SOLR-11807: Restoring collection now treats maxShardsPerNode=-1 as unlimited

solr/CHANGES.txt

@@ -98,6 +98,8 @@ Bug Fixes
 
 * SOLR-11216: Race condition in PeerSync (Cao Manh Dat)
 
+* SOLR-11807: Restoring collection now treats maxShardsPerNode=-1 as unlimited (Varun Thacker)
+
 Optimizations
 ----------------------
 

solr/core/src/java/org/apache/solr/cloud/api/collections/RestoreCmd.java

@@ -128,7 +128,7 @@ public class RestoreCmd implements OverseerCollectionMessageHandler.Cmd {
     
     int maxShardsPerNode = message.getInt(MAX_SHARDS_PER_NODE, backupCollectionState.getMaxShardsPerNode());
     int availableNodeCount = nodeList.size();
-    if ((numShards * totalReplicasPerShard) > (availableNodeCount * maxShardsPerNode)) {
+    if (maxShardsPerNode != -1 && (numShards * totalReplicasPerShard) > (availableNodeCount * maxShardsPerNode)) {
       throw new SolrException(ErrorCode.BAD_REQUEST,
           String.format(Locale.ROOT, "Solr cloud with available number of nodes:%d is insufficient for"
               + " restoring a collection with %d shards, total replicas per shard %d and maxShardsPerNode %d."

solr/core/src/test/org/apache/solr/cloud/api/collections/AbstractCloudBackupRestoreTestCase.java

@@ -100,7 +100,9 @@ public abstract class AbstractCloudBackupRestoreTestCase extends SolrCloudTestCa
         CollectionAdminRequest.createCollectionWithImplicitRouter(getCollectionName(), "conf1", "shard1,shard2", replFactor, numTlogReplicas, numPullReplicas) :
         CollectionAdminRequest.createCollection(getCollectionName(), "conf1", NUM_SHARDS, replFactor, numTlogReplicas, numPullReplicas);
 
-    if (NUM_SHARDS * (backupReplFactor) > cluster.getJettySolrRunners().size() || random().nextBoolean()) {
+    if (random().nextBoolean()) {
+      create.setMaxShardsPerNode(-1);
+    } else if (NUM_SHARDS * (backupReplFactor) > cluster.getJettySolrRunners().size() || random().nextBoolean()) {
       create.setMaxShardsPerNode((int)Math.ceil(NUM_SHARDS * backupReplFactor / (double) cluster.getJettySolrRunners().size()));//just to assert it survives the restoration
       if (doSplitShardOperation) {
         create.setMaxShardsPerNode(create.getMaxShardsPerNode() * 2);
@@ -265,8 +267,7 @@ public abstract class AbstractCloudBackupRestoreTestCase extends SolrCloudTestCa
     int restoreReplFactor = restoreReplcationFactor + restoreTlogReplicas + restorePullReplicas;
 
     boolean isMaxShardsPerNodeExternal = false;
-    int restoreMaxShardsPerNode = -1;
+    boolean isMaxShardsUnlimited = false;
-    {
     CollectionAdminRequest.Restore restore = CollectionAdminRequest.restoreCollection(restoreCollectionName, backupName)
         .setLocation(backupLocation).setRepositoryName(getBackupRepoName());
 
@@ -276,16 +277,20 @@ public abstract class AbstractCloudBackupRestoreTestCase extends SolrCloudTestCa
       restore.setTlogReplicas(restoreTlogReplicas);
       restore.setPullReplicas(restorePullReplicas);
     }
-
+    final int restoreMaxShardsPerNode = (int) Math.ceil((restoreReplFactor * numShards/(double) cluster.getJettySolrRunners().size()));
 
     if (restoreReplFactor > backupReplFactor) { //else the backup maxShardsPerNode should be enough
-        restoreMaxShardsPerNode = (int)Math.ceil((restoreReplFactor * numShards/(double) cluster.getJettySolrRunners().size()));
       log.info("numShards={} restoreReplFactor={} maxShardsPerNode={} totalNodes={}",
           numShards, restoreReplFactor, restoreMaxShardsPerNode, cluster.getJettySolrRunners().size());
-        isMaxShardsPerNodeExternal = true;
 
+      if (random().nextBoolean()) { //set it to -1
+        isMaxShardsUnlimited = true;
+        restore.setMaxShardsPerNode(-1);
+      } else {
+        isMaxShardsPerNodeExternal = true;
         restore.setMaxShardsPerNode(restoreMaxShardsPerNode);
       }
+    }
 
     if (rarely()) { // Try with createNodeSet configuration
       int nodeSetSize = cluster.getJettySolrRunners().size() / 2;
@@ -318,7 +323,6 @@ public abstract class AbstractCloudBackupRestoreTestCase extends SolrCloudTestCa
     }
     AbstractDistribZkTestBase.waitForRecoveriesToFinish(
         restoreCollectionName, cluster.getSolrClient().getZkStateReader(), log.isDebugEnabled(), true, 30);
-    }
 
     //Check the number of results are the same
     DocCollection restoreCollection = client.getZkStateReader().getClusterState().getCollection(restoreCollectionName);
@@ -339,8 +343,8 @@ public abstract class AbstractCloudBackupRestoreTestCase extends SolrCloudTestCa
       numReplicasByNodeName.put(x.getNodeName(), numReplicasByNodeName.getOrDefault(x.getNodeName(), 0) + 1);
     });
     numReplicasByNodeName.forEach((k, v) -> {
-      assertTrue("Node " + k + " has " + v + " replicas. Expected num replicas : " + restoreCollection.getMaxShardsPerNode() ,
+      assertTrue("Node " + k + " has " + v + " replicas. Expected num replicas : " + restoreMaxShardsPerNode
-          v <= restoreCollection.getMaxShardsPerNode());
+              + " state file \n" + restoreCollection, v <= restoreMaxShardsPerNode);
     });
 
     assertEquals(restoreCollection.toString(), restoreReplcationFactor, restoreCollection.getReplicationFactor().intValue());
@@ -349,6 +353,8 @@ public abstract class AbstractCloudBackupRestoreTestCase extends SolrCloudTestCa
     assertEquals(restoreCollection.toString(), restoreTlogReplicas, restoreCollection.getNumTlogReplicas().intValue());
     if (isMaxShardsPerNodeExternal) {
       assertEquals(restoreCollectionName, restoreMaxShardsPerNode, restoreCollection.getMaxShardsPerNode());
+    } else if (isMaxShardsUnlimited){
+      assertEquals(restoreCollectionName, -1, restoreCollection.getMaxShardsPerNode());
     } else {
       assertEquals(restoreCollectionName, backupCollection.getMaxShardsPerNode(), restoreCollection.getMaxShardsPerNode());
     }