Bug#927471: curl: Regression that fails to exhaust socket data

[Guillem Jover]

Source: curl
Source-Version: 7.64.0-2
Severity: serious
Control: affects -1 rtorrent

Hi!

I've started noticing rtorrent busy-looping at some points after
finishing a torrent. stracing and gdb'ing the process it was doing
that in its main loop, spamming on gettimeofday() and epoll_wait().

Checking the rtorrent dependencies I've not seen much suspicious
updated recently, except for curl. I checked then the upstream bug
trackers and noticed quite many instances of "100% cpu usage" reports,
such as . And that
one points to old and new curl issue. The last instance of it appears
to have been fixed with 
in 7.64.1.

So I think curl should be either upgraded to that release, or the
relevant commits cherry-picked.

Thanks,
Guillem

Bug#927471: curl: Regression that fails to exhaust socket data

[Alessandro Ghedini]

On Sat, Apr 20, 2019 at 01:39:36PM +0200, Guillem Jover wrote:
> Source: curl
> Source-Version: 7.64.0-2
> Severity: serious
> Control: affects -1 rtorrent
> 
> Hi!

Hello,

> I've started noticing rtorrent busy-looping at some points after
> finishing a torrent. stracing and gdb'ing the process it was doing
> that in its main loop, spamming on gettimeofday() and epoll_wait().
> 
> Checking the rtorrent dependencies I've not seen much suspicious
> updated recently, except for curl. I checked then the upstream bug
> trackers and noticed quite many instances of "100% cpu usage" reports,
> such as . And that
> one points to old and new curl issue. The last instance of it appears
> to have been fixed with 
> in 7.64.1.
> 
> So I think curl should be either upgraded to that release, or the
> relevant commits cherry-picked.

I prepared an update for the curl package with the two upstream patches applied
(38d8e1b and 4015fae), and uploaded the result for amd64 at:
https://people.debian.org/~ghedo/libcurl4_7.64.0-3_amd64.deb

Guillem, could you try this version of the package and see if it fixes the
issue for you? Once that's confirmed I'll upload to sid and ask for unblock.

Cheers


signature.asc
Description: PGP signature

Bug#927471: curl: Regression that fails to exhaust socket data

[Guillem Jover]

Hi!

On Sat, 2019-05-04 at 13:02:13 +0100, Alessandro Ghedini wrote:
> On Sat, Apr 20, 2019 at 01:39:36PM +0200, Guillem Jover wrote:
> > Source: curl
> > Source-Version: 7.64.0-2
> > Severity: serious
> > Control: affects -1 rtorrent

> > I've started noticing rtorrent busy-looping at some points after
> > finishing a torrent. stracing and gdb'ing the process it was doing
> > that in its main loop, spamming on gettimeofday() and epoll_wait().
> > 
> > Checking the rtorrent dependencies I've not seen much suspicious
> > updated recently, except for curl. I checked then the upstream bug
> > trackers and noticed quite many instances of "100% cpu usage" reports,
> > such as . And that
> > one points to old and new curl issue. The last instance of it appears
> > to have been fixed with 
> > in 7.64.1.
> > 
> > So I think curl should be either upgraded to that release, or the
> > relevant commits cherry-picked.
> 
> I prepared an update for the curl package with the two upstream patches 
> applied
> (38d8e1b and 4015fae), and uploaded the result for amd64 at:
> https://people.debian.org/~ghedo/libcurl4_7.64.0-3_amd64.deb
> 
> Guillem, could you try this version of the package and see if it fixes the
> issue for you? Once that's confirmed I'll upload to sid and ask for unblock.

Thanks! I tried yesterday and today to get a deterministic reproducer
with the version from unstable, but I'm not sure what's the trigger,
perhaps the torrent peers or the chunk arrival order or something. In
any case I was able to get the problem while redownloading some torrents
I had around. Installed the new package, and redownloaded several old
torrents and several new ones, and none have shown the problem since.

But even though I was getting that pretty often before, that does not
mean I might have been lucky until now! So, while it does look like the
issue is gone, I cannot say that with 100% assurance. :)

Will keep checking and report again in a day or two.

Thanks,
Guillem

Bug#927471: curl: Regression that fails to exhaust socket data

[Guillem Jover]

Hi!

On Tue, 2019-05-07 at 14:57:53 +0200, Guillem Jover wrote:
> Thanks! I tried yesterday and today to get a deterministic reproducer
> with the version from unstable, but I'm not sure what's the trigger,
> perhaps the torrent peers or the chunk arrival order or something. In
> any case I was able to get the problem while redownloading some torrents
> I had around. Installed the new package, and redownloaded several old
> torrents and several new ones, and none have shown the problem since.
> 
> But even though I was getting that pretty often before, that does not
> mean I might have been lucky until now! So, while it does look like the
> issue is gone, I cannot say that with 100% assurance. :)
> 
> Will keep checking and report again in a day or two.

Ok, been using it for a while, and I've not seen the busy-looping
since, so it LGTM!

Thanks,
Guillem