[2/3] incubator-trafodion git commit: Add comments to explain the use of stack variable to minimize impact on runtime code based optimization.

[db...@apache.org]

Add comments to explain the use of stack variable to minimize impact on runtime code based optimization.


Project: http://git-wip-us.apache.org/repos/asf/incubator-trafodion/repo
Commit: http://git-wip-us.apache.org/repos/asf/incubator-trafodion/commit/263b90cc
Tree: http://git-wip-us.apache.org/repos/asf/incubator-trafodion/tree/263b90cc
Diff: http://git-wip-us.apache.org/repos/asf/incubator-trafodion/diff/263b90cc

Branch: refs/heads/master
Commit: 263b90ccd96eab24cb697d8cacc1b669f4ce225c
Parents: 96bcc40
Author: Eric Owhadi <er...@esgyn.com>
Authored: Mon Mar 21 18:06:47 2016 +0000
Committer: Eric Owhadi <er...@esgyn.com>
Committed: Mon Mar 21 18:06:47 2016 +0000

----------------------------------------------------------------------
 core/sql/executor/ExHbaseSelect.cpp | 9 +++++++++
 1 file changed, 9 insertions(+)
----------------------------------------------------------------------


http://git-wip-us.apache.org/repos/asf/incubator-trafodion/blob/263b90cc/core/sql/executor/ExHbaseSelect.cpp
----------------------------------------------------------------------
diff --git a/core/sql/executor/ExHbaseSelect.cpp b/core/sql/executor/ExHbaseSelect.cpp
index 04478f0..2bd904b 100644
--- a/core/sql/executor/ExHbaseSelect.cpp
+++ b/core/sql/executor/ExHbaseSelect.cpp
@@ -445,6 +445,15 @@ ExWorkProcRetcode ExHbaseScanSQTaskTcb::work(short &rc)
   rc = 0;
   Lng32 remainingInBatch = batchSize_;
   NABoolean isFirstBatch = false;
+  // isFirstInBatch is a stack variable for optimization reason. It is used for the mdam small scanner optimization heuristic that
+  // is performed at runtime. Since this function is invoke intensively for all scan (mdam or regular scan), minimizing CPU/memory access
+  // impact on runtime code to a strict minimum is attempted. Given that we are trying to detect if the actual scan is bellow the size
+  // of an HBase block, having the runtime logic performing the detection only affect the first work invoke looks like the right idea.
+  // and leveraging an existing counter (remainingInBatch) instead of creating a new one. The reasonable asumption to allow this is that
+  // 1- batchSize_ being 8K, most likely times the row size, we are good in assuming that first hbase block will fit in batchSize
+  // 2- parent buffer size will be large enough to deal with one HBAse_Block_size without having to rely on re-invoking work in the middle.
+  // and anyway, if none of the reasonable assumption is true, then that's fine, the heuristic won't work, and we will use regular scanner,
+  // meaning optimization is off for the scan part of MDAM (still on for the probe side of it).
 
   while (1)
     {