Hi,
On Tue, Sep 11, 2018 at 12:58:40PM +0100, David King wrote:
> Hi,
>
> > I just tested, and it seems to happen with the latest 1.8, but not with
> 1.9.
> > Not using kqueue (by using -dk) seems to work around the issue.
> > It's quite interesting it doesn't happen on Centos, it means it is
> probably
> > kqueue-specific (or that it works by accident with epoll/poll/select).
> > I'm investigating.
> >
> > Regards,
> >
> > Olivier
>
>
> so i've been running some tests from builds from source, it seems to be
> 1.8.13 specific, as 1.8.12 works fine, and as Oliver said 1.9 is OK as well
>
> Thanks
>
> Dave
Ok I think I figured it out.
The bug was present in master too, but was masked for some reason.
The patches attached should fix this. The first one is for master, and the
second one for 1.8, as the master patch didn't apply cleanly on 1.8.
Regards,
Olivier
>From d950da31340528c37173fc74d1c0f635c977cd03 Mon Sep 17 00:00:00 2001
From: Olivier Houchard <ohouch...@haproxy.com>
Date: Tue, 11 Sep 2018 14:44:51 +0200
Subject: [PATCH] BUG/MAJOR: kqueue: Don't reset the changes number by
accident.
In _update_fd(), if the fd wasn't polled, and we don't want it to be polled,
we just returned 0, however, we should return changes instead, or all previous
changes will be lost.
This should be backported to 1.8.
---
src/ev_kqueue.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/src/ev_kqueue.c b/src/ev_kqueue.c
index 087a07e7..e2f04f70 100644
--- a/src/ev_kqueue.c
+++ b/src/ev_kqueue.c
@@ -44,7 +44,7 @@ static int _update_fd(int fd, int start)
if (!(fdtab[fd].thread_mask & tid_bit) || !(en & FD_EV_POLLED_RW)) {
if (!(polled_mask[fd] & tid_bit)) {
/* fd was not watched, it's still not */
- return 0;
+ return changes;
}
/* fd totally removed from poll list */
EV_SET(&kev[changes++], fd, EVFILT_READ, EV_DELETE, 0, 0, NULL);
--
2.14.3
>From e987a5bda41a1b560a352b4ec2d54a8ebcd5965a Mon Sep 17 00:00:00 2001
From: Olivier Houchard <ohouch...@haproxy.com>
Date: Tue, 11 Sep 2018 14:44:51 +0200
Subject: [PATCH] BUG/MAJOR: kqueue: Don't reset the changes number by
accident.
In _update_fd(), if the fd wasn't polled, and we don't want it to be polled,
we just returned 0, however, we should return changes instead, or all previous
changes will be lost.
This should be backported to 1.8.
---
src/ev_kqueue.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/src/ev_kqueue.c b/src/ev_kqueue.c
index 1f4762e6..c88cf6f3 100644
--- a/src/ev_kqueue.c
+++ b/src/ev_kqueue.c
@@ -44,7 +44,7 @@ static int _update_fd(int fd, int start)
if (!(fdtab[fd].thread_mask & tid_bit) || !(en & FD_EV_POLLED_RW)) {
if (!(fdtab[fd].polled_mask & tid_bit)) {
/* fd was not watched, it's still not */
- return 0;
+ return changes;
}
/* fd totally removed from poll list */
EV_SET(&kev[changes++], fd, EVFILT_READ, EV_DELETE, 0, 0, NULL);
--
2.14.3
