Hello Fujii-san,

On Fri, 30 Jul 2021 02:01:08 +0900
Fujii Masao <masao.fu...@oss.nttdata.com> wrote:

> 
> 
> On 2021/07/29 2:23, Fujii Masao wrote:
> > 
> > 
> > On 2021/07/28 16:15, Yugo NAGATA wrote:
> >>> I found another disconnect_all().
> >>>
> >>>     /* XXX should this be connection time? */
> >>>     disconnect_all(state, nclients);
> >>>
> >>> The measurement is also not necessary here.
> >>> So the above comment should be removed or updated?
> >>
> >> I think this disconnect_all will be a no-op because all connections should
> >> be already closed in threadRun(), but I left it just to be sure that
> >> connections are all cleaned-up. I updated the comment for explaining above.
> >>
> >> I attached the updated patch. Could you please look at this?
> > 
> > Thanks for updating the patch! LGTM.
> 
> This patch needs to be back-patched because it fixes the bug
> in measurement of disconnection delays. Thought?

This patch fixes three issues of connection time measurement:

1. The initial connection time is measured and stored into conn_duration
   but the result is never used.
2. The disconnection time are not measured even when -C is specified.
3. The disconnection time measurement at the end of threadRun() is
   not necessary.

The first one exists only in v14 and master, but others are also in v13 and
before. So, I think we can back-patch these fixes.

> But the patch fails to be back-patched to v13 or before because
> pgbench's time logic was changed by commit 547f04e734.
> Do you have the patches that can be back-patched to v13 or before?

I attached the patch for v13.

Regards,
Yugo Nagata

-- 
Yugo NAGATA <nag...@sraoss.co.jp>
diff --git a/src/bin/pgbench/pgbench.c b/src/bin/pgbench/pgbench.c
index 338c0152a2..3411556df8 100644
--- a/src/bin/pgbench/pgbench.c
+++ b/src/bin/pgbench/pgbench.c
@@ -3285,8 +3285,12 @@ advanceConnectionState(TState *thread, CState *st, StatsData *agg)
 
 				if (is_connect)
 				{
+					instr_time	start = now;
+
+					INSTR_TIME_SET_CURRENT_LAZY(start);
 					finishCon(st);
-					INSTR_TIME_SET_ZERO(now);
+					INSTR_TIME_SET_CURRENT(now);
+					INSTR_TIME_ACCUM_DIFF(thread->conn_time, now, start);
 				}
 
 				if ((st->cnt >= nxacts && duration <= 0) || timer_exceeded)
@@ -3310,6 +3314,17 @@ advanceConnectionState(TState *thread, CState *st, StatsData *agg)
 				 */
 			case CSTATE_ABORTED:
 			case CSTATE_FINISHED:
+				/*
+				 * In CSTATE_FINISHED state, this finishCon() is a no-op
+				 * under -C/--connect because all the connections that this
+				 * thread established should have already been closed at the end
+				 * of transactions. So we don't need to measure the disconnection
+				 * delays here.
+				 *
+				 * In CSTATE_ABORTED state, the measurement is no longer
+				 * necessary because we cannot report complete results anyways
+				 * in this case.
+				 */
 				finishCon(st);
 				return;
 		}
@@ -6210,6 +6225,12 @@ main(int argc, char **argv)
 		latency_late += thread->latency_late;
 		INSTR_TIME_ADD(conn_total_time, thread->conn_time);
 	}
+
+	/*
+	 * All connections should be already closed in threadRun(), so this
+	 * disconnect_all() will be a no-op, but clean up the connecions just
+	 * to be sure. We don't need to measure the disconnection delays here.
+	 */
 	disconnect_all(state, nclients);
 
 	/*
@@ -6237,8 +6258,7 @@ threadRun(void *arg)
 {
 	TState	   *thread = (TState *) arg;
 	CState	   *state = thread->state;
-	instr_time	start,
-				end;
+	instr_time	start;
 	int			nstate = thread->nstate;
 	int			remains = nstate;	/* number of remaining clients */
 	socket_set *sockets = alloc_socket_set(nstate);
@@ -6502,10 +6522,7 @@ threadRun(void *arg)
 	}
 
 done:
-	INSTR_TIME_SET_CURRENT(start);
 	disconnect_all(state, nstate);
-	INSTR_TIME_SET_CURRENT(end);
-	INSTR_TIME_ACCUM_DIFF(thread->conn_time, end, start);
 	if (thread->logfile)
 	{
 		if (agg_interval > 0)

Reply via email to