HDDS-5252. PipelinePlacementPolicy filter out datanodes with not enough space.

hadoop-hdds/server-scm/src/main/java/org/apache/hadoop/hdds/scm/pipeline/PipelinePlacementPolicy.java

@@ -22,6 +22,7 @@
 import com.google.common.base.Preconditions;
 import org.apache.hadoop.hdds.client.RatisReplicationConfig;
 import org.apache.hadoop.hdds.conf.ConfigurationSource;
+import org.apache.hadoop.hdds.conf.StorageUnit;
 import org.apache.hadoop.hdds.protocol.DatanodeDetails;
 import org.apache.hadoop.hdds.protocol.proto.HddsProtos;
 import org.apache.hadoop.hdds.protocol.proto.HddsProtos.ReplicationFactor;
@@ -45,9 +46,10 @@
  * and network topology to supply pipeline creation.
  * <p>
  * 1. get a list of healthy nodes
- * 2. filter out nodes that are not too heavily engaged in other pipelines
- * 3. Choose an anchor node among the viable nodes.
- * 4. Choose other nodes around the anchor node based on network topology
+ * 2. filter out nodes that have space less than container size.
+ * 3. filter out nodes that are not too heavily engaged in other pipelines
+ * 4. Choose an anchor node among the viable nodes.
+ * 5. Choose other nodes around the anchor node based on network topology
  */
 public final class PipelinePlacementPolicy extends SCMCommonPlacementPolicy {
   @VisibleForTesting
@@ -151,11 +153,30 @@ List<DatanodeDetails> filterViableNodes(
           SCMException.ResultCodes.FAILED_TO_FIND_SUITABLE_NODE);
     }
 
+    long sizeRequired = (long) conf.getStorageSize(
+        ScmConfigKeys.OZONE_SCM_CONTAINER_SIZE,
+        ScmConfigKeys.OZONE_SCM_CONTAINER_SIZE_DEFAULT,
+        StorageUnit.BYTES);
+
+    // filter nodes that don't even have space for one container
+    List<DatanodeDetails> canHoldList = healthyNodes.stream().filter(d ->
+        hasEnoughSpace(d, sizeRequired)).collect(Collectors.toList());
+
+    if (canHoldList.size() < nodesRequired) {
+      msg = String.format("Pipeline creation failed due to no sufficient" +
+          " healthy datanodes with enough space for even a single container." +
+          " Required %d. Found %d. Container size %d.",
+          nodesRequired, canHoldList.size(), sizeRequired);
+      LOG.warn(msg);
+      throw new SCMException(msg,
+          SCMException.ResultCodes.FAILED_TO_FIND_SUITABLE_NODE);
+    }
+
     // filter nodes that meet the size and pipeline engagement criteria.
     // Pipeline placement doesn't take node space left into account.
     // Sort the DNs by pipeline load.
     // TODO check if sorting could cause performance issue: HDDS-3466.
-    List<DatanodeDetails> healthyList = healthyNodes.stream()
+    List<DatanodeDetails> healthyList = canHoldList.stream()
         .map(d ->
             new DnWithPipelines(d, currentPipelineCount(d, nodesRequired)))
         .filter(d ->

hadoop-hdds/server-scm/src/test/java/org/apache/hadoop/hdds/scm/node/TestDeadNodeHandler.java

@@ -61,6 +61,7 @@
 import org.apache.hadoop.hdds.server.events.EventPublisher;
 
 import org.apache.hadoop.hdds.server.events.EventQueue;
+import org.apache.hadoop.ozone.OzoneConsts;
 import org.apache.hadoop.security.authentication.client
     .AuthenticationException;
 import org.apache.hadoop.test.GenericTestUtils;
@@ -131,7 +132,8 @@ public void testOnMessage() throws Exception {
         .concat("/" + datanode1.getUuidString());
 
     StorageReportProto storageOne = TestUtils.createStorageReport(
-        datanode1.getUuid(), storagePath, 100, 10, 90, null);
+        datanode1.getUuid(), storagePath, 100 * OzoneConsts.TB,
+        10 * OzoneConsts.TB, 90 * OzoneConsts.TB, null);
 
     // Exit safemode, as otherwise the safemode precheck will prevent pipelines
     // from getting created. Due to how this test is wired up, safemode will

hadoop-hdds/server-scm/src/test/java/org/apache/hadoop/hdds/scm/pipeline/TestPipelinePlacementPolicy.java

@@ -58,6 +58,7 @@
 import static junit.framework.TestCase.assertEquals;
 import static junit.framework.TestCase.assertTrue;
 import static org.apache.hadoop.hdds.scm.ScmConfigKeys.OZONE_DATANODE_PIPELINE_LIMIT;
+import static org.apache.hadoop.hdds.scm.ScmConfigKeys.OZONE_SCM_CONTAINER_SIZE;
 import static org.apache.hadoop.hdds.scm.net.NetConstants.LEAF_SCHEMA;
 import static org.apache.hadoop.hdds.scm.net.NetConstants.RACK_SCHEMA;
 import static org.apache.hadoop.hdds.scm.net.NetConstants.ROOT_SCHEMA;
@@ -174,6 +175,30 @@ public void testChooseNodeWithSingleNodeRack() throws SCMException {
     Assert.assertNotEquals(results.get(1).getNetworkLocation(),
         results.get(2).getNetworkLocation());
   }
+
+  @Test
+  public void testChooseNodeNotEnoughSpace() throws SCMException {
+    // A huge container size
+    conf.set(OZONE_SCM_CONTAINER_SIZE, "10TB");
+
+    // There is only one node on 3 racks altogether.
+    List<DatanodeDetails> datanodes = new ArrayList<>();
+    for (Node node : SINGLE_NODE_RACK) {
+      DatanodeDetails datanode = overwriteLocationInNode(
+          MockDatanodeDetails.randomDatanodeDetails(), node);
+      datanodes.add(datanode);
+    }
+    MockNodeManager localNodeManager = new MockNodeManager(initTopology(),
+        datanodes, false, datanodes.size());
+    PipelinePlacementPolicy localPlacementPolicy = new PipelinePlacementPolicy(
+        localNodeManager, new PipelineStateManager(), conf);
+    int nodesRequired = HddsProtos.ReplicationFactor.THREE.getNumber();
+
+    thrownExp.expect(SCMException.class);
+    thrownExp.expectMessage("enough space for even a single container");
+    localPlacementPolicy.chooseDatanodes(new ArrayList<>(datanodes.size()),
+        new ArrayList<>(datanodes.size()), nodesRequired, 0);
+  }
 
   @Test
   public void testPickLowestLoadAnchor() throws IOException{