Merge pull request #1350 from virtualcell/sim-control-test-2

Sim control test 2
virtualcell · Oct 11, 2024 · 2b091cf · 2b091cf
2 parents 09fcba6 + d3744b8
commit 2b091cf
Show file tree

Hide file tree

Showing 35 changed files with 2,251 additions and 1,144 deletions.
diff --git a/docker/build/Dockerfile-sched-dev b/docker/build/Dockerfile-sched-dev
@@ -68,7 +68,11 @@ ENV softwareVersion=SOFTWARE-VERSION-NOT-SET \
 	maxOdeJobsPerUser="max-ode-jobs-per-user-not-set" \
 	vcell_ssh_cmd_cmdtimeout="cmdSrvcSshCmdTimeoutMS-not-set" \
     vcell_ssh_cmd_restoretimeout="cmdSrvcSshCmdRestoreTimeoutFactor-not-set" \
-	maxPdeJobsPerUser="max-pde-jobs-per-user-not-set"
+	maxPdeJobsPerUser="max-pde-jobs-per-user-not-set" \
+	htcMinMemoryMB="htc-min-memory-not-set" \
+	htcMaxMemoryMB="htc-max-memory-not-set" \
+	htcPowerUserMemoryFloorMB="htc-power-user-memory-floor-not-set" \
+    htcPowerUserMemoryMaxMB="htc-power-user-memory-max-not-set"
 
 ENV dbpswdfile=/run/secrets/dbpswd \
     jmspswdfile=/run/secrets/jmspswd \
@@ -119,4 +123,8 @@ ENTRYPOINT java \
 	-Dvcell.server.maxPdeJobsPerUser=${maxPdeJobsPerUser} \
 	-Dvcell.ssh.cmd.cmdtimeout=${vcell_ssh_cmd_cmdtimeout} \
 	-Dvcell.ssh.cmd.restoretimeout=${vcell_ssh_cmd_restoretimeout} \
-	-cp "./lib/*" cbit.vcell.message.server.dispatcher.SimulationDispatcher
+	-Dvcell.htc.memory.min.mb=${htcMinMemoryMB} \
+	-Dvcell.htc.memory.max.mb=${htcMaxMemoryMB} \
+	-Dvcell.htc.memory.pu.floor.mb=${htcPowerUserMemoryFloorMB} \
+    -Dvcell.htc.memory.pu.max.mb=${htcPowerUserMemoryMaxMB} \
+	-cp "./lib/*" cbit.vcell.message.server.dispatcher.SimulationDispatcherMain
diff --git a/docker/build/Dockerfile-submit-dev b/docker/build/Dockerfile-submit-dev
@@ -91,7 +91,11 @@ ENV softwareVersion=SOFTWARE-VERSION-NOT-SET \
     vcell_ssh_cmd_cmdtimeout="cmdSrvcSshCmdTimeoutMS-not-set" \
     vcell_ssh_cmd_restoretimeout="cmdSrvcSshCmdRestoreTimeoutFactor-not-set" \
     simdatadir_archive_external="simdatadir_archive_external-not-set" \
-    simdatadir_archive_internal="simdatadir_archive_internal-not-set"
+    simdatadir_archive_internal="simdatadir_archive_internal-not-set" \
+    htcMinMemoryMB="htc-min-memory-not-set" \
+    htcMaxMemoryMB="htc-max-memory-not-set" \
+    htcPowerUserMemoryFloorMB="htc-power-user-memory-floor-not-set" \
+    htcPowerUserMemoryMaxMB="htc-power-user-memory-max-not-set"
 
 ENV jmspswdfile=/run/secrets/jmspswd \
 	jmsrestpswdfile=/run/secrets/jmsrestpswd \
@@ -170,4 +174,8 @@ ENTRYPOINT java \
 	-Dvcell.simdatadir.archive.external=${simdatadir_archive_external} \
 	-Dvcell.ssh.cmd.cmdtimeout=${vcell_ssh_cmd_cmdtimeout} \
 	-Dvcell.ssh.cmd.restoretimeout=${vcell_ssh_cmd_restoretimeout} \
+	-Dvcell.htc.memory.min.mb=${htcMinMemoryMB} \
+    -Dvcell.htc.memory.max.mb=${htcMaxMemoryMB} \
+    -Dvcell.htc.memory.pu.floor.mb=${htcPowerUserMemoryFloorMB} \
+    -Dvcell.htc.memory.pu.max.mb=${htcPowerUserMemoryMaxMB} \
 	-cp "./lib/*" cbit.vcell.message.server.batch.sim.HtcSimulationWorker
diff --git a/docker/build/build.sh b/docker/build/build.sh
@@ -130,6 +130,8 @@ build_webapp() {
   if [[ $? -ne 0 ]]; then echo "failed to build prod"; exit 1; fi
   build_webapp_common island
   if [[ $? -ne 0 ]]; then echo "failed to build island"; exit 1; fi
+  build_webapp_common remote
+  if [[ $? -ne 0 ]]; then echo "failed to build remote"; exit 1; fi
 }
 
 build_batch() {

diff --git a/vcell-core/src/main/java/cbit/rmi/event/WorkerEvent.java b/vcell-core/src/main/java/cbit/rmi/event/WorkerEvent.java
@@ -19,6 +19,8 @@
 import cbit.vcell.solver.VCSimulationIdentifier;
 import cbit.vcell.solver.server.SimulationMessage;
 
+import java.util.ArrayList;
+
 /**
  * Insert the type's description here.
  * Creation date: (2/5/2004 12:35:20 PM)
@@ -36,6 +38,10 @@ public class WorkerEvent extends MessageEvent {
 	public static final int JOB_WORKER_EXIT_NORMAL = 1015;
 	public static final int JOB_WORKER_EXIT_ERROR = 1016;
 
+	public static final ArrayList<Integer> ALL_JOB_EVENTS = new ArrayList<>(){{add(JOB_ACCEPTED);
+	add(JOB_STARTING); add(JOB_DATA); add(JOB_PROGRESS); add(JOB_FAILURE); add(JOB_COMPLETED); add(JOB_WORKER_ALIVE);
+	add(JOB_WORKER_EXIT_NORMAL); add(JOB_WORKER_EXIT_ERROR);}};
+
 	private VCSimulationIdentifier vcSimulationIdentifier = null;
 	private int jobIndex = -1;
 	private String hostName = null;

diff --git a/vcell-core/src/main/java/cbit/vcell/resource/PropertyLoader.java b/vcell-core/src/main/java/cbit/vcell/resource/PropertyLoader.java
@@ -80,6 +80,10 @@ public static void setConfigProvider(VCellConfigProvider configProvider) {
 	public static final String htcPbsHome		 			= record("vcell.htc.pbs.home",ValueType.GEN);
 	public static final String htcSgeHome		 			= record("vcell.htc.sge.home",ValueType.GEN);
 	public static final String htcNodeList                   = record("vcell.htc.nodelist",ValueType.GEN);
+	public static final String htcMinMemoryMB				= record("vcell.htc.memory.min.mb", ValueType.INT); // minimum memory request in MB, currently 4g
+	public static final String htcMaxMemoryMB				= record("vcell.htc.memory.max.mb", ValueType.INT); // maximum memory request in MB
+	public static final String htcPowerUserMemoryFloorMB	= record("vcell.htc.memory.pu.floor.mb", ValueType.INT); // MIN memory allowed if declared to be a power user, currently 50g (Previously Existing Value)
+	public static final String htcPowerUserMemoryMaxMB	    = record("vcell.htc.memory.pu.max.mb", ValueType.INT); // MAX memory allowed if declared to be a power user
 
 	public static final String htc_vcellfvsolver_docker_name = 	record("vcell.htc.vcellfvsolver.docker.name",ValueType.GEN);
 	public static final String htc_vcellfvsolver_solver_list =	record("vcell.htc.vcellfvsolver.solver.list",ValueType.GEN);

diff --git a/vcell-server/src/main/java/cbit/vcell/message/server/batch/sim/HtcSimulationWorker.java b/vcell-server/src/main/java/cbit/vcell/message/server/batch/sim/HtcSimulationWorker.java
@@ -678,7 +678,11 @@ public static void main(String[] args) throws IOException {
 		PropertyLoader.slurm_qos,
 		PropertyLoader.slurm_partition_pu,
 		PropertyLoader.slurm_reservation_pu,
-		PropertyLoader.slurm_qos_pu
+		PropertyLoader.slurm_qos_pu,
+		PropertyLoader.htcMinMemoryMB,
+		PropertyLoader.htcMaxMemoryMB,
+		PropertyLoader.htcPowerUserMemoryFloorMB,
+		PropertyLoader.htcPowerUserMemoryMaxMB
 	};
 
 

diff --git a/vcell-server/src/main/java/cbit/vcell/message/server/batch/sim/SolverPostprocessor.java b/vcell-server/src/main/java/cbit/vcell/message/server/batch/sim/SolverPostprocessor.java
@@ -146,7 +146,11 @@ private static Exception runPostprocessingCommands(String filename, Logger lg) {
 	}
 	private static final String POST_PROCESSOR_PROPERTIES[] = {
 			PropertyLoader.primarySimDataDirInternalProperty,
-			PropertyLoader.secondarySimDataDirInternalProperty
+			PropertyLoader.secondarySimDataDirInternalProperty,
+			PropertyLoader.mongodbDatabase,
+			PropertyLoader.jmsSimHostInternal,
+			PropertyLoader.jmsSimPortInternal,
+			PropertyLoader.jmsBlobMessageUseMongo
 		};
 
 }