[DRILL-3936] We don't handle out of memory condition during build phase of hash join

• Type: Bug
• Status: Open
• Priority: Major
• Resolution: Unresolved
• Component/s: Execution - Relational Operators
• Assignee: Boaz Ben-Zvi
• Reporter: Victoria Markman
• Created: 14/Oct/15 23:13
• Updated: 12/Dec/17 03:35

Description

It looks like we just fall through ( see excerpt from HashJoinBatch.java below )

  public void executeBuildPhase() throws SchemaChangeException, ClassTransformationException, IOException {
    //Setup the underlying hash table

    // skip first batch if count is zero, as it may be an empty schema batch
    if (right.getRecordCount() == 0) {
      for (final VectorWrapper<?> w : right) {
        w.clear();
      }
      rightUpstream = next(right);
    }

    boolean moreData = true;

    while (moreData) {
      switch (rightUpstream) {
      case OUT_OF_MEMORY:
      case NONE:
      case NOT_YET:
      case STOP:
        moreData = false;
        continue;
...

We don't handle it later either:

  public IterOutcome innerNext() {
    try {
      /* If we are here for the first time, execute the build phase of the
       * hash join and setup the run time generated class for the probe side
       */
      if (state == BatchState.FIRST) {
        // Build the hash table, using the build side record batches.
        executeBuildPhase();
        //                IterOutcome next = next(HashJoinHelper.LEFT_INPUT, left);
        hashJoinProbe.setupHashJoinProbe(context, hyperContainer, left, left.getRecordCount(), this, hashTable,
            hjHelper, joinType);

        // Update the hash table related stats for the operator
        updateStats(this.hashTable);
      }
....