Gittrac #5480: Initial version of submitting NOOP node jobs works

(it works if there's a real submit file for the node, but not
otherwise).  Commit includes temporary code.

src/condor_dagman/dag.cpp

@@ -91,8 +91,8 @@ Dag::Dag( /* const */ StringList &dagFiles,
 		  const CondorID *DAGManJobID,
 		  bool prohibitMultiJobs, bool submitDepthFirst,
 		  const char *defaultNodeLog, bool generateSubdagSubmits,
-		  SubmitDagDeepOptions *submitDagDeepOpts, bool isSplice,
-		  const MyString &spliceScope ) :
+		  SubmitDagDeepOptions *submitDagDeepOpts, bool submitNoopJobs,
+		  bool isSplice, const MyString &spliceScope ) :
     _maxPreScripts        (maxPreScripts),
     _maxPostScripts       (maxPostScripts),
 	MAX_SIGNAL			  (64),
@@ -134,6 +134,7 @@ Dag::Dag( /* const */ StringList &dagFiles,
 	_reject			  (false),
 	_alwaysRunPost		  (true),
 	_defaultPriority	  (0),
+	_submitNoopJobs		  (submitNoopJobs),
 	_metrics			  (NULL)
 {
 
@@ -1337,7 +1338,7 @@ Job * Dag::FindNodeByEventID ( const CondorID condorID ) const {
 				check_warning_strictness( DAG_STRICT_3 );
 			}
 		}
-		ASSERT( isNoop == node->GetNoop() );
+		//TEMPTEMP? ASSERT( isNoop == node->GetNoop() );
 	}
 
 	return node;
@@ -3673,7 +3674,7 @@ Dag::LogEventNodeLookup( const ULogEvent* event,
 						// inserted it when we did the condor_submit.)
 					Job *tmpNode = NULL;
 					bool isNoop = JobIsNoop( condorID );
-					ASSERT( isNoop == node->GetNoop() );
+					//TEMPTEMP ASSERT( isNoop == node->GetNoop() );
 					int id = GetIndexID( condorID );
 					HashTable<int, Job *> *ht =
 								GetEventIDHash( isNoop );
@@ -3859,9 +3860,9 @@ Dag::SanityCheckSubmitEvent( const CondorID condorID, const Job* node )
 
 //---------------------------------------------------------------------------
 HashTable<int, Job *> *
-Dag::GetEventIDHash(bool isNoop)
+Dag::GetEventIDHash( bool isNoop )
 {
-	if ( isNoop ) {
+	if ( isNoop && !_submitNoopJobs ) {
 		return &_noopIDHash;
 	}
 
@@ -3961,7 +3962,8 @@ Dag::SubmitNodeJob( const Dagman &dm, Job *node, CondorID &condorID )
 	bool submit_success = false;
 
  	node->_submitTries++;
-	if ( node->GetNoop() ) {
+	//TEMPTEMP? if ( node->GetNoop() ) {
+	if ( node->GetNoop() && !_submitNoopJobs ) {//TEMPTEMP?
    		submit_success = fake_condor_submit( condorID, 0,
 					node->GetJobName(), node->GetDirectory(),
 					_defaultNodeLog );
@@ -3975,7 +3977,8 @@ Dag::SubmitNodeJob( const Dagman &dm, Job *node, CondorID &condorID )
 					node->GetJobName(), parents,
 					node->varsFromDag, node->GetRetries(),
 					node->GetDirectory(), _defaultNodeLog,
-					node->NumChildren() > 0 && dm._claim_hold_time > 0 );
+					node->NumChildren() > 0 && dm._claim_hold_time > 0,
+					node->GetNoop() );
 	}
 
 	result = submit_success ? SUBMIT_RESULT_OK : SUBMIT_RESULT_FAILED;
@@ -4012,7 +4015,7 @@ Dag::ProcessSuccessfulSubmit( Job *node, const CondorID &condorID )
         // since we won't have seen the submit command stdout...)
 
 	node->SetCondorID( condorID );
-	ASSERT( JobIsNoop( node->GetID() ) == node->GetNoop() );
+	//TEMPTEMP ASSERT( JobIsNoop( node->GetID() ) == node->GetNoop() );
 	int id = GetIndexID( node->GetID() );
 	int insertResult = GetEventIDHash( node->GetNoop() )->insert( id, node );
 	ASSERT( insertResult == 0 );

src/condor_dagman/dag.h

@@ -135,6 +135,7 @@ class Dag {
 		 bool prohibitMultiJobs, bool submitDepthFirst,
 		 const char *defaultNodeLog, bool generateSubdagSubmits,
 		 SubmitDagDeepOptions *submitDagDeepOpts,
+		 bool submitNoopJobs,
 		 bool isSplice = false, const MyString &spliceScope = "root" );
 
     ///
@@ -1193,6 +1194,10 @@ class Dag {
 		// The default priority for nodes in this DAG. (defaults to 0)
 	int _defaultPriority;
 
+		//TEMPTEMP -- document
+		//TEMPTEMP -- hmm -- should this be in submitDagDeepOpts or something?
+	bool _submitNoopJobs;
+
 		// Whether the DAG is currently halted.
 	bool _dagIsHalted;
 

src/condor_dagman/dagman_main.cpp

@@ -142,6 +142,7 @@ Dagman::Dagman() :
 	_claim_hold_time(20),
 	_doRecovery(false),
 	_suppressJobLogs(false),
+	_submitNoopJobs(true),
 	_dagmanClassad(NULL)
 {
     debug_level = DEBUG_VERBOSE;  // Default debug level is verbose output
@@ -451,6 +452,12 @@ Dagman::Config()
 	debug_printf( DEBUG_NORMAL, "DAGMAN_SUPPRESS_JOB_LOGS setting: %s\n",
 				_suppressJobLogs ? "True" : "False" );
 
+	_submitNoopJobs = 
+				param_boolean( "DAGMAN_SUBMIT_NOOP_JOBS",
+				_submitNoopJobs );
+	debug_printf( DEBUG_NORMAL, "DAGMAN_SUBMIT_NOOP_JOBS setting: %s\n",
+				_submitNoopJobs ? "True" : "False" );
+
 	// enable up the debug cache if needed
 	if (debug_cache_enabled) {
 		debug_cache_set_size(debug_cache_size);
@@ -1093,6 +1100,7 @@ void main_init (int argc, char ** const argv) {
 						  dagman._defaultNodeLog.Value(),
 						  dagman._generateSubdagSubmits,
 						  &dagman._submitDagDeepOpts,
+						  dagman._submitNoopJobs,
 						  false ); /* toplevel dag! */
 
     if( dagman.dag == NULL ) {

src/condor_dagman/dagman_main.h

@@ -226,6 +226,9 @@ class Dagman {
 		// log files specified in their submit files (see gittrac #4353).
 	bool _suppressJobLogs;
 
+		//TEMPTEMP -- document
+	bool _submitNoopJobs;
+
 	DagmanClassad *_dagmanClassad;
 };
 

src/condor_dagman/dagman_submit.cpp

@@ -201,12 +201,14 @@ do_submit( ArgList &args, CondorID &condorID, bool prohibitMultiJobs )
 }
 
 //-------------------------------------------------------------------------
+//TEMPTEMP -- probably start by just appending "noop_job = true" here if job is a noop; then add stuff to cons up the whole submit file...
+//TEMPTEMP -- for a noop job, do we want to skip a bunch of the stuff like appending the parent node list?  or are we better off not creating two different code paths so much?
 bool
 condor_submit( const Dagman &dm, const char* cmdFile, CondorID& condorID,
 			   const char* DAGNodeName, MyString &DAGParentNodeNames,
 			   List<Job::NodeVar> *vars, int retry,
 			   const char* directory, const char *workflowLogFile,
-			   bool hold_claim )
+			   bool hold_claim, bool noopNode )
 {
 	TmpDir		tmpDir;
 	MyString	errMsg;
@@ -234,6 +236,11 @@ condor_submit( const Dagman &dm, const char* cmdFile, CondorID& condorID,
 
 	args.AppendArg( dm.condorSubmitExe );
 
+	if ( noopNode ) {
+		args.AppendArg( "-a" );
+		args.AppendArg( "noop_job=true" );
+	}
+
 	args.AppendArg( "-a" );
 	MyString nodeName = MyString(ATTR_DAG_NODE_NAME_ALT) + " = " + DAGNodeName;
 	args.AppendArg( nodeName.Value() );
@@ -392,6 +399,8 @@ get_fake_condorID()
 {
 	return _subprocID;
 }
+
+//TEMPTEMP -- hmm -- we probably don't want to call this at all if we are submitting a noop node; do we want to call the "real" submit method with a noop flag, or make a new method for submitting noop nodes?
 //-------------------------------------------------------------------------
 bool
 fake_condor_submit( CondorID& condorID, Job* job, const char* DAGNodeName, 

src/condor_dagman/parse.cpp

@@ -1728,6 +1728,7 @@ parse_splice(
 							dag->DefaultNodeLog(),
 							dag->GenerateSubdagSubmits(),
 							NULL, // this Dag will never submit a job
+							false, //TEMPTEMP -- submit noops
 							true, /* we are a splice! */
 							current_splice_scope() );
 

src/condor_dagman/submit.h

@@ -47,14 +47,15 @@
 	@param log file to force this job to use (should be null if submit
 		file specifies log file)
 	@param hold_claim is true if DAGMAN_HOLD_CLAIM_IDLE is positive
+	@param noopNode TEMPTEMP
 	@return true on success, false on failure
 */
 
 bool condor_submit( const Dagman &dm, const char* cmdFile, CondorID& condorID,
 					const char* DAGNodeName, MyString &DAGParentNodeNames,
 					List<Job::NodeVar> *vars, int retry,
 					const char* directory, const char *worflowLogFile,
-					bool hold_claim );
+					bool hold_claim, bool noopNode );
 
 void set_fake_condorID( int subprocID );
 

src/condor_tests/job_dagman_basic.dag

@@ -1,3 +1,4 @@
-Job     NODERET         job_dagman_basic-node.cmd
+#TEMPTEMP Job     NODERET         job_dagman_basic-node.cmd
+Job     NODERET         job_dagman_basic-node.cmd NOOP
 Script PRE NODERET     ./x_dagman_premonitor.pl 
 Script POST NODERET     ./job_dagman_monitor.pl $RETURN

src/condor_utils/param_info.in

@@ -1193,6 +1193,15 @@ friendly_name=Dagman Write Partial Rescue
 review=?
 tags=dagman,dagman_main
 
+[DAGMAN_SUBMIT_NOOP_JOBS]
+default=true
+type=bool
+reconfig=false
+customization=expert
+friendly_name=Dagman Write Partial Rescue
+review=?
+tags=dagman,dagman_main
+
 [CKPT_SERVER_DIR]
 default=
 type=string