BUG-8056: place an upper bound on number of transactions processed

[controller.git] / opendaylight / md-sal / sal-distributed-datastore / src / main / java / org / opendaylight / controller / cluster / datastore / ShardDataTree.java

diff --git a/opendaylight/md-sal/sal-distributed-datastore/src/main/java/org/opendaylight/controller/cluster/datastore/ShardDataTree.java b/opendaylight/md-sal/sal-distributed-datastore/src/main/java/org/opendaylight/controller/cluster/datastore/ShardDataTree.java
index e36bc98f99115b32c36d85010cbf088b98533225..dcf632d215e930fa2dbc1be3af65e5bb02645298 100644 (file)
--- a/opendaylight/md-sal/sal-distributed-datastore/src/main/java/org/opendaylight/controller/cluster/datastore/ShardDataTree.java
+++ b/opendaylight/md-sal/sal-distributed-datastore/src/main/java/org/opendaylight/controller/cluster/datastore/ShardDataTree.java
@@ -106,8 +106,15 @@ public class ShardDataTree extends ShardDataTreeTransactionParent {
     private static final Timeout COMMIT_STEP_TIMEOUT = new Timeout(Duration.create(5, TimeUnit.SECONDS));
     private static final Logger LOG = LoggerFactory.getLogger(ShardDataTree.class);
 
-    private final Map<LocalHistoryIdentifier, ShardDataTreeTransactionChain> transactionChains = new HashMap<>();
+    /**
+     * Process this many transactions in a single batched run. If we exceed this limit, we need to schedule later
+     * execution to finish up the batch. This is necessary in case of a long list of transactions which progress
+     * immediately through their preCommit phase -- if that happens, their completion eats up stack frames and could
+     * result in StackOverflowError.
+     */
+    private static final int MAX_TRANSACTION_BATCH = 100;
 
+    private final Map<LocalHistoryIdentifier, ShardDataTreeTransactionChain> transactionChains = new HashMap<>();
     private final DataTreeCohortActorRegistry cohortRegistry = new DataTreeCohortActorRegistry();
     private final Queue<CommitEntry> pendingTransactions = new ArrayDeque<>();
     private final Queue<CommitEntry> pendingCommits = new ArrayDeque<>();
@@ -136,6 +143,8 @@ public class ShardDataTree extends ShardDataTreeTransactionParent {
 
     private SchemaContext schemaContext;
 
+    private int currentTransactionBatch;
+
     ShardDataTree(final Shard shard, final SchemaContext schemaContext, final TipProducingDataTree dataTree,
             final ShardDataTreeChangeListenerPublisher treeChangeListenerPublisher,
             final ShardDataChangeListenerPublisher dataChangeListenerPublisher, final String logContext,
@@ -188,6 +197,10 @@ public class ShardDataTree extends ShardDataTreeTransactionParent {
         this.schemaContext = Preconditions.checkNotNull(newSchemaContext);
     }
 
+    void resetTransactionBatch() {
+        currentTransactionBatch = 0;
+    }
+
     /**
      * Take a snapshot of current state for later recovery.
      *
@@ -699,8 +712,23 @@ public class ShardDataTree extends ShardDataTreeTransactionParent {
         return ret;
     }
 
+    /**
+     * Called some time after {@link #processNextPendingTransaction()} decides to stop processing.
+     */
+    void resumeNextPendingTransaction() {
+        LOG.debug("{}: attempting to resume transaction processing", logContext);
+        processNextPending();
+    }
+
     @SuppressWarnings("checkstyle:IllegalCatch")
     private void processNextPendingTransaction() {
+        ++currentTransactionBatch;
+        if (currentTransactionBatch > MAX_TRANSACTION_BATCH) {
+            LOG.debug("{}: Already processed {}, scheduling continuation", logContext, currentTransactionBatch);
+            shard.scheduleNextPendingTransaction();
+            return;
+        }
+
         processNextPending(pendingTransactions, State.CAN_COMMIT_PENDING, entry -> {
             final SimpleShardDataTreeCohort cohort = entry.cohort;
             final DataTreeModification modification = cohort.getDataTreeModification();