Jeff Janes wrote:
> I'm attaching a patch for each option. Each one independently solves the
> problem. But I think we should do both. There is no point in issuing
> unnecessary kill system calls, and there may also be more spurious wake-ups
> than just these ones.
I modified patch 1 a bit -- result attached. I would like to back-patch
this all the way back to 9.4, but there are a few conflicts due to
c29aff959dc so I think I'm going to do pg10 only unless I get some votes
that it should go further back. I think it should be fairly trivial to
resolve.
The other patch should be applied to master only IMO.
--
Álvaro Herrera https://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services
>From e482f52f792df3685f4f9d955d54e04570db5cc5 Mon Sep 17 00:00:00 2001
From: Alvaro Herrera <alvhe...@alvh.no-ip.org>
Date: Mon, 4 Sep 2017 13:18:44 +0200
Subject: [PATCH] fix basebackup throttle
---
src/backend/replication/basebackup.c | 34 +++++++++++++++++++++++-----------
1 file changed, 23 insertions(+), 11 deletions(-)
diff --git a/src/backend/replication/basebackup.c
b/src/backend/replication/basebackup.c
index 9776858f03..4db79d733c 100644
--- a/src/backend/replication/basebackup.c
+++ b/src/backend/replication/basebackup.c
@@ -1336,10 +1336,7 @@ _tarWriteDir(const char *pathbuf, int basepathlen,
struct stat *statbuf,
static void
throttle(size_t increment)
{
- TimeOffset elapsed,
- elapsed_min,
- sleep;
- int wait_result;
+ TimeOffset elapsed_min;
if (throttling_counter < 0)
return;
@@ -1348,14 +1345,29 @@ throttle(size_t increment)
if (throttling_counter < throttling_sample)
return;
- /* Time elapsed since the last measurement (and possible wake up). */
- elapsed = GetCurrentTimestamp() - throttled_last;
- /* How much should have elapsed at minimum? */
- elapsed_min = elapsed_min_unit * (throttling_counter /
throttling_sample);
- sleep = elapsed_min - elapsed;
- /* Only sleep if the transfer is faster than it should be. */
- if (sleep > 0)
+ /* How much time should have elapsed at minimum? */
+ elapsed_min = elapsed_min_unit *
+ (throttling_counter / throttling_sample);
+
+ /*
+ * Since the latch could be set repeatedly because of concurrently WAL
+ * activity, sleep enough times in a loop to ensure enough time has
+ * passed.
+ */
+ for (;;)
{
+ TimeOffset elapsed,
+ sleep;
+ int wait_result;
+
+ /* Time elapsed since the last measurement (and possible wake
up). */
+ elapsed = GetCurrentTimestamp() - throttled_last;
+
+ /* sleep if the transfer is faster than it should be */
+ sleep = elapsed_min - elapsed;
+ if (sleep <= 0)
+ break;
+
ResetLatch(MyLatch);
/* We're eating a potentially set latch, so check for
interrupts */
--
2.11.0
--
Sent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-hackers
