Match the buffer usage tracking for leader and worker backends.
authorAmit Kapila
Fri, 3 Aug 2018 04:37:56 +0000 (10:07 +0530)
committerAmit Kapila
Fri, 3 Aug 2018 04:37:56 +0000 (10:07 +0530)
In the leader backend, we don't track the buffer usage for ExecutorStart
phase whereas in worker backend we track it for ExecutorStart phase as
well.  This leads to different value for buffer usage stats for the
parallel and non-parallel query.  Change the code so that worker backend
also starts tracking buffer usage after ExecutorStart.

Author: Amit Kapila and Robert Haas
Reviewed-by: Robert Haas and Andres Freund
Backpatch-through: 9.6 where this code was introduced
Discussion: https://postgr.es/m/86137f17-1dfb-42f9-7421-82fd786b04a1@anayrat.info

src/backend/executor/execParallel.c

index 6eed6db33556a3efb52d28ef8c5726bed5f6ad67..b1a3d91a1969045fb58f4ef407cb1a1897849964 100644 (file)
@@ -735,12 +735,19 @@ ParallelQueryMain(dsm_segment *seg, shm_toc *toc)
        instrument_options = instrumentation->instrument_options;
    queryDesc = ExecParallelGetQueryDesc(toc, receiver, instrument_options);
 
-   /* Prepare to track buffer usage during query execution. */
-   InstrStartParallelQuery();
-
    /* Start up the executor, have it run the plan, and then shut it down. */
    ExecutorStart(queryDesc, 0);
    ExecParallelInitializeWorker(queryDesc->planstate, toc);
+
+   /*
+    * Prepare to track buffer usage during query execution.
+    *
+    * We do this after starting up the executor to match what happens in the
+    * leader, which also doesn't count buffer accesses that occur during
+    * executor startup.
+    */
+   InstrStartParallelQuery();
+
    ExecutorRun(queryDesc, ForwardScanDirection, 0L);
    ExecutorFinish(queryDesc);