[HBASE-10805] Speed up KeyValueHeap.next() a bit

• Type: Bug
• Status: Closed
• Priority: Major
• Resolution: Fixed
• Fix Version/s: 0.98.1, 0.99.0, 0.94.19, 0.96.3
• Hadoop Flags: Reviewed
• Assignee: Lars Hofhansl
• Reporter: Lars Hofhansl
• Created: 21/Mar/14 04:41
• Updated: 28/Apr/14 22:29
• Resolved: 22/Mar/14 04:09

Description

See discussion on HBASE-9969.
This was brought up by mcorgan. Even though I didn't believe him first.

      KeyValueScanner topScanner = this.heap.peek();
      if (topScanner == null ||
          this.comparator.compare(kvNext, topScanner.peek()) >= 0) {
        this.heap.add(this.current);
        this.current = pollRealKV();
      }

We already have the invariant everywhere this.current always has a real KV polled. So adding current back to the heap followed by pollRealKV() is a no-op if this.current is the only scanner anyway.

This can be changed to:

      ...
      if (topScanner != null &&
          this.comparator.compare(kvNext, topScanner.peek()) >= 0) {
      ...

With 20m rows, 5 cols each, everything in the cache, fully compacted - i.e. one HFile per store, a scan that filters everything at the server through all 100m KVs takes 15.1s without the change and 13.3s with it, so a 12% improvement.