temp commit

apache · Jun 27, 2022 · 301b210 · 301b210
1 parent 5feb168
commit 301b210
Show file tree

Hide file tree

Showing 5 changed files with 58 additions and 34 deletions.
diff --git a/...ver/server-api/src/main/java/org/apache/submarine/server/api/spec/ExperimentTaskSpec.java b/...ver/server-api/src/main/java/org/apache/submarine/server/api/spec/ExperimentTaskSpec.java
@@ -37,7 +37,7 @@ public class ExperimentTaskSpec {
   private String cmd;
   private Map<String, String> envVars;
 
-  // should ignored in JSON Serialization
+  // should be ignored in JSON Serialization
   private Map<String, String> resourceMap;
 
   public ExperimentTaskSpec() {

diff --git a/...src/main/java/org/apache/submarine/server/submitter/k8s/model/mljob/MLJobReplicaSpec.java b/...src/main/java/org/apache/submarine/server/submitter/k8s/model/mljob/MLJobReplicaSpec.java
@@ -101,7 +101,8 @@ public String getContainerMemMB() {
     V1PodTemplateSpec podSpec = getTemplate();
     return String.join(" ",
         podSpec.getSpec().getContainers().get(0)
-            .getResources().getLimits().get("memory").
+//            .getResources().getLimits().get("memory").
+            .getResources().getRequests().get("memory").
             getNumber().divide(BigDecimal.valueOf(1000000)).toString() + "M");
   }
 

diff --git a/.../src/main/java/org/apache/submarine/server/submitter/k8s/parser/ExperimentSpecParser.java b/.../src/main/java/org/apache/submarine/server/submitter/k8s/parser/ExperimentSpecParser.java
@@ -100,12 +100,12 @@ public static XGBoostJobSpec parseXGBoostJobSpec(ExperimentSpec experimentSpec)
     for (Map.Entry<String, ExperimentTaskSpec> entry : experimentSpec.getSpec().entrySet()) {
       String replicaType = entry.getKey();
       ExperimentTaskSpec taskSpec = entry.getValue();
-      
+
       if (TFJobReplicaType.isSupportedReplicaType(replicaType)) {
         MLJobReplicaSpec replicaSpec = new MLJobReplicaSpec();
         replicaSpec.setReplicas(taskSpec.getReplicas());
         V1PodTemplateSpec podTemplateSpec = parseTemplateSpec(taskSpec, experimentSpec);
-        
+
         replicaSpec.setTemplate(podTemplateSpec);
         replicaSpecMap.put(XGBoostJobReplicaType.valueOf(replicaType), replicaSpec);
       } else {
@@ -140,7 +140,7 @@ public static PyTorchJobSpec parsePyTorchJobSpec(ExperimentSpec experimentSpec)
         MLJobReplicaSpec replicaSpec = new MLJobReplicaSpec();
         replicaSpec.setReplicas(taskSpec.getReplicas());
         V1PodTemplateSpec podTemplateSpec = parseTemplateSpec(taskSpec, experimentSpec);
-      
+
         replicaSpec.setTemplate(podTemplateSpec);
         replicaSpecMap.put(PyTorchJobReplicaType.valueOf(replicaType), replicaSpec);
       } else {
@@ -181,12 +181,12 @@ private static TFJobSpec parseTFJobSpec(ExperimentSpec experimentSpec)
     for (Map.Entry<String, ExperimentTaskSpec> entry : experimentSpec.getSpec().entrySet()) {
       String replicaType = entry.getKey();
       ExperimentTaskSpec taskSpec = entry.getValue();
-      
+
       if (TFJobReplicaType.isSupportedReplicaType(replicaType)) {
         MLJobReplicaSpec replicaSpec = new MLJobReplicaSpec();
         replicaSpec.setReplicas(taskSpec.getReplicas());
         V1PodTemplateSpec podTemplateSpec = parseTemplateSpec(taskSpec, experimentSpec);
-        
+
         replicaSpec.setTemplate(podTemplateSpec);
         replicaSpecMap.put(TFJobReplicaType.valueOf(replicaType), replicaSpec);
       } else {

diff --git a/...k8s/src/test/java/org/apache/submarine/server/submitter/k8s/ExperimentSpecParserTest.java b/...k8s/src/test/java/org/apache/submarine/server/submitter/k8s/ExperimentSpecParserTest.java
@@ -61,30 +61,23 @@ public class ExperimentSpecParserTest extends SpecBuilder {
   private static SubmarineConfiguration conf =
       SubmarineConfiguration.getInstance();
 
-
   @Test
-  public void testValidTensorFlowExperiment() {
-    int x = 100;
-    Assert.assertTrue(x == (int)100.0);
-  }
+  public void testValidTensorFlowExperiment() throws IOException,
+      URISyntaxException, InvalidSpecException {
+    ExperimentSpec experimentSpec = (ExperimentSpec) buildFromJsonFile(ExperimentSpec.class, tfJobReqFile);
+    TFJob tfJob = (TFJob) ExperimentSpecParser.parseJob(experimentSpec);
+    validateMetadata(experimentSpec.getMeta(), tfJob.getMetadata(),
+        ExperimentMeta.SupportedMLFramework.TENSORFLOW.getName().toLowerCase()
+    );
+    // Validate ExperimentMeta without envVars. Related to SUBMARINE-534.
+    experimentSpec.getMeta().setEnvVars(null);
+    validateMetadata(experimentSpec.getMeta(), tfJob.getMetadata(),
+            ExperimentMeta.SupportedMLFramework.TENSORFLOW.getName().toLowerCase()
+    );
 
-  // @Test
-  // public void testValidTensorFlowExperiment() throws IOException,
-  //     URISyntaxException, InvalidSpecException {
-  //   ExperimentSpec experimentSpec = (ExperimentSpec) buildFromJsonFile(ExperimentSpec.class, tfJobReqFile);
-  //   TFJob tfJob = (TFJob) ExperimentSpecParser.parseJob(experimentSpec);
-  //   validateMetadata(experimentSpec.getMeta(), tfJob.getMetadata(),
-  //       ExperimentMeta.SupportedMLFramework.TENSORFLOW.getName().toLowerCase()
-  //   );
-  //   // Validate ExperimentMeta without envVars. Related to SUBMARINE-534.
-  //   experimentSpec.getMeta().setEnvVars(null);
-  //   validateMetadata(experimentSpec.getMeta(), tfJob.getMetadata(),
-  //           ExperimentMeta.SupportedMLFramework.TENSORFLOW.getName().toLowerCase()
-  //   );
-
-  //   validateReplicaSpec(experimentSpec, tfJob, TFJobReplicaType.Ps);
-  //   validateReplicaSpec(experimentSpec, tfJob, TFJobReplicaType.Worker);
-  // }
+    validateReplicaSpec(experimentSpec, tfJob, TFJobReplicaType.Ps);
+    validateReplicaSpec(experimentSpec, tfJob, TFJobReplicaType.Worker);
+  }
 
   @Test
   public void testInvalidTensorFlowExperiment() throws IOException,
@@ -160,7 +153,8 @@ public void testInvalidPyTorchJobSpec() throws IOException,
   @Test
   public void testValidXGBoostExperiment() throws IOException,
       URISyntaxException, InvalidSpecException {
-    ExperimentSpec experimentSpec = (ExperimentSpec) buildFromJsonFile(ExperimentSpec.class, xgboostJobReqFile);
+    ExperimentSpec experimentSpec = (ExperimentSpec) buildFromJsonFile(ExperimentSpec.class,
+        xgboostJobReqFile);
     XGBoostJob xgboostJob = (XGBoostJob) ExperimentSpecParser.parseJob(experimentSpec);
     validateMetadata(experimentSpec.getMeta(), xgboostJob.getMetadata(),
         ExperimentMeta.SupportedMLFramework.XGBOOST.getName().toLowerCase()
@@ -173,7 +167,8 @@ public void testValidXGBoostExperiment() throws IOException,
   @Test
   public void testInvalidXGBoostExperiment() throws IOException,
       URISyntaxException {
-    ExperimentSpec experimentSpec = (ExperimentSpec) buildFromJsonFile(ExperimentSpec.class, xgboostJobReqFile);
+    ExperimentSpec experimentSpec = (ExperimentSpec) buildFromJsonFile(ExperimentSpec.class,
+        xgboostJobReqFile);
     // Case 1. Invalid framework name
     experimentSpec.getMeta().setFramework("fooframework");
     try {
@@ -185,7 +180,8 @@ public void testInvalidXGBoostExperiment() throws IOException,
 
     // Case 2. Invalid XGBoost Replica name. It can only be "master" and "worker"
     experimentSpec = (ExperimentSpec) buildFromJsonFile(ExperimentSpec.class, xgboostJobReqFile);
-    experimentSpec.getSpec().put("foo", experimentSpec.getSpec().get(XGBoostJobReplicaType.Master.getTypeName()));
+    experimentSpec.getSpec().put("foo", experimentSpec.getSpec().get(
+        XGBoostJobReplicaType.Master.getTypeName()));
     experimentSpec.getSpec().remove(XGBoostJobReplicaType.Master.getTypeName());
     try {
       ExperimentSpecParser.parseJob(experimentSpec);
@@ -213,14 +209,15 @@ private void validateReplicaSpec(ExperimentSpec experimentSpec,
       mlJobReplicaSpec = ((XGBoostJob) mlJob).getSpec().getReplicaSpecs().get(type);
     }
     Assert.assertNotNull(mlJobReplicaSpec);
-
+    
     ExperimentTaskSpec definedPyTorchMasterTask = experimentSpec.getSpec().
         get(type.getTypeName());
+
     // replica
     int expectedMasterReplica = definedPyTorchMasterTask.getReplicas();
     Assert.assertEquals(expectedMasterReplica,
         (int) mlJobReplicaSpec.getReplicas());
-    // Image
+    // image
     String expectedMasterImage = definedPyTorchMasterTask.getImage() == null ?
         experimentSpec.getEnvironment().getImage() : definedPyTorchMasterTask.getImage();
     String actualMasterImage = mlJobReplicaSpec.getContainerImageName();

diff --git a/submarine-server/server-submitter/submitter-k8s/src/test/resources/xgboost_job_req.json b/submarine-server/server-submitter/submitter-k8s/src/test/resources/xgboost_job_req.json
@@ -0,0 +1,26 @@
+{
+  "meta": {
+    "name": "xgboost-dist-mnist",
+    "namespace": "submarine",
+    "framework": "XGBoost",
+    "cmd": "python /var/mnist.py --backend gloo",
+    "envVars": {
+      "ENV_1": "ENV1"
+    }
+  },
+  "environment": {
+    "image": "apache/submarine:pytorch-dist-mnist-1.0"
+  },
+  "spec": {
+    "Master": {
+      "name": "master",
+      "replicas": 1,
+      "resources": "cpu=2,memory=2048M"
+    },
+    "Worker": {
+      "name": "worker",
+      "replicas": 2,
+      "resources": "cpu=1,memory=1024M"
+    }
+  }
+}