Re: Problems with recent versions of s6-rc

[Casper Ti. Vector]

I just found myself being extremely careless this time: when I said "the
alpine testing repository", I really meant `edge/testing', but s6-rc
v0.2.1.1 should be in `edge/main', so probably I just installed s6-rc
v0.2.0.1 from `v3.6/main', and rediscovered the original problems with
the old alpine package.  The v0.2.1.1 package seems to work fine now.
I am very sorry for the disturbance :(

On Tue, Jul 04, 2017 at 11:59:01PM +, Laurent Bercot wrote:
>   I cannot reproduce the issue

-- 
My current OpenPGP key:
RSA4096/0x227E8CAAB7AA186C (expires: 2020.10.19)
7077 7781 B859 5166 AE07 0286 227E 8CAA B7AA 186C

Re: Problems with recent versions of s6-rc

[Casper Ti. Vector]

The s6-rc-compile segfault and the s6-rc failure still occur with `s6-rc
0.2.1.1-r0' from the alpine testing repository.

-- 
My current OpenPGP key:
RSA4096/0x227E8CAAB7AA186C (expires: 2020.10.19)
7077 7781 B859 5166 AE07 0286 227E 8CAA B7AA 186C

Re: Problems with recent versions of s6-rc

[Casper Ti. Vector]

Thanks.  Seems I need to pin the s6-related packages at alpine 3.5.x on
my servers for a longer time...

On Sun, Jun 04, 2017 at 11:21:30AM +, Laurent Bercot wrote:
>   We have confirmed that it is an Alpine problem. We're not sure
> exactly where the problem occurs (the usual suspect, grsec, has been
> innocented), but we know it's related to the Alpine package build
> and to virtualization.
>   In any case, the s6-rc code is good. You can confirm it by building
> it yourself.

-- 
My current OpenPGP key:
RSA4096/0x227E8CAAB7AA186C (expires: 2020.10.19)
7077 7781 B859 5166 AE07 0286 227E 8CAA B7AA 186C

Re: Problems with recent versions of s6-rc

[Laurent Bercot]

Hello, is there any progress on this issue?


 We have confirmed that it is an Alpine problem. We're not sure
exactly where the problem occurs (the usual suspect, grsec, has been
innocented), but we know it's related to the Alpine package build
and to virtualization.
 In any case, the s6-rc code is good. You can confirm it by building
it yourself.

--
 Laurent

Re: Problems with recent versions of s6-rc

[Casper Ti. Vector]

Hello, is there any progress on this issue?

On Fri, May 26, 2017 at 04:00:06AM +, Laurent Bercot wrote:
>   Tomorrow I'll report that to the Alpine devs, and investigate what
> could have gone wrong. And I'll have to either pay more attention to
> Alpine quality control, or drop maintenance of the Alpine packages, because
> I lost half a night on this and I can't afford to waste time on f*cking
> distro bugs.

-- 
My current OpenPGP key:
RSA4096/0x227E8CAAB7AA186C (expires: 2020.10.19)
7077 7781 B859 5166 AE07 0286 227E 8CAA B7AA 186C

Re: Problems with recent versions of s6-rc

[Laurent Bercot]

Not even in a VirtualBox VM?  Seems more interesting than I thought...


 I hadn't tried... I cannot reproduce when compiling everything by hand
on my machines, but I followed the procedure you outlined and I am 
indeed

able to reproduce on a Virtualbox VM running the Alpine packages.

 The fact that the outcome of s6-rc-compile is random (segfault,
busyloop, success, or even random failure) seems to point to a memory
corruption, but I can reproducibly successfully compile the same source
database on other machines.

 Add that the s6-rc failure you reported is caused by s6-ftrigrd getting
a SIGILL, which normally only happens on miscompiled programs, and I'm
starting to look extremely suspiciously at the Alpine packages.

 Tomorrow I'll report that to the Alpine devs, and investigate what
could have gone wrong. And I'll have to either pay more attention to
Alpine quality control, or drop maintenance of the Alpine packages, 
because

I lost half a night on this and I can't afford to waste time on f*cking
distro bugs.

--
 Laurent

Re: Problems with recent versions of s6-rc

[Casper Ti. Vector]

On Fri, May 26, 2017 at 12:15:39AM +, Laurent Bercot wrote:
>   When it's about debug information such as strace outputs
 ;)

>   Except that when you send a file to a mailing-list
Well this is the problem I neglected: it is not Gentoo or not, but bug
trackers have their own persistent storages that are not duplicated by
mails.  I will post debug info to pastebins then.

-- 
My current OpenPGP key:
RSA4096/0x227E8CAAB7AA186C (expires: 2020.10.19)
7077 7781 B859 5166 AE07 0286 227E 8CAA B7AA 186C

Re: Problems with recent versions of s6-rc

[Laurent Bercot]

Well I think pastebins really tend to be unreliable (most importantly,
content disappearing at unexpected time), so small files (whether text
or binary, as long as they are relevant) are best posted directly to 
the

mail list


 No. When it's about debug information such as strace outputs, the
information is transient - it loses importance once the bug is fixed.
So it doesn't matter that a pastebin doesn't last forever.



 and since compression saves space, tarballs and zipballs are
suitable for files longer than a few dozens of lines.  This practice is
recommended on some FOSS bug trackers, eg. .


 Except that when you send a file to a mailing-list, the file is
duplicated to every recipient, in a "push" way, i.e. recipients have
no control over it. When you publish it on a site, be it a pastebin
or any other regular Web site, it's only duplicated in a "pull" way,
by clients who are actively interested in the contents of the file.
Publication should be done in a "pull" way as much as possible; a
mailing-list only works because all subscribers are generally interested
in the mailing-list contents, and I doubt all subscribers are interested
in verbose s6-rc strace outputs.

 I don't care what other people recommend. (Especially Gentoo.)
If I did what other people recommend instead of thinking for myself,
skarnet.org wouldn't exist. Or suck just as much as everything else.

--
 Laurent

Re: Problems with recent versions of s6-rc

[Casper Ti. Vector]

Well I think pastebins really tend to be unreliable (most importantly,
content disappearing at unexpected time), so small files (whether text
or binary, as long as they are relevant) are best posted directly to the
mail list; and since compression saves space, tarballs and zipballs are
suitable for files longer than a few dozens of lines.  This practice is
recommended on some FOSS bug trackers, eg. .

On Thu, May 25, 2017 at 03:19:51PM +, Laurent Bercot wrote:
>   It is intentional that binary attachments are stripped when sent to the
> list. I don't want to spam every subscriber with binaries. When I say
> "please send a strace", I really mean "please paste a strace to some
> website such as pastebin, and send a link".
>   Please don't try and circumvent the attachment stripping, or I'll
> have to tighten it. :P

-- 
My current OpenPGP key:
RSA4096/0x227E8CAAB7AA186C (expires: 2020.10.19)
7077 7781 B859 5166 AE07 0286 227E 8CAA B7AA 186C

Re: Problems with recent versions of s6-rc

[Laurent Bercot]

A person on this list kindly told that no attachment appeared; but
actually I *did* attach a gzip'ed version of the log as 
`rc-err.log.gz',
so we are perhaps facing a problem with some steps in the mail 
delivery.


 It is intentional that binary attachments are stripped when sent to the
list. I don't want to spam every subscriber with binaries. When I say
"please send a strace", I really mean "please paste a strace to some
website such as pastebin, and send a link".

 Please don't try and circumvent the attachment stripping, or I'll
have to tighten it. :P
 (The attachment made it this time, I'll study it. Thanks!)

--
 Laurent

Re: Problems with recent versions of s6-rc

[Casper Ti. Vector]

A person on this list kindly told that no attachment appeared; but
actually I *did* attach a gzip'ed version of the log as `rc-err.log.gz',
so we are perhaps facing a problem with some steps in the mail delivery.
Attached is a tgz version; please do tell me if it disappears again.

On Thu, May 25, 2017 at 09:19:26PM +0800, Casper Ti. Vector wrote:
> attached is the log.

-- 
My current OpenPGP key:
RSA4096/0x227E8CAAB7AA186C (expires: 2020.10.19)
7077 7781 B859 5166 AE07 0286 227E 8CAA B7AA 186C



rc-log.tgz
Description: GNU Unix tar archive

Re: Problems with recent versions of s6-rc

[Casper Ti. Vector]

After reboot-looping a test VM for many times, this is also reproduced
successfully; attached is the log.

On Thu, May 25, 2017 at 09:07:26PM +0800, Casper Ti. Vector wrote:
> The issue with `s6-rc' seems to be difficult to reproduce when `s6-rc'
> is debugged by strace, so I am unable to produce a log for it now.

-- 
My current OpenPGP key:
RSA4096/0x227E8CAAB7AA186C (expires: 2020.10.19)
7077 7781 B859 5166 AE07 0286 227E 8CAA B7AA 186C

Re: Problems with recent versions of s6-rc

[Casper Ti. Vector]

Not even in a VirtualBox VM?  Seems more interesting than I thought...
Attached is the logs of `s6-rc-compile' for the three cases.  The issue
with `s6-rc' seems to be difficult to reproduce when `s6-rc' is debugged
by strace, so I am unable to produce a log for it now.

On Thu, May 25, 2017 at 11:43:45AM +, Laurent Bercot wrote:
>   Can you then please send me a strace of the failing programs? I can't
> reproduce.

-- 
My current OpenPGP key:
RSA4096/0x227E8CAAB7AA186C (expires: 2020.10.19)
7077 7781 B859 5166 AE07 0286 227E 8CAA B7AA 186C



compile-log.tgz
Description: GNU Unix tar archive

Re: Problems with recent versions of s6-rc

[Laurent Bercot]

I experienced this when updating Alpine from v3.5.2 to v3.6.0 on these
machines; the v3.6.0 release have already covered skalibs v2.5.1.1.


 Can you then please send me a strace of the failing programs? I can't
reproduce.

--
 Laurent

Re: Problems with recent versions of s6-rc

[Casper Ti. Vector]

On Thu, May 25, 2017 at 10:17:22AM +, Laurent Bercot wrote:
>   All the problems you report should be fixed with skalibs-2.5.1.1, which
> should already be available in Alpine. Are you still experiencing issues
> with skalibs-2.5.1.1 ?

I experienced this when updating Alpine from v3.5.2 to v3.6.0 on these
machines; the v3.6.0 release have already covered skalibs v2.5.1.1.

>   It is expected to recompile the database on major upgrades.
> 0.1.0.0 to 0.2.0.0 is a major upgrade (else the release would have been
> called 0.1.1.0). Sorry if that was not clear; I will make it
> explicit when it happens again.

I see; thanks then.

-- 
My current OpenPGP key:
RSA4096/0x227E8CAAB7AA186C (expires: 2020.10.19)
7077 7781 B859 5166 AE07 0286 227E 8CAA B7AA 186C

Re: Problems with recent versions of s6-rc

[Laurent Bercot]

 All the problems you report should be fixed with skalibs-2.5.1.1, which
should already be available in Alpine. Are you still experiencing issues
with skalibs-2.5.1.1 ?


* Incompatibility of the compiled database: s6-rc commit 543c1405 
breaks

 the backwards compatibility of the compiled service database, which is
 unfortunately not mentioned in `upgrade.html' (or is it expected to
 recompile the database whenever s6-rc is updated?).


 It is expected to recompile the database on major upgrades.
0.1.0.0 to 0.2.0.0 is a major upgrade (else the release would have been
called 0.1.1.0). Sorry if that was not clear; I will make it
explicit when it happens again.

--
 Laurent

Re: Problems with recent versions of s6-rc

[Casper Ti. Vector]

Additionally, on my machines, which now use freshly compiled databases,
when running `s6-rc -u change default' in `rc.init', with a small but
significant probability (one in a few) the procedure ends up with this
error:
> s6-svlisten1: fatal: unable to ftrigr_update: Broken pipe
> s6-rc: warning: unable to start start service s6rc-oneshot-runner: command 
> exited 111

On Thu, May 25, 2017 at 12:58:30PM +0800, Casper Ti. Vector wrote:
> * Incompatibility of the compiled database: s6-rc commit 543c1405 breaks
>   the backwards compatibility of the compiled service database, which is
>   unfortunately not mentioned in `upgrade.html' (or is it expected to
>   recompile the database whenever s6-rc is updated?).

-- 
My current OpenPGP key:
RSA4096/0x227E8CAAB7AA186C (expires: 2020.10.19)
7077 7781 B859 5166 AE07 0286 227E 8CAA B7AA 186C