NOT_STARTED;
delete sock;
return -1;
}
/* Continue reading from rsock the rest of the protcol for this encapsulation
method */
int
TransferD::accept_transfer_request_encapsulation_old_classads(Stream *sock)
{
int i;
ClassAd *ad = NULL;
TransferRequest *treq = NULL;
MyString cap;
ClassAd respad;
dprintf(D_ALWAYS,
"Entering "
"TransferD::accept_transfer_request_encapsulation_old_classads()\n");
sock->decode();
/////////////////////////////////////////////////////////////////////////
// Accept the transfer request from the schedd.
/////////////////////////////////////////////////////////////////////////
/* read the transfer request header packet upon construction */
ad = new ClassAd();
if (getClassAd(sock, *ad) == false) {
// XXX don't fail here, just go back to daemoncore
EXCEPT("XXX Couldn't init initial ad from stream!");
}
sock->end_of_message();
dprintf(D_ALWAYS, "Read treq header.\n");
// initialize the header information of the TransferRequest object.
treq = new TransferRequest(ad);
if (treq == NULL) {
EXCEPT("Out of memory!");
}
/* read the information packet which describes the rest of the protocol */
if (treq->get_num_transfers() <= 0) {
EXCEPT("Protocol error!");
}
// read all the work ads associated with this TransferRequest
for (i = 0; i < treq->get_num_transfers(); i++) {
ad = new ClassAd();
if (ad == NULL) {
EXCEPT("Out of memory!");
}
if (getClassAd(sock, *ad) == false) {
EXCEPT("Expected %d transfer job ads, got %d instead.",
treq->get_num_transfers(), i);
}
sock->end_of_message();
dprintf(D_ALWAYS, "Read treq job ad[%d].\n", i);
treq->append_task(ad);
}
sock->end_of_message();
sock->encode();
/////////////////////////////////////////////////////////////////////////
// See if I can honor this request's protocol choice
/////////////////////////////////////////////////////////////////////////
switch(treq->get_xfer_protocol())
{
case FTP_CFTP: // Transferd may use the FileTransfer Object protocol
respad.Assign(ATTR_TREQ_INVALID_REQUEST, FALSE);
break;
default:
dprintf(D_ALWAYS, "Transfer Request uses an unsupported file "
"transfer protocol. Rejecting it.\n");
// Currently, I don't support anything else....
respad.Assign(ATTR_TREQ_INVALID_REQUEST, TRUE);
respad.Assign(ATTR_TREQ_INVALID_REASON,
"Transferd doesn't support client required file transfer "
"protocol.");
// tell the schedd we don't want to do this request
putClassAd(sock, respad);
sock->end_of_message();
delete treq;
// wait for the next request to come in....
return KEEP_STREAM;
break;
}
/////////////////////////////////////////////////////////////////////////
// Create a capability for this request, making sure it is unique to all
// rest of them, then send it back.
/////////////////////////////////////////////////////////////////////////
cap = gen_capability();
treq->set_capability(cap);
respad.Assign(ATTR_TREQ_CAPABILITY, cap);
dprintf(D_ALWAYS, "Assigned capability to treq: %s.\n", cap.Value());
// This respose ad will contain:
//
// ATTR_TREQ_INVALID_REQUEST (set to true)
// ATTR_TREQ_INVALID_REASON
//
// OR
//
// ATTR_TREQ_INVALID_REQUEST (set to false)
// ATTR_TREQ_CAPABILITY
//
putClassAd(sock, respad);
sock->end_of_message();
dprintf(D_ALWAYS, "Reported capability back to schedd.\n");
// If nothing times out or broke connection, and I think the schedd has
// gotten the above information, then queue this request to deal with
// at the appropriate time.
m_treqs.insert(cap, treq);
// get ready to read another treq.
sock->decode();
dprintf(D_ALWAYS, "Waiting for another transfer request from schedd.\n");
dprintf(D_ALWAYS,
"Leaving "
"TransferD::accept_transfer_request_encapsulation_old_classads()\n");
return KEEP_STREAM;
}
// Read history from a remote schedd
static void readHistoryRemote(classad::ExprTree *constraintExpr)
{
printHeader();
if(longformat && use_xml) {
std::string out;
AddClassAdXMLFileHeader(out);
printf("%s\n", out.c_str());
}
classad::ClassAd ad;
classad::ExprList *projList(new classad::ExprList());
classad::ExprTree *projTree = static_cast<classad::ExprTree*>(projList);
ad.Insert(ATTR_PROJECTION, projTree);
ad.Insert(ATTR_REQUIREMENTS, constraintExpr);
ad.InsertAttr(ATTR_NUM_MATCHES, specifiedMatch <= 0 ? -1 : specifiedMatch);
DCSchedd schedd(g_name.size() ? g_name.c_str() : NULL, g_pool.size() ? g_pool.c_str() : NULL);
if (!schedd.locate(Daemon::LOCATE_FOR_LOOKUP)) {
fprintf(stderr, "Unable to locate remote schedd (name=%s, pool=%s).\n", g_name.c_str(), g_pool.c_str());
exit(1);
}
Sock* sock;
if (!(sock = schedd.startCommand(QUERY_SCHEDD_HISTORY, Stream::reli_sock, 0))) {
fprintf(stderr, "Unable to send history command to remote schedd;\n"
"Typically, either the schedd is not responding, does not authorize you, or does not support remote history.\n");
exit(1);
}
classad_shared_ptr<Sock> sock_sentry(sock);
if (!putClassAd(sock, ad) || !sock->end_of_message()) {
fprintf(stderr, "Unable to send request to remote schedd; likely a server or network error.\n");
exit(1);
}
while (true) {
compat_classad::ClassAd ad;
if (!getClassAd(sock, ad)) {
fprintf(stderr, "Failed to recieve remote ad.\n");
exit(1);
}
long long intVal;
if (ad.EvaluateAttrInt(ATTR_OWNER, intVal) && (intVal == 0)) { // Last ad.
if (!sock->end_of_message()) {
fprintf(stderr, "Unable to close remote socket.\n");
}
sock->close();
std::string errorMsg;
if (ad.EvaluateAttrInt(ATTR_ERROR_CODE, intVal) && intVal && ad.EvaluateAttrString(ATTR_ERROR_STRING, errorMsg)) {
fprintf(stderr, "Error %lld: %s\n", intVal, errorMsg.c_str());
exit(intVal);
}
if (ad.EvaluateAttrInt("MalformedAds", intVal) && intVal) {
fprintf(stderr, "Remote side had parse errors on history file");
exit(1);
}
if (!ad.EvaluateAttrInt(ATTR_NUM_MATCHES, intVal) || (intVal != matchCount)) {
fprintf(stderr, "Client and server do not agree on number of ads sent;\n"
"Indicates lost network packets or an internal error\n");
exit(1);
}
break;
}
matchCount++;
printJob(ad);
}
if(longformat && use_xml) {
std::string out;
AddClassAdXMLFileFooter(out);
printf("%s\n", out.c_str());
}
}
int
do_Q_request(ReliSock *syscall_sock,bool &may_fork)
{
int request_num = -1;
int rval;
syscall_sock->decode();
assert( syscall_sock->code(request_num) );
dprintf(D_SYSCALLS, "Got request #%d\n", request_num);
switch( request_num ) {
case CONDOR_InitializeConnection:
{
// dprintf( D_ALWAYS, "InitializeConnection()\n" );
bool authenticated = true;
// Authenticate socket, if not already done by daemonCore
if( !syscall_sock->triedAuthentication() ) {
if( IsDebugLevel(D_SECURITY) ) {
MyString methods;
SecMan::getAuthenticationMethods( WRITE, &methods );
dprintf(D_SECURITY,"Calling authenticate(%s) in qmgmt_receivers\n", methods.Value());
}
CondorError errstack;
if( ! SecMan::authenticate_sock(syscall_sock, WRITE, &errstack) ) {
// Failed to authenticate
dprintf( D_ALWAYS, "SCHEDD: authentication failed: %s\n",
errstack.getFullText().c_str() );
authenticated = false;
}
}
if ( authenticated ) {
InitializeConnection( syscall_sock->getOwner(),
syscall_sock->getDomain() );
} else {
InitializeConnection( NULL, NULL );
}
return 0;
}
case CONDOR_InitializeReadOnlyConnection:
{
// dprintf( D_ALWAYS, "InitializeReadOnlyConnection()\n" );
// Since InitializeConnection() does nothing, and we need
// to record the fact that this is a read-only connection,
// but we have to do it in the socket (since we don't have
// any other persistent data structure, and it's probably
// the right place anyway), set the FQU.
//
// We need to record if this is a read-only connection so that
// we can avoid expanding $$ in GetJobAd; simply checking if the
// connection is authenticated isn't sufficient, because the
// security session cache means that read-only connection could
// be authenticated by a previous authenticated connection from
// the same address (when using host-based security) less than
// the expiration period ago.
syscall_sock->setFullyQualifiedUser( "read-only" );
// same as InitializeConnection but no authenticate()
InitializeConnection( NULL, NULL );
may_fork = true;
return 0;
}
case CONDOR_SetEffectiveOwner:
{
MyString owner;
int terrno;
assert( syscall_sock->get(owner) );
assert( syscall_sock->end_of_message() );
rval = QmgmtSetEffectiveOwner( owner.Value() );
terrno = errno;
syscall_sock->encode();
assert( syscall_sock->code(rval) );
if( rval < 0 ) {
assert( syscall_sock->code(terrno) );
}
assert( syscall_sock->end_of_message() );
char const *fqu = syscall_sock->getFullyQualifiedUser();
dprintf(D_SYSCALLS, "\tSetEffectiveOwner\n");
dprintf(D_SYSCALLS, "\tauthenticated user = '******'\n", fqu ? fqu : "");
dprintf(D_SYSCALLS, "\trequested owner = '%s'\n", owner.Value());
dprintf(D_SYSCALLS, "\trval %d, errno %d\n", rval, terrno);
return 0;
}
case CONDOR_NewCluster:
{
int terrno;
assert( syscall_sock->end_of_message() );;
errno = 0;
rval = NewCluster( );
terrno = errno;
dprintf(D_SYSCALLS,
"\tNewCluster: rval = %d, errno = %d\n",rval,terrno );
if ( rval > 0 ) {
dprintf( D_AUDIT, *syscall_sock,
"Submitting new job %d.0\n", rval );
}
syscall_sock->encode();
assert( syscall_sock->code(rval) );
if( rval < 0 ) {
assert( syscall_sock->code(terrno) );
}
assert( syscall_sock->end_of_message() );;
dprintf(D_FULLDEBUG,"schedd: NewCluster rval %d errno %d\n",rval,terrno);
return 0;
}
case CONDOR_NewProc:
{
int cluster_id = -1;
int terrno;
assert( syscall_sock->code(cluster_id) );
dprintf( D_SYSCALLS, " cluster_id = %d\n", cluster_id );
assert( syscall_sock->end_of_message() );;
errno = 0;
rval = NewProc( cluster_id );
terrno = errno;
dprintf( D_SYSCALLS, "\trval = %d, errno = %d\n", rval, terrno );
if ( rval > 0 ) {
dprintf( D_AUDIT, *syscall_sock,
"Submitting new job %d.%d\n", cluster_id, rval );
}
syscall_sock->encode();
assert( syscall_sock->code(rval) );
if( rval < 0 ) {
assert( syscall_sock->code(terrno) );
}
assert( syscall_sock->end_of_message() );;
dprintf(D_FULLDEBUG,"schedd: NewProc rval %d errno %d\n",rval,terrno);
return 0;
}
case CONDOR_DestroyProc:
{
int cluster_id = -1;
int proc_id = -1;
int terrno;
assert( syscall_sock->code(cluster_id) );
dprintf( D_SYSCALLS, " cluster_id = %d\n", cluster_id );
assert( syscall_sock->code(proc_id) );
dprintf( D_SYSCALLS, " proc_id = %d\n", proc_id );
assert( syscall_sock->end_of_message() );;
errno = 0;
rval = DestroyProc( cluster_id, proc_id );
terrno = errno;
dprintf( D_SYSCALLS, "\trval = %d, errno = %d\n", rval, terrno );
syscall_sock->encode();
assert( syscall_sock->code(rval) );
if( rval < 0 ) {
assert( syscall_sock->code(terrno) );
}
assert( syscall_sock->end_of_message() );;
dprintf(D_FULLDEBUG,"schedd: DestroyProc cluster %d proc %d rval %d errno %d\n",cluster_id,proc_id,rval,terrno);
return 0;
}
case CONDOR_DestroyCluster:
{
int cluster_id = -1;
int terrno;
assert( syscall_sock->code(cluster_id) );
dprintf( D_SYSCALLS, " cluster_id = %d\n", cluster_id );
assert( syscall_sock->end_of_message() );;
errno = 0;
rval = DestroyCluster( cluster_id );
terrno = errno;
dprintf( D_SYSCALLS, "\trval = %d, errno = %d\n", rval, terrno );
syscall_sock->encode();
assert( syscall_sock->code(rval) );
if( rval < 0 ) {
assert( syscall_sock->code(terrno) );
}
assert( syscall_sock->end_of_message() );;
return 0;
}
#if 0
case CONDOR_DestroyClusterByConstraint:
{
char *constraint=NULL;
int terrno;
assert( syscall_sock->code(constraint) );
assert( syscall_sock->end_of_message() );;
errno = 0;
rval = DestroyClusterByConstraint( constraint );
terrno = errno;
dprintf( D_SYSCALLS, "\trval = %d, errno = %d\n", rval, terrno );
syscall_sock->encode();
assert( syscall_sock->code(rval) );
if( rval < 0 ) {
assert( syscall_sock->code(terrno) );
}
free( (char *)constraint );
assert( syscall_sock->end_of_message() );;
return 0;
}
#endif
case CONDOR_SetAttributeByConstraint:
case CONDOR_SetAttributeByConstraint2:
{
char *attr_name=NULL;
char *attr_value=NULL;
char *constraint=NULL;
int terrno;
SetAttributeFlags_t flags = 0;
assert( syscall_sock->code(constraint) );
dprintf( D_SYSCALLS, " constraint = %s\n",constraint);
assert( syscall_sock->code(attr_value) );
assert( syscall_sock->code(attr_name) );
if( request_num == CONDOR_SetAttributeByConstraint2 ) {
assert( syscall_sock->code( flags ) );
}
assert( syscall_sock->end_of_message() );;
if (strcmp (attr_name, ATTR_MYPROXY_PASSWORD) == 0) {
errno = 0;
dprintf( D_SYSCALLS, "SetAttributeByConstraint (MyProxyPassword) not supported...\n");
rval = 0;
terrno = errno;
} else {
errno = 0;
rval = SetAttributeByConstraint( constraint, attr_name, attr_value, flags );
terrno = errno;
dprintf( D_SYSCALLS, "\trval = %d, errno = %d\n", rval, terrno );
if ( rval == 0 ) {
dprintf( D_AUDIT, *syscall_sock,
"Set Attribute By Constraint %s, "
"%s = %s\n",
constraint, attr_name, attr_value);
}
}
syscall_sock->encode();
assert( syscall_sock->code(rval) );
if( rval < 0 ) {
assert( syscall_sock->code(terrno) );
}
free( (char *)constraint );
free( (char *)attr_value );
free( (char *)attr_name );
assert( syscall_sock->end_of_message() );;
return 0;
}
case CONDOR_SetAttribute:
case CONDOR_SetAttribute2:
{
int cluster_id = -1;
int proc_id = -1;
char *attr_name=NULL;
char *attr_value=NULL;
int terrno;
SetAttributeFlags_t flags = 0;
const char *users_username;
const char *condor_username;
assert( syscall_sock->code(cluster_id) );
dprintf( D_SYSCALLS, " cluster_id = %d\n", cluster_id );
assert( syscall_sock->code(proc_id) );
dprintf( D_SYSCALLS, " proc_id = %d\n", proc_id );
assert( syscall_sock->code(attr_value) );
assert( syscall_sock->code(attr_name) );
if( request_num == CONDOR_SetAttribute2 ) {
assert( syscall_sock->code( flags ) );
}
users_username = syscall_sock->getOwner();
condor_username = get_condor_username();
if (attr_name) dprintf(D_SYSCALLS,"\tattr_name = %s\n",attr_name);
if (attr_value) dprintf(D_SYSCALLS,"\tattr_value = %s\n",attr_value);
assert( syscall_sock->end_of_message() );;
// ckireyev:
// We do NOT want to include MyProxy password in the ClassAd (since it's a secret)
// I'm not sure if this is the best place to do this, but....
if (attr_name && attr_value && strcmp (attr_name, ATTR_MYPROXY_PASSWORD) == 0) {
errno = 0;
dprintf( D_SYSCALLS, "Got MyProxyPassword, stashing...\n");
rval = SetMyProxyPassword (cluster_id, proc_id, attr_value);
terrno = errno;
dprintf( D_SYSCALLS, "\trval = %d, errno = %d\n", rval, terrno );
}
else {
errno = 0;
rval = SetAttribute( cluster_id, proc_id, attr_name, attr_value, flags );
terrno = errno;
dprintf( D_SYSCALLS, "\trval = %d, errno = %d\n", rval, terrno );
// If we're modifying a previously-submitted job AND either
// the client's username is not HTCondor's (i.e. not a
// daemon) OR the client says we should log...
if( (cluster_id != active_cluster_num) && (rval == 0) &&
( strcmp(users_username, condor_username) || (flags & SHOULDLOG) ) ) {
dprintf( D_AUDIT, *syscall_sock,
"Set Attribute for job %d.%d, "
"%s = %s\n",
cluster_id, proc_id, attr_name, attr_value);
}
}
free( (char *)attr_value );
free( (char *)attr_name );
if( flags & SetAttribute_NoAck ) {
if( rval < 0 ) {
return -1;
}
}
else {
syscall_sock->encode();
assert( syscall_sock->code(rval) );
if( rval < 0 ) {
assert( syscall_sock->code(terrno) );
}
assert( syscall_sock->end_of_message() );
}
return 0;
}
case CONDOR_SetTimerAttribute:
{
int cluster_id = -1;
int proc_id = -1;
char *attr_name=NULL;
int duration = 0;
int terrno;
assert( syscall_sock->code(cluster_id) );
dprintf( D_SYSCALLS, " cluster_id = %d\n", cluster_id );
assert( syscall_sock->code(proc_id) );
dprintf( D_SYSCALLS, " proc_id = %d\n", proc_id );
assert( syscall_sock->code(attr_name) );
if (attr_name) dprintf(D_SYSCALLS,"\tattr_name = %s\n",attr_name);
assert( syscall_sock->code(duration) );
dprintf(D_SYSCALLS,"\tduration = %d\n",duration);
assert( syscall_sock->end_of_message() );;
errno = 0;
rval = SetTimerAttribute( cluster_id, proc_id, attr_name, duration );
terrno = errno;
dprintf( D_SYSCALLS, "\trval = %d, errno = %d\n", rval, terrno );
dprintf( D_AUDIT, *syscall_sock,
"Set Timer Attribute for job %d.%d, "
"attr_name = %s, duration = %d\n",
cluster_id, proc_id, attr_name, duration);
syscall_sock->encode();
assert( syscall_sock->code(rval) );
if( rval < 0 ) {
assert( syscall_sock->code(terrno) );
}
free( (char *)attr_name );
assert( syscall_sock->end_of_message() );;
return 0;
}
case CONDOR_BeginTransaction:
{
int terrno;
assert( syscall_sock->end_of_message() );;
errno = 0;
rval = 0; // BeginTransaction returns void (sigh), so always success
BeginTransaction( );
terrno = errno;
dprintf( D_SYSCALLS, "\trval = %d, errno = %d\n", rval, terrno );
syscall_sock->encode();
assert( syscall_sock->code(rval) );
if( rval < 0 ) {
assert( syscall_sock->code(terrno) );
}
assert( syscall_sock->end_of_message() );;
return 0;
}
case CONDOR_AbortTransaction:
{
int terrno;
assert( syscall_sock->end_of_message() );;
errno = 0;
rval = 0; // AbortTransaction returns void (sigh), so always success
AbortTransaction( );
terrno = errno;
dprintf( D_SYSCALLS, "\trval = %d, errno = %d\n", rval, terrno );
syscall_sock->encode();
assert( syscall_sock->code(rval) );
if( rval < 0 ) {
assert( syscall_sock->code(terrno) );
}
assert( syscall_sock->end_of_message() );;
return 0;
}
case CONDOR_CommitTransactionNoFlags:
case CONDOR_CommitTransaction:
{
int terrno;
int flags;
if( request_num == CONDOR_CommitTransaction ) {
assert( syscall_sock->code(flags) );
}
else {
flags = 0;
}
assert( syscall_sock->end_of_message() );;
errno = 0;
CondorError errstack;
rval = CheckTransaction( flags, & errstack );
terrno = errno;
dprintf( D_SYSCALLS, "\tflags = %d, rval = %d, errno = %d\n", flags, rval, terrno );
if( rval >= 0 ) {
errno = 0;
CommitTransaction( flags );
// CommitTransaction() never returns on failure
rval = 0;
terrno = errno;
dprintf( D_SYSCALLS, "\tflags = %d, rval = %d, errno = %d\n", flags, rval, terrno );
}
syscall_sock->encode();
assert( syscall_sock->code(rval) );
if( rval < 0 ) {
assert( syscall_sock->code(terrno) );
const CondorVersionInfo *vers = syscall_sock->get_peer_version();
if (vers && vers->built_since_version(8, 3, 4))
{
// Send a classad, for less backwards-incompatibility.
int code = 1;
const char * reason = "QMGMT rejected job submission.";
if( errstack.subsys() ) {
code = 2;
reason = errstack.message();
}
ClassAd reply;
reply.Assign( "ErrorCode", code );
reply.Assign( "ErrorReason", reason );
assert( putClassAd( syscall_sock, reply ) );
}
}
assert( syscall_sock->end_of_message() );;
return 0;
}
case CONDOR_GetAttributeFloat:
{
int cluster_id = -1;
int proc_id = -1;
char *attr_name=NULL;
float value = 0.0;
int terrno;
assert( syscall_sock->code(cluster_id) );
dprintf( D_SYSCALLS, " cluster_id = %d\n", cluster_id );
assert( syscall_sock->code(proc_id) );
dprintf( D_SYSCALLS, " proc_id = %d\n", proc_id );
assert( syscall_sock->code(attr_name) );
assert( syscall_sock->end_of_message() );;
errno = 0;
if( QmgmtMayAccessAttribute( attr_name ) ) {
rval = GetAttributeFloat( cluster_id, proc_id, attr_name, &value );
}
else {
rval = -1;
}
terrno = errno;
dprintf( D_SYSCALLS, "\trval = %d, errno = %d\n", rval, terrno );
syscall_sock->encode();
assert( syscall_sock->code(rval) );
if( rval < 0 ) {
assert( syscall_sock->code(terrno) );
}
if( rval >= 0 ) {
assert( syscall_sock->code(value) );
}
free( (char *)attr_name );
assert( syscall_sock->end_of_message() );;
return 0;
}
case CONDOR_GetAttributeInt:
{
int cluster_id = -1;
int proc_id = -1;
char *attr_name=NULL;
int value = 0;
int terrno;
assert( syscall_sock->code(cluster_id) );
dprintf( D_SYSCALLS, " cluster_id = %d\n", cluster_id );
assert( syscall_sock->code(proc_id) );
dprintf( D_SYSCALLS, " proc_id = %d\n", proc_id );
assert( syscall_sock->code(attr_name) );
dprintf( D_SYSCALLS, " attr_name = %s\n", attr_name );
assert( syscall_sock->end_of_message() );;
errno = 0;
if( QmgmtMayAccessAttribute( attr_name ) ) {
rval = GetAttributeInt( cluster_id, proc_id, attr_name, &value );
}
else {
rval = -1;
}
terrno = errno;
if (rval < 0) {
dprintf( D_SYSCALLS, "GetAttributeInt(%d, %d, %s) not found.\n",
cluster_id, proc_id, attr_name);
} else {
dprintf( D_SYSCALLS, " value: %d\n", value );
dprintf( D_SYSCALLS, "\trval = %d, errno = %d\n", rval, terrno );
}
syscall_sock->encode();
assert( syscall_sock->code(rval) );
if( rval < 0 ) {
assert( syscall_sock->code(terrno) );
}
if( rval >= 0 ) {
assert( syscall_sock->code(value) );
}
free( (char *)attr_name );
assert( syscall_sock->end_of_message() );;
return 0;
}
case CONDOR_GetAttributeString:
{
int cluster_id = -1;
int proc_id = -1;
char *attr_name=NULL;
char *value = NULL;
int terrno;
assert( syscall_sock->code(cluster_id) );
dprintf( D_SYSCALLS, " cluster_id = %d\n", cluster_id );
assert( syscall_sock->code(proc_id) );
dprintf( D_SYSCALLS, " proc_id = %d\n", proc_id );
assert( syscall_sock->code(attr_name) );
assert( syscall_sock->end_of_message() );;
errno = 0;
if( QmgmtMayAccessAttribute( attr_name ) ) {
rval = GetAttributeStringNew( cluster_id, proc_id, attr_name, &value );
}
else {
rval = -1;
}
terrno = errno;
dprintf( D_SYSCALLS, "\trval = %d, errno = %d\n", rval, terrno );
syscall_sock->encode();
assert( syscall_sock->code(rval) );
if( rval < 0 ) {
assert( syscall_sock->code(terrno) );
}
if( rval >= 0 ) {
assert( syscall_sock->code(value) );
}
free( (char *)value );
free( (char *)attr_name );
assert( syscall_sock->end_of_message() );;
return 0;
}
case CONDOR_GetAttributeExpr:
{
int cluster_id = -1;
int proc_id = -1;
char *attr_name=NULL;
int terrno;
assert( syscall_sock->code(cluster_id) );
dprintf( D_SYSCALLS, " cluster_id = %d\n", cluster_id );
assert( syscall_sock->code(proc_id) );
dprintf( D_SYSCALLS, " proc_id = %d\n", proc_id );
assert( syscall_sock->code(attr_name) );
assert( syscall_sock->end_of_message() );;
char *value = NULL;
errno = 0;
if( QmgmtMayAccessAttribute( attr_name ) ) {
rval = GetAttributeExprNew( cluster_id, proc_id, attr_name, &value );
}
else {
rval = -1;
}
terrno = errno;
dprintf( D_SYSCALLS, "\trval = %d, errno = %d\n", rval, terrno );
syscall_sock->encode();
if ( !syscall_sock->code(rval) ) {
free(value);
return -1;
}
if( rval < 0 ) {
if ( !syscall_sock->code(terrno) ) {
free(value);
return -1;
}
}
if( rval >= 0 ) {
if ( !syscall_sock->code(value) ) {
free(value);
return -1;
}
}
free( (char *)value );
free( (char *)attr_name );
assert( syscall_sock->end_of_message() );;
return 0;
}
case CONDOR_GetDirtyAttributes:
{
int cluster_id = -1;
int proc_id = -1;
ClassAd updates;
int terrno;
assert( syscall_sock->code(cluster_id) );
dprintf( D_SYSCALLS, " cluster_id = %d\n", cluster_id );
assert( syscall_sock->code(proc_id) );
dprintf( D_SYSCALLS, " proc_id = %d\n", proc_id );
assert( syscall_sock->end_of_message() );;
errno = 0;
rval = GetDirtyAttributes( cluster_id, proc_id, &updates );
terrno = errno;
dprintf( D_SYSCALLS, "\trval = %d, errno = %d\n", rval, terrno );
syscall_sock->encode();
if ( !syscall_sock->code(rval) ) {
return -1;
}
if( rval < 0 ) {
if ( !syscall_sock->code(terrno) ) {
return -1;
}
}
if( rval >= 0 ) {
assert( putClassAd(syscall_sock, updates) );
}
assert( syscall_sock->end_of_message() );;
return 0;
}
case CONDOR_DeleteAttribute:
{
int cluster_id = -1;
int proc_id = -1;
char *attr_name=NULL;
int terrno;
assert( syscall_sock->code(cluster_id) );
dprintf( D_SYSCALLS, " cluster_id = %d\n", cluster_id );
assert( syscall_sock->code(proc_id) );
dprintf( D_SYSCALLS, " proc_id = %d\n", proc_id );
assert( syscall_sock->code(attr_name) );
assert( syscall_sock->end_of_message() );;
errno = 0;
rval = DeleteAttribute( cluster_id, proc_id, attr_name );
terrno = errno;
dprintf( D_SYSCALLS, "\trval = %d, errno = %d\n", rval, terrno );
syscall_sock->encode();
assert( syscall_sock->code(rval) );
if( rval < 0 ) {
assert( syscall_sock->code(terrno) );
}
free( (char *)attr_name );
assert( syscall_sock->end_of_message() );;
return 0;
}
case CONDOR_GetJobAd:
{
int cluster_id = -1;
int proc_id = -1;
ClassAd *ad = NULL;
int terrno;
bool delete_ad = false;
assert( syscall_sock->code(cluster_id) );
dprintf( D_SYSCALLS, " cluster_id = %d\n", cluster_id );
assert( syscall_sock->code(proc_id) );
dprintf( D_SYSCALLS, " proc_id = %d\n", proc_id );
assert( syscall_sock->end_of_message() );;
// dprintf( D_ALWAYS, "(%d.%d) isAuthenticated() = %d\n", cluster_id, proc_id, syscall_sock->isAuthenticated() );
// dprintf( D_ALWAYS, "(%d.%d) getOwner() = %s\n", cluster_id, proc_id, syscall_sock->getOwner() );
errno = 0;
// Only fetch the jobad for legal values of cluster/proc
if( cluster_id >= 1 ) {
if( proc_id >= 0 ) {
const char * fqu = syscall_sock->getFullyQualifiedUser();
if( fqu != NULL && strcmp( fqu, "read-only" ) != 0 ) {
// expand $$() macros in the jobad as required by GridManager.
// The GridManager depends on the fact that the following call
// expands $$ and saves the expansions to disk in case of
// restart.
ad = GetJobAd_as_ClassAd( cluster_id, proc_id, true, true );
delete_ad = true;
// note : since we expanded the ad, ad is now a deep
// copy of the ad in memory, so we must delete it below.
} else {
ad = GetJobAd_as_ClassAd( cluster_id, proc_id, false, false );
}
} else if( proc_id == -1 ) {
// allow cluster ad to be queried as required by preen, but
// do NOT ask to expand $$() macros in a cluster ad!
ad = GetJobAd_as_ClassAd( cluster_id, proc_id, false, false );
}
}
terrno = errno;
rval = ad ? 0 : -1;
dprintf( D_SYSCALLS, "\trval = %d, errno = %d\n", rval, terrno );
syscall_sock->encode();
assert( syscall_sock->code(rval) );
if( rval < 0 ) {
assert( syscall_sock->code(terrno) );
}
if( rval >= 0 ) {
assert( putClassAd(syscall_sock, *ad, PUT_CLASSAD_NO_PRIVATE) );
}
// If we called GetJobAd() with the third bool argument set
// to True (expandedAd), it does a deep copy of the ad in the
// queue in order to expand the $$() attributes. So we must
// delete it.
if (delete_ad) delete ad;
assert( syscall_sock->end_of_message() );;
return 0;
}
case CONDOR_GetJobByConstraint:
{
char *constraint=NULL;
ClassAd *ad;
int terrno;
assert( syscall_sock->code(constraint) );
assert( syscall_sock->end_of_message() );;
errno = 0;
ad = GetJobByConstraint_as_ClassAd( constraint );
terrno = errno;
rval = ad ? 0 : -1;
dprintf( D_SYSCALLS, "\trval = %d, errno = %d\n", rval, terrno );
syscall_sock->encode();
assert( syscall_sock->code(rval) );
if( rval < 0 ) {
assert( syscall_sock->code(terrno) );
}
if( rval >= 0 ) {
assert( putClassAd(syscall_sock, *ad, PUT_CLASSAD_NO_PRIVATE) );
}
FreeJobAd(ad);
free( (char *)constraint );
assert( syscall_sock->end_of_message() );;
return 0;
}
case CONDOR_GetNextJob:
{
ClassAd *ad;
int initScan = 0;
int terrno;
assert( syscall_sock->code(initScan) );
dprintf( D_SYSCALLS, " initScan = %d\n", initScan );
assert( syscall_sock->end_of_message() );;
errno = 0;
ad = GetNextJob( initScan );
terrno = errno;
rval = ad ? 0 : -1;
dprintf( D_SYSCALLS, "\trval = %d, errno = %d\n", rval, terrno );
syscall_sock->encode();
assert( syscall_sock->code(rval) );
if( rval < 0 ) {
assert( syscall_sock->code(terrno) );
}
if( rval >= 0 ) {
assert( putClassAd(syscall_sock, *ad, PUT_CLASSAD_NO_PRIVATE) );
}
FreeJobAd(ad);
assert( syscall_sock->end_of_message() );;
return 0;
}
case CONDOR_GetNextJobByConstraint:
{
char *constraint=NULL;
ClassAd *ad;
int initScan = 0;
int terrno;
assert( syscall_sock->code(initScan) );
dprintf( D_SYSCALLS, " initScan = %d\n", initScan );
if ( !(syscall_sock->code(constraint)) ) {
if (constraint != NULL) {
free(constraint);
constraint = NULL;
}
return -1;
}
assert( syscall_sock->end_of_message() );;
errno = 0;
ad = GetNextJobByConstraint( constraint, initScan );
terrno = errno;
rval = ad ? 0 : -1;
dprintf( D_SYSCALLS, "\trval = %d, errno = %d\n", rval, terrno );
syscall_sock->encode();
assert( syscall_sock->code(rval) );
if( rval < 0 ) {
assert( syscall_sock->code(terrno) );
}
if( rval >= 0 ) {
assert( putClassAd(syscall_sock, *ad, PUT_CLASSAD_NO_PRIVATE) );
}
FreeJobAd(ad);
free( (char *)constraint );
assert( syscall_sock->end_of_message() );;
return 0;
}
case CONDOR_GetNextDirtyJobByConstraint:
{
char *constraint=NULL;
ClassAd *ad;
int initScan = 0;
int terrno;
assert( syscall_sock->code(initScan) );
dprintf( D_SYSCALLS, " initScan = %d\n", initScan );
if ( !(syscall_sock->code(constraint)) ) {
if (constraint != NULL) {
free(constraint);
constraint = NULL;
}
return -1;
}
assert( syscall_sock->end_of_message() );
errno = 0;
ad = GetNextDirtyJobByConstraint( constraint, initScan );
terrno = errno;
rval = ad ? 0 : -1;
dprintf( D_SYSCALLS, "\trval = %d, errno = %d\n", rval, terrno );
syscall_sock->encode();
assert( syscall_sock->code(rval) );
if( rval < 0 ) {
assert( syscall_sock->code(terrno) );
}
if( rval >= 0 ) {
assert( putClassAd(syscall_sock, *ad, PUT_CLASSAD_NO_PRIVATE) );
}
FreeJobAd(ad);
free( (char *)constraint );
assert( syscall_sock->end_of_message() );
return 0;
}
case CONDOR_SendSpoolFile:
{
char *filename=NULL;
int terrno;
assert( syscall_sock->code(filename) );
assert( syscall_sock->end_of_message() );;
errno = 0;
rval = SendSpoolFile( filename );
terrno = errno;
dprintf( D_SYSCALLS, "\trval = %d, errno = %d\n", rval, terrno );
#if 0
syscall_sock->encode();
assert( syscall_sock->code(rval) );
if( rval < 0 ) {
assert( syscall_sock->code(terrno) );
}
assert( syscall_sock->end_of_message() );;
#endif
free( (char *)filename );
return 0;
}
case CONDOR_SendSpoolFileIfNeeded:
{
int terrno;
ClassAd ad;
assert( getClassAd(syscall_sock, ad) );
assert( syscall_sock->end_of_message() );;
errno = 0;
rval = SendSpoolFileIfNeeded(ad);
terrno = errno;
dprintf( D_SYSCALLS, "\trval = %d, errno = %d\n", rval, terrno );
return 0;
}
case CONDOR_GetAllJobsByConstraint:
{
char *constraint=NULL;
char *projection=NULL;
ClassAd *ad;
int terrno;
int initScan = 1;
classad::References proj;
if ( !(syscall_sock->code(constraint)) ) {
if (constraint != NULL) {
free(constraint);
constraint = NULL;
}
return -1;
}
if ( !(syscall_sock->code(projection)) ) {
if (projection != NULL) {
free(constraint);
free(projection);
projection = NULL;
}
return -1;
}
dprintf( D_SYSCALLS, " constraint = %s\n", constraint );
dprintf( D_SYSCALLS, " projection = %s\n", projection ? projection : "");
assert( syscall_sock->end_of_message() );;
// if there is a projection, convert it into a set of attribute names
if (projection) {
StringTokenIterator list(projection);
const std::string * attr;
while ((attr = list.next_string())) { proj.insert(*attr); }
}
syscall_sock->encode();
do {
errno = 0;
ad = GetNextJobByConstraint( constraint, initScan );
initScan=0; // one first time through, otherwise 0
terrno = errno;
rval = ad ? 0 : -1;
dprintf( D_SYSCALLS, "\trval = %d, errno = %d\n", rval, terrno );
assert( syscall_sock->code(rval) );
if( rval < 0 ) {
assert( syscall_sock->code(terrno) );
}
if( rval >= 0 ) {
assert( putClassAd(syscall_sock, *ad, PUT_CLASSAD_NO_PRIVATE, proj.empty() ? NULL : &proj) );
FreeJobAd(ad);
}
} while (rval >= 0);
assert( syscall_sock->end_of_message() );;
free( (char *)constraint );
free( (char *)projection );
return 0;
}
case CONDOR_CloseSocket:
{
assert( syscall_sock->end_of_message() );;
return -1;
}
} /* End of switch */
return -1;
} /* End of function */
ClassAd*
DCSchedd::actOnJobs( JobAction action,
const char* constraint, StringList* ids,
const char* reason, const char* reason_attr,
const char* reason_code, const char* reason_code_attr,
action_result_type_t result_type,
bool notify_scheduler,
CondorError * errstack )
{
char* tmp = NULL;
char buf[512];
int size, reply;
ReliSock rsock;
// // // // // // // //
// Construct the ad we want to send
// // // // // // // //
ClassAd cmd_ad;
sprintf( buf, "%s = %d", ATTR_JOB_ACTION, action );
cmd_ad.Insert( buf );
sprintf( buf, "%s = %d", ATTR_ACTION_RESULT_TYPE,
(int)result_type );
cmd_ad.Insert( buf );
sprintf( buf, "%s = %s", ATTR_NOTIFY_JOB_SCHEDULER,
notify_scheduler ? "True" : "False" );
cmd_ad.Insert( buf );
if( constraint ) {
if( ids ) {
// This is a programming error, not a run-time one
EXCEPT( "DCSchedd::actOnJobs has both constraint and ids!" );
}
size = strlen(constraint) + strlen(ATTR_ACTION_CONSTRAINT) + 4;
tmp = (char*) malloc( size*sizeof(char) );
if( !tmp ) {
EXCEPT( "Out of memory!" );
}
sprintf( tmp, "%s = %s", ATTR_ACTION_CONSTRAINT, constraint );
if( ! cmd_ad.Insert(tmp) ) {
dprintf( D_ALWAYS, "DCSchedd::actOnJobs: "
"Can't insert constraint (%s) into ClassAd!\n",
constraint );
free( tmp );
return NULL;
}
free( tmp );
tmp = NULL;
} else if( ids ) {
char* action_ids = ids->print_to_string();
if ( action_ids ) {
size = strlen(action_ids) + strlen(ATTR_ACTION_IDS) + 7;
tmp = (char*) malloc( size*sizeof(char) );
if( !tmp ) {
EXCEPT( "Out of memory!" );
}
sprintf( tmp, "%s = \"%s\"", ATTR_ACTION_IDS, action_ids );
cmd_ad.Insert( tmp );
free( tmp );
tmp = NULL;
free(action_ids);
action_ids = NULL;
}
} else {
EXCEPT( "DCSchedd::actOnJobs called without constraint or ids" );
}
if( reason_attr && reason ) {
size = strlen(reason_attr) + strlen(reason) + 7;
tmp = (char*) malloc( size*sizeof(char) );
if( !tmp ) {
EXCEPT( "Out of memory!" );
}
sprintf( tmp, "%s = \"%s\"", reason_attr, reason );
cmd_ad.Insert( tmp );
free( tmp );
tmp = NULL;
}
if( reason_code_attr && reason_code ) {
cmd_ad.AssignExpr(reason_code_attr,reason_code);
}
// // // // // // // //
// On the wire protocol
// // // // // // // //
rsock.timeout(20); // years of research... :)
if( ! rsock.connect(_addr) ) {
dprintf( D_ALWAYS, "DCSchedd::actOnJobs: "
"Failed to connect to schedd (%s)\n", _addr );
return NULL;
}
if( ! startCommand(ACT_ON_JOBS, (Sock*)&rsock, 0, errstack) ) {
dprintf( D_ALWAYS, "DCSchedd::actOnJobs: "
"Failed to send command (ACT_ON_JOBS) to the schedd\n" );
return NULL;
}
// First, if we're not already authenticated, force that now.
if (!forceAuthentication( &rsock, errstack )) {
dprintf( D_ALWAYS, "DCSchedd: authentication failure: %s\n",
errstack->getFullText().c_str() );
return NULL;
}
// Now, put the command classad on the wire
if( ! (putClassAd(&rsock, cmd_ad) && rsock.end_of_message()) ) {
dprintf( D_ALWAYS, "DCSchedd:actOnJobs: Can't send classad\n" );
return NULL;
}
// Next, we need to read the reply from the schedd if things
// are ok and it's going to go forward. If the schedd can't
// read our reply to this ClassAd, it assumes we got killed
// and it should abort its transaction
rsock.decode();
ClassAd* result_ad = new ClassAd();
if( ! (getClassAd(&rsock, *result_ad) && rsock.end_of_message()) ) {
dprintf( D_ALWAYS, "DCSchedd:actOnJobs: "
"Can't read response ad from %s\n", _addr );
delete( result_ad );
return NULL;
}
// If the action totally failed, the schedd will already have
// aborted the transaction and closed up shop, so there's no
// reason trying to continue. However, we still want to
// return the result ad we got back so that our caller can
// figure out what went wrong.
reply = FALSE;
result_ad->LookupInteger( ATTR_ACTION_RESULT, reply );
if( reply != OK ) {
dprintf( D_ALWAYS, "DCSchedd:actOnJobs: Action failed\n" );
return result_ad;
}
// Tell the schedd we're still here and ready to go
rsock.encode();
int answer = OK;
if( ! (rsock.code(answer) && rsock.end_of_message()) ) {
dprintf( D_ALWAYS, "DCSchedd:actOnJobs: Can't send reply\n" );
delete( result_ad );
return NULL;
}
// finally, make sure the schedd didn't blow up trying to
// commit these changes to the job queue...
rsock.decode();
if( ! (rsock.code(reply) && rsock.end_of_message()) ) {
dprintf( D_ALWAYS, "DCSchedd:actOnJobs: "
"Can't read confirmation from %s\n", _addr );
delete( result_ad );
return NULL;
}
return result_ad;
}
int
CCBServer::HandleRegistration(int cmd,Stream *stream)
{
ReliSock *sock = (ReliSock *)stream;
ASSERT( cmd == CCB_REGISTER );
// Avoid lengthy blocking on communication with our peer.
// This command-handler should not get called until data
// is ready to read.
sock->timeout(1);
ClassAd msg;
sock->decode();
if( !getClassAd( sock, msg ) || !sock->end_of_message() ) {
dprintf(D_ALWAYS,
"CCB: failed to receive registration "
"from %s.\n", sock->peer_description() );
return FALSE;
}
SetSmallBuffers(sock);
MyString name;
if( msg.LookupString(ATTR_NAME,name) ) {
// target daemon name is purely for debugging purposes
name.formatstr_cat(" on %s",sock->peer_description());
sock->set_peer_description(name.Value());
}
CCBTarget *target = new CCBTarget(sock);
MyString reconnect_cookie_str,reconnect_ccbid_str;
CCBID reconnect_cookie,reconnect_ccbid;
bool reconnected = false;
if( msg.LookupString(ATTR_CLAIM_ID,reconnect_cookie_str) &&
CCBIDFromString(reconnect_cookie,reconnect_cookie_str.Value()) &&
msg.LookupString( ATTR_CCBID,reconnect_ccbid_str) &&
CCBIDFromContactString(reconnect_ccbid,reconnect_ccbid_str.Value()) )
{
target->setCCBID( reconnect_ccbid );
reconnected = ReconnectTarget( target, reconnect_cookie );
}
if( !reconnected ) {
AddTarget( target );
}
CCBReconnectInfo *reconnect_info = GetReconnectInfo( target->getCCBID() );
ASSERT( reconnect_info );
sock->encode();
ClassAd reply_msg;
MyString ccb_contact;
CCBIDToString( reconnect_info->getReconnectCookie(),reconnect_cookie_str );
// We send our address as part of the CCB contact string, rather
// than letting the target daemon fill it in. This is to give us
// potential flexibility on the CCB server side to do things like
// assign different targets to different CCB server sub-processes,
// each with their own command port.
CCBIDToContactString( m_address.Value(), target->getCCBID(), ccb_contact );
reply_msg.Assign(ATTR_CCBID,ccb_contact.Value());
reply_msg.Assign(ATTR_COMMAND,CCB_REGISTER);
reply_msg.Assign(ATTR_CLAIM_ID,reconnect_cookie_str.Value());
if( !putClassAd( sock, reply_msg ) || !sock->end_of_message() ) {
dprintf(D_ALWAYS,
"CCB: failed to send registration response "
"to %s.\n", sock->peer_description() );
RemoveTarget( target );
return KEEP_STREAM; // we have already closed this socket
}
return KEEP_STREAM;
}
bool
DCStarter::peek(bool transfer_stdout, ssize_t &stdout_offset, bool transfer_stderr, ssize_t &stderr_offset, const std::vector<std::string> &filenames, std::vector<ssize_t> &offsets, size_t max_bytes, bool &retry_sensible, PeekGetFD &next, std::string &error_msg, unsigned timeout, const std::string &sec_session_id, DCTransferQueue *xfer_q)
{
compat_classad::ClassAd ad;
ad.InsertAttr(ATTR_JOB_OUTPUT, transfer_stdout);
ad.InsertAttr("OutOffset", stdout_offset);
ad.InsertAttr(ATTR_JOB_ERROR, transfer_stderr);
ad.InsertAttr("ErrOffset", stderr_offset);
ad.InsertAttr(ATTR_VERSION, CondorVersion());
size_t total_files = 0;
total_files += transfer_stdout ? 1 : 0;
total_files += transfer_stderr ? 1 : 0;
if (filenames.size())
{
total_files += filenames.size();
std::vector<classad::ExprTree *> filelist; filelist.reserve(filenames.size());
std::vector<classad::ExprTree *> offsetlist; offsetlist.reserve(filenames.size());
std::vector<ssize_t>::const_iterator it2 = offsets.begin();
for (std::vector<std::string>::const_iterator it = filenames.begin();
it != filenames.end() && it2 != offsets.end();
it++, it2++)
{
classad::Value value;
value.SetStringValue(*it);
filelist.push_back(classad::Literal::MakeLiteral(value));
value.SetIntegerValue(*it2);
offsetlist.push_back(classad::Literal::MakeLiteral(value));
}
classad::ExprTree *list(classad::ExprList::MakeExprList(filelist));
ad.Insert("TransferFiles", list);
list = classad::ExprList::MakeExprList(offsetlist);
ad.Insert("TransferOffsets", list);
}
ad.InsertAttr(ATTR_MAX_TRANSFER_BYTES, static_cast<long long>(max_bytes));
ReliSock sock;
if( !connectSock(&sock, timeout, NULL) ) {
error_msg = "Failed to connect to starter";
return false;
}
if( !startCommand(STARTER_PEEK, &sock, timeout, NULL, NULL, false, sec_session_id.c_str()) ) {
error_msg = "Failed to send START_PEEK to starter";
return false;
}
sock.encode();
if (!putClassAd(&sock, ad) || !sock.end_of_message()) {
error_msg = "Failed to send request to starter";
return false;
}
compat_classad::ClassAd response;
sock.decode();
if (!getClassAd(&sock, response) || !sock.end_of_message())
{
error_msg = "Failed to read response for peeking at logs.";
return false;
}
dPrintAd(D_FULLDEBUG, response);
bool success = false;
if (!response.EvaluateAttrBool(ATTR_RESULT, success) || !success)
{
response.EvaluateAttrBool(ATTR_RETRY, retry_sensible);
error_msg = "Remote operation failed.";
response.EvaluateAttrString(ATTR_ERROR_STRING, error_msg);
return false;
}
classad::Value valueX;
classad_shared_ptr<classad::ExprList> list;
if (!response.EvaluateAttr("TransferFiles", valueX) || !valueX.IsSListValue(list))
{
error_msg = "Unable to evaluate starter response";
return false;
}
classad_shared_ptr<classad::ExprList> offlist;
if (!response.EvaluateAttr("TransferOffsets", valueX) || !valueX.IsSListValue(offlist))
{
error_msg = "Unable to evaluate starter response (missing offsets)";
return false;
}
size_t remaining = max_bytes;
size_t file_count = 0;
classad::ExprList::const_iterator it2 = offlist->begin();
for (classad::ExprList::const_iterator it = list->begin();
it != list->end() && it2 != offlist->end();
it++, it2++)
{
classad::Value value;
(*it2)->Evaluate(value);
off_t off = -1;
value.IsIntegerValue(off);
(*it)->Evaluate(value);
std::string filename;
int64_t xfer_fd = -1;
if (!value.IsStringValue(filename) && value.IsIntegerValue(xfer_fd))
{
if (xfer_fd == 0) filename = "_condor_stdout";
if (xfer_fd == 1) filename = "_condor_stderr";
}
int fd = next.getNextFD(filename);
filesize_t size = -1;
int retval;
if ((retval = sock.get_file(&size, fd, false, false, remaining, xfer_q)) && (retval != GET_FILE_MAX_BYTES_EXCEEDED))
{
error_msg = "Internal error when transferring file " + filename;
}
else if (size >= 0)
{
remaining -= max_bytes;
file_count++;
off += size;
}
else
{
error_msg = "Failed to transfer file " + filename;
}
if (xfer_fd == 0)
{
stdout_offset = off;
//dprintf(D_FULLDEBUG, "New stdout offset: %ld\n", stdout_offset);
}
else if (xfer_fd == 1)
{
stderr_offset = off;
}
else
{
std::vector<ssize_t>::iterator it4 = offsets.begin();
for (std::vector<std::string>::const_iterator it3 = filenames.begin();
it3 != filenames.end() && it4 != offsets.end();
it3++, it4++)
{
if (*it3 == filename) *it4 = off;
}
}
}
size_t remote_file_count;
if (!sock.get(remote_file_count) || !sock.end_of_message())
{
error_msg = "Unable to get remote file count.";
return false;
}
if (file_count != remote_file_count)
{
std::stringstream ss;
ss << "Recieved " << file_count << " files, but remote side thought it sent " << remote_file_count << " files";
error_msg = ss.str();
return false;
}
if ((total_files != file_count) && !error_msg.size())
{
error_msg = "At least one file transfer failed.";
return false;
}
return true;
}
bool DCStarter::startSSHD(char const *known_hosts_file,char const *private_client_key_file,char const *preferred_shells,char const *slot_name,char const *ssh_keygen_args,ReliSock &sock,int timeout,char const *sec_session_id,MyString &remote_user,MyString &error_msg,bool &retry_is_sensible)
{
retry_is_sensible = false;
#ifndef HAVE_SSH_TO_JOB
error_msg = "This version of Condor does not support ssh key exchange.";
return false;
#else
if( !connectSock(&sock, timeout, NULL) ) {
error_msg = "Failed to connect to starter";
return false;
}
if( !startCommand(START_SSHD, &sock,timeout,NULL,NULL,false,sec_session_id) ) {
error_msg = "Failed to send START_SSHD to starter";
return false;
}
ClassAd input;
if( preferred_shells && *preferred_shells ) {
input.Assign(ATTR_SHELL,preferred_shells);
}
if( slot_name && *slot_name ) {
// This is a little silly.
// We are telling the remote side the name of the slot so
// that it can put it in the welcome message.
input.Assign(ATTR_NAME,slot_name);
}
if( ssh_keygen_args && *ssh_keygen_args ) {
input.Assign(ATTR_SSH_KEYGEN_ARGS,ssh_keygen_args);
}
sock.encode();
if( !putClassAd(&sock, input) || !sock.end_of_message() ) {
error_msg = "Failed to send START_SSHD request to starter";
return false;
}
ClassAd result;
sock.decode();
if( !getClassAd(&sock, result) || !sock.end_of_message() ) {
error_msg = "Failed to read response to START_SSHD from starter";
return false;
}
bool success = false;
result.LookupBool(ATTR_RESULT,success);
if( !success ) {
std::string remote_error_msg;
result.LookupString(ATTR_ERROR_STRING,remote_error_msg);
error_msg.formatstr("%s: %s",slot_name,remote_error_msg.c_str());
retry_is_sensible = false;
result.LookupBool(ATTR_RETRY,retry_is_sensible);
return false;
}
result.LookupString(ATTR_REMOTE_USER,remote_user);
std::string public_server_key;
if( !result.LookupString(ATTR_SSH_PUBLIC_SERVER_KEY,public_server_key) ) {
error_msg = "No public ssh server key received in reply to START_SSHD";
return false;
}
std::string private_client_key;
if( !result.LookupString(ATTR_SSH_PRIVATE_CLIENT_KEY,private_client_key) ) {
error_msg = "No ssh client key received in reply to START_SSHD";
return false;
}
// store the private client key
unsigned char *decode_buf = NULL;
int length = -1;
condor_base64_decode(private_client_key.c_str(),&decode_buf,&length);
if( !decode_buf ) {
error_msg = "Error decoding ssh client key.";
return false;
}
FILE *fp = safe_fcreate_fail_if_exists(private_client_key_file,"a",0400);
if( !fp ) {
error_msg.formatstr("Failed to create %s: %s",
private_client_key_file,strerror(errno));
free( decode_buf );
return false;
}
if( fwrite(decode_buf,length,1,fp)!=1 ) {
error_msg.formatstr("Failed to write to %s: %s",
private_client_key_file,strerror(errno));
fclose( fp );
free( decode_buf );
return false;
}
if( fclose(fp)!=0 ) {
error_msg.formatstr("Failed to close %s: %s",
private_client_key_file,strerror(errno));
free( decode_buf );
return false;
}
fp = NULL;
free( decode_buf );
decode_buf = NULL;
// store the public server key in the known_hosts file
length = -1;
condor_base64_decode(public_server_key.c_str(),&decode_buf,&length);
if( !decode_buf ) {
error_msg = "Error decoding ssh server key.";
return false;
}
fp = safe_fcreate_fail_if_exists(known_hosts_file,"a",0600);
if( !fp ) {
error_msg.formatstr("Failed to create %s: %s",
known_hosts_file,strerror(errno));
free( decode_buf );
return false;
}
// prepend a host name pattern (*) to the public key to make a valid
// record in the known_hosts file
fprintf(fp,"* ");
if( fwrite(decode_buf,length,1,fp)!=1 ) {
error_msg.formatstr("Failed to write to %s: %s",
known_hosts_file,strerror(errno));
fclose( fp );
free( decode_buf );
return false;
}
if( fclose(fp)!=0 ) {
error_msg.formatstr("Failed to close %s: %s",
known_hosts_file,strerror(errno));
free( decode_buf );
return false;
}
fp = NULL;
free( decode_buf );
decode_buf = NULL;
return true;
#endif
}
bool DCSchedd::getJobConnectInfo(
PROC_ID jobid,
int subproc,
char const *session_info,
int timeout,
CondorError *errstack,
MyString &starter_addr,
MyString &starter_claim_id,
MyString &starter_version,
MyString &slot_name,
MyString &error_msg,
bool &retry_is_sensible,
int &job_status,
MyString &hold_reason)
{
ClassAd input;
ClassAd output;
input.Assign(ATTR_CLUSTER_ID,jobid.cluster);
input.Assign(ATTR_PROC_ID,jobid.proc);
if( subproc != -1 ) {
input.Assign(ATTR_SUB_PROC_ID,subproc);
}
input.Assign(ATTR_SESSION_INFO,session_info);
ReliSock sock;
if( !connectSock(&sock,timeout,errstack) ) {
error_msg = "Failed to connect to schedd";
dprintf( D_ALWAYS, "%s\n",error_msg.Value());
return false;
}
if( !startCommand(GET_JOB_CONNECT_INFO, &sock, timeout, errstack) ) {
error_msg = "Failed to send GET_JOB_CONNECT_INFO to schedd";
dprintf( D_ALWAYS, "%s\n",error_msg.Value());
return false;
}
if( !forceAuthentication(&sock, errstack) ) {
error_msg = "Failed to authenticate";
dprintf( D_ALWAYS, "%s\n",error_msg.Value());
return false;
}
sock.encode();
if( !putClassAd(&sock, input) || !sock.end_of_message() ) {
error_msg = "Failed to send GET_JOB_CONNECT_INFO to schedd";
dprintf( D_ALWAYS, "%s\n",error_msg.Value());
return false;
}
sock.decode();
if( !getClassAd(&sock, output) || !sock.end_of_message() ) {
error_msg = "Failed to get response from schedd";
dprintf( D_ALWAYS, "%s\n",error_msg.Value());
return false;
}
if( IsFulldebug(D_FULLDEBUG) ) {
std::string adstr;
sPrintAd(adstr, output, true);
dprintf(D_FULLDEBUG,"Response for GET_JOB_CONNECT_INFO:\n%s\n",
adstr.c_str());
}
bool result=false;
output.LookupBool(ATTR_RESULT,result);
if( !result ) {
output.LookupString(ATTR_HOLD_REASON,hold_reason);
output.LookupString(ATTR_ERROR_STRING,error_msg);
retry_is_sensible = false;
output.LookupBool(ATTR_RETRY,retry_is_sensible);
output.LookupInteger(ATTR_JOB_STATUS,job_status);
}
else {
output.LookupString(ATTR_STARTER_IP_ADDR,starter_addr);
output.LookupString(ATTR_CLAIM_ID,starter_claim_id);
output.LookupString(ATTR_VERSION,starter_version);
output.LookupString(ATTR_REMOTE_HOST,slot_name);
}
return result;
}
bool
ScheddNegotiate::sendJobInfo(Sock *sock, bool just_sig_attrs)
{
// The Negotiator wants us to send it a job.
sock->encode();
if( m_current_job_id.cluster == -1 && !nextJob() ) {
if( !sock->snd_int(NO_MORE_JOBS,TRUE) ) {
dprintf( D_ALWAYS, "Can't send NO_MORE_JOBS to mgr\n" );
return false;
}
m_negotiation_finished = true;
return true;
}
if( !sock->put(JOB_INFO) ) {
dprintf( D_ALWAYS, "Can't send JOB_INFO to mgr\n" );
return false;
}
// If schedd wants pslot preemption, advertise here
m_current_job_ad.Assign(ATTR_WANT_PSLOT_PREEMPTION,
param_boolean("ALLOW_PSLOT_PREEMPTION", false));
// request match diagnostics
// 0 = no match diagnostics
// 1 = match diagnostics string
// 2 = match diagnostics string decorated w/ autocluster + jobid
m_current_job_ad.Assign(ATTR_WANT_MATCH_DIAGNOSTICS, (int) 2);
m_current_job_ad.Assign(ATTR_WANT_PSLOT_PREEMPTION,
param_boolean("ALLOW_PSLOT_PREEMPTION", false));
// Send the ad to the negotiator
int putad_result = 0;
std::string auto_cluster_attrs;
if ( just_sig_attrs &&
m_current_job_ad.LookupString(ATTR_AUTO_CLUSTER_ATTRS, auto_cluster_attrs) )
{
// don't send the entire job ad; just send significant attrs
classad::References sig_attrs;
StringTokenIterator list(auto_cluster_attrs);
const std::string *attr;
while ((attr = list.next_string())) { sig_attrs.insert(*attr); }
// besides significant attrs, we also always want to send these attrs cuz
// the matchmaker explicitly looks for them (for dprintfs or whatever).
sig_attrs.insert(ATTR_OWNER);
sig_attrs.insert(ATTR_CLUSTER_ID);
sig_attrs.insert(ATTR_PROC_ID);
sig_attrs.insert(ATTR_RESOURCE_REQUEST_COUNT);
sig_attrs.insert(ATTR_GLOBAL_JOB_ID);
sig_attrs.insert(ATTR_AUTO_CLUSTER_ID);
sig_attrs.insert(ATTR_WANT_MATCH_DIAGNOSTICS);
sig_attrs.insert(ATTR_WANT_PSLOT_PREEMPTION);
sig_attrs.insert(ATTR_WANT_CLAIMING); // used for Condor-G matchmaking
// ship it!
putad_result = putClassAd(sock, m_current_job_ad, 0, &sig_attrs);
} else {
// send the entire classad. perhaps we are doing this because the
// ad does not have ATTR_AUTO_CLUSTER_ATTRS defined for some reason,
// or perhaps we are doing this because we were explicitly told to do so.
putad_result = putClassAd(sock, m_current_job_ad);
}
if( !putad_result ) {
dprintf( D_ALWAYS,
"Can't send job ad to mgr\n" );
sock->end_of_message();
return false;
}
if( !sock->end_of_message() ) {
dprintf( D_ALWAYS,
"Can't send job eom to mgr\n" );
return false;
}
m_current_resources_delivered = 0;
m_current_resources_requested = 1;
m_current_job_ad.LookupInteger(ATTR_RESOURCE_REQUEST_COUNT,m_current_resources_requested);
dprintf( D_FULLDEBUG,
"Sent job %d.%d (autocluster=%d resources_requested=%d) to the negotiator\n",
m_current_job_id.cluster, m_current_job_id.proc,
m_current_auto_cluster_id, m_current_resources_requested );
return true;
}
// when a transferd registers itself, it identifies who it is. The connection
// is then held open and the schedd periodically might send more transfer
// requests to the transferd. Also, if the transferd dies, the schedd is
// informed quickly and reliably due to the closed connection.
bool
DCSchedd::register_transferd(MyString sinful, MyString id, int timeout,
ReliSock **regsock_ptr, CondorError *errstack)
{
ReliSock *rsock;
int invalid_request = 0;
ClassAd regad;
ClassAd respad;
std::string errstr;
std::string reason;
if (regsock_ptr != NULL) {
// Our caller wants a pointer to the socket we used to succesfully
// register the claim. The NULL pointer will represent failure and
// this will only be set to something real if everything was ok.
*regsock_ptr = NULL;
}
// This call with automatically connect to _addr, which was set in the
// constructor of this object to be the schedd in question.
rsock = (ReliSock*)startCommand(TRANSFERD_REGISTER, Stream::reli_sock,
timeout, errstack);
if( ! rsock ) {
dprintf( D_ALWAYS, "DCSchedd::register_transferd: "
"Failed to send command (TRANSFERD_REGISTER) "
"to the schedd\n" );
errstack->push("DC_SCHEDD", 1,
"Failed to start a TRANSFERD_REGISTER command.");
return false;
}
// First, if we're not already authenticated, force that now.
if (!forceAuthentication( rsock, errstack )) {
dprintf( D_ALWAYS, "DCSchedd::register_transferd authentication "
"failure: %s\n", errstack->getFullText().c_str() );
errstack->push("DC_SCHEDD", 1,
"Failed to authenticate properly.");
return false;
}
rsock->encode();
// set up my registration request.
regad.Assign(ATTR_TREQ_TD_SINFUL, sinful);
regad.Assign(ATTR_TREQ_TD_ID, id);
// This is the initial registration identification ad to the schedd
// It contains:
// ATTR_TREQ_TD_SINFUL
// ATTR_TREQ_TD_ID
putClassAd(rsock, regad);
rsock->end_of_message();
// Get the response from the schedd.
rsock->decode();
// This is the response ad from the schedd:
// It contains:
// ATTR_TREQ_INVALID_REQUEST
//
// OR
//
// ATTR_TREQ_INVALID_REQUEST
// ATTR_TREQ_INVALID_REASON
getClassAd(rsock, respad);
rsock->end_of_message();
respad.LookupInteger(ATTR_TREQ_INVALID_REQUEST, invalid_request);
if (invalid_request == FALSE) {
// not an invalid request
if (regsock_ptr)
*regsock_ptr = rsock;
return true;
}
respad.LookupString(ATTR_TREQ_INVALID_REASON, reason);
errstack->pushf("DC_SCHEDD", 1, "Schedd refused registration: %s", reason.c_str());
return false;
}