[gentoo-dev] systemd + postgresql is non-obvious to me

[gentoo-dev] systemd + postgresql is non-obvious to me

[Alexander Berntsen]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA256

# emerge postgresql-server
  ...

# emerge --config dev-db/postgresql-server:9.3
  ...

# systemctl start postgresql-server-9.3
  Job for postgresql-9.3.service failed. See 'systemctl status
  postgresql-9.3.service' and 'journalctl -xn' for details.

# systemctl status postgresql-9.3.service
  ● postgresql-9.3.service - PostgreSQL database server
  ...
  Process: 24145 ExecStart=/usr/lib/postgresql-9.3/bin/pg_ctl start -D
${PGDATA} -s -l ${DATA_DIR}/postmaster.log -o -p ${PGPORT}
- --data-directory=${DATA_DIR} -w -t 300 (code=exited, status=1/FAILURE)
  ...
  Jul 15 13:17:23 hackintosh pg_ctl[24145]: pg_ctl: could not start
  server
  Jul 15 13:17:23 hackintosh pg_ctl[24145]: Examine the log output.
  Jul 15 13:17:23 hackintosh systemd[1]: postgresql-9.3.service:
  control process exited, code=exited status=1
  Jul 15 13:17:23 hackintosh systemd[1]: Failed to start PostgreSQL
  database server.
  Jul 15 13:17:23 hackintosh systemd[1]: Unit postgresql-9.3.service
  entered failed state.


# cat /var/lib/postgresql/9.3/data/postmaster.log
  FATAL:  could not create lock file "/run/postgresql
  /.s.PGSQL.5432.lock": No such file or directory


***

To alleviate this I needed to run "systemd-tmpfiles --create". This
was non-obvious to me. Sounds like a packaging issue that I need to do
it in the first place?
- -- 
Alexander
bernalex@gentoo.org
https://secure.plaimi.net/~alexander
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v2.0.22 (GNU/Linux)
Comment: Using GnuPG with Thunderbird - http://www.enigmail.net/

iF4EAREIAAYFAlPFEQcACgkQRtClrXBQc7U73QD/bsI/9hp9XhNT9rBuNwiZUVu/
u4YbjfuN5KztqwM8XR4BAJZvsdAz9AxkpIJoOgeOfCKTEytTv27RmiGsSZnLND96
=9lQr
-----END PGP SIGNATURE-----

Re: [gentoo-dev] systemd + postgresql is non-obvious to me

[Pacho Ramos]

El mar, 15-07-2014 a las 13:31 +0200, Alexander Berntsen escribió:
[...]
> To alleviate this I needed to run "systemd-tmpfiles --create". This
> was non-obvious to me. Sounds like a packaging issue that I need to do
> it in the first place?

It's:
https://bugs.gentoo.org/show_bug.cgi?id=462118

The problem is that it's not clear to us how to make it automatically
without needing to call it manually from every ebuild installing a
tmpfiles.d file :(

Re: [gentoo-dev] systemd + postgresql is non-obvious to me

[Pacho Ramos]

El mar, 15-07-2014 a las 07:40 -0400, Rich Freeman escribió:
> On Tue, Jul 15, 2014 at 7:36 AM, Pacho Ramos <pacho@gentoo.org> wrote:
> > El mar, 15-07-2014 a las 13:31 +0200, Alexander Berntsen escribió:
> > [...]
> >> To alleviate this I needed to run "systemd-tmpfiles --create". This
> >> was non-obvious to me. Sounds like a packaging issue that I need to do
> >> it in the first place?
> >
> > It's:
> > https://bugs.gentoo.org/show_bug.cgi?id=462118
> >
> > The problem is that it's not clear to us how to make it automatically
> > without needing to call it manually from every ebuild installing a
> > tmpfiles.d file :(
> >
> 
> Isn't this a bit like automatically getting all open shells to source
> /etc/profile?  (Obviously that is even harder to do, so we don't
> bother trying.)  For that we're satisfied with a notice to the user.
> 
> Rich

It would be like:
https://bugs.gentoo.org/show_bug.cgi?id=487504

that could be used by icon/mime caches and similar too ;)

Re: [gentoo-dev] systemd + postgresql is non-obvious to me

[Alexander Berntsen]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA256

On 15/07/14 13:36, Pacho Ramos wrote:
> It's: https://bugs.gentoo.org/show_bug.cgi?id=462118
Thank you for the info.

> The problem is that it's not clear to us how to make it 
> automatically without needing to call it manually from every
> ebuild installing a tmpfiles.d file :(
Maybe consider adding ebuild output for ebuilds where this step is
somewhat critical, like for PostgreSQL.
- -- 
Alexander
bernalex@gentoo.org
https://secure.plaimi.net/~alexander
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v2.0.22 (GNU/Linux)
Comment: Using GnuPG with Thunderbird - http://www.enigmail.net/

iF0EAREIAAYFAlPFI8UACgkQRtClrXBQc7XRwQD8D+Who0CD2pavniG/D67d/w3j
VmBmhwVjMTMDJIk9At4A+MNTozzlGDR97dsCZGs9Sq5ha4IjzBBTEJg8P5prFes=
=fyIP
-----END PGP SIGNATURE-----

Re: [gentoo-dev] systemd + postgresql is non-obvious to me

[Ian Stakenvicius]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA256

On 15/07/14 07:36 AM, Pacho Ramos wrote:
> El mar, 15-07-2014 a las 13:31 +0200, Alexander Berntsen escribió: 
> [...]
>> To alleviate this I needed to run "systemd-tmpfiles --create".
>> This was non-obvious to me. Sounds like a packaging issue that I
>> need to do it in the first place?
> 
> It's: https://bugs.gentoo.org/show_bug.cgi?id=462118
> 
> The problem is that it's not clear to us how to make it
> automatically without needing to call it manually from every ebuild
> installing a tmpfiles.d file :(
> 
> 

Wasn't there a plan to make an eclass helper to process tmpfiles.d
files that get installed, during pkg_postinst ?

...and now that I think about it, did I say that I was going to write it?
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v2.0.22 (GNU/Linux)

iF4EAREIAAYFAlPFJCIACgkQ2ugaI38ACPDQZQD9HpwM9B/UfJBGY9e0bH4FQkXI
v558HKt9u39E2TTWUioBAJpi8xjnvOpCYIufe5YUvId46D0npaUuycn3RgQPEJlS
=drKc
-----END PGP SIGNATURE-----

Re: [gentoo-dev] systemd + postgresql is non-obvious to me

[Aaron W. Swenson]

[-- Attachment #1: Type: text/plain, Size: 1744 bytes --]

On 2014-07-15 08:52, Ian Stakenvicius wrote:
> On 15/07/14 07:36 AM, Pacho Ramos wrote:
> > El mar, 15-07-2014 a las 13:31 +0200, Alexander Berntsen escribió: 
> > [...]
> >> To alleviate this I needed to run "systemd-tmpfiles --create".
> >> This was non-obvious to me. Sounds like a packaging issue that I
> >> need to do it in the first place?
> > 
> > It's: https://bugs.gentoo.org/show_bug.cgi?id=462118
> > 
> > The problem is that it's not clear to us how to make it
> > automatically without needing to call it manually from every ebuild
> > installing a tmpfiles.d file :(
> > 
> > 
> 
> Wasn't there a plan to make an eclass helper to process tmpfiles.d
> files that get installed, during pkg_postinst ?
> 
> ...and now that I think about it, did I say that I was going to write it?
> 

There's a bit more to it, as well.

On the whole, I'm displeased with the systemd alternative for
controlling PostgreSQL. It's significantly hampered and doesn't allow
as much flexibility as the initscript. The major issue being trying to
nicely shut down the server instead of jumping straight to murder.

I really don't want to get into an OpenRC versus systemd flame-war,
but I really hate the systemd offering we have for PostgreSQL. It's
dumb and I recommend against using it.

I haven't used systemd and do not plan on using it until it actually
fixes a problem that's non-trivial, which is most likely never for me.

So, if any of you would like to improve our systemd offering, it'd be
welcomed enthusiastically.

-- 
Mr. Aaron W. Swenson
Gentoo Linux Developer
PostgreSQL Herd Bull
Email : titanofold@gentoo.org
GnuPG FP : 2C00 7719 4F85 FB07 A49C 0E31 5713 AA03 D1BB FDA0
GnuPG ID : D1BBFDA0

[-- Attachment #2: Digital signature --]
[-- Type: application/pgp-signature, Size: 213 bytes --]

Re: [gentoo-dev] systemd + postgresql is non-obvious to me

[Rich Freeman]

On Fri, Aug 22, 2014 at 1:23 PM, Aaron W. Swenson <titanofold@gentoo.org> wrote:
> On the whole, I'm displeased with the systemd alternative for
> controlling PostgreSQL. It's significantly hampered and doesn't allow
> as much flexibility as the initscript. The major issue being trying to
> nicely shut down the server instead of jumping straight to murder.
>

It looks like the package installs a service file provided by
upstream, so you might want to direct your complaint there unless it
really is a systemd limitation.

Checking the latest version in portage it calls:
ExecStop=/usr/@LIBDIR@/postgresql-@SLOT@/bin/pg_ctl stop -D
${DATA_DIR} -s -m fast

I'm no postresql expert, but according to the manpage that should
bring the server to a screeching, but still controlled, halt.  Perhaps
you would prefer setting it to -m smart instead of -m fast.

If so override it in /etc/systemd/system.  I generally recommend using
drop-ins for this, but I'm not sure if doing a drop-in for ExecStop
will override the existing value, or simply cause systemd to run both
commands (which means that whichever runs first will control how it
stops).

Systemd shouldn't begin killing processes without mercy until the
process invoked in ExecStop terminates, so it should not terminate
until the process has finished graceful shutdown.

Rich

Re: [gentoo-dev] systemd + postgresql is non-obvious to me

[Canek Peláez Valdés]

On Fri, Aug 22, 2014 at 1:07 PM, Rich Freeman <rich0@gentoo.org> wrote:
> On Fri, Aug 22, 2014 at 1:23 PM, Aaron W. Swenson <titanofold@gentoo.org> wrote:
>> On the whole, I'm displeased with the systemd alternative for
>> controlling PostgreSQL. It's significantly hampered and doesn't allow
>> as much flexibility as the initscript. The major issue being trying to
>> nicely shut down the server instead of jumping straight to murder.
>>
>
> It looks like the package installs a service file provided by
> upstream, so you might want to direct your complaint there unless it
> really is a systemd limitation.
>
> Checking the latest version in portage it calls:
> ExecStop=/usr/@LIBDIR@/postgresql-@SLOT@/bin/pg_ctl stop -D
> ${DATA_DIR} -s -m fast
>
> I'm no postresql expert, but according to the manpage that should
> bring the server to a screeching, but still controlled, halt.  Perhaps
> you would prefer setting it to -m smart instead of -m fast.
>
> If so override it in /etc/systemd/system.  I generally recommend using
> drop-ins for this, but I'm not sure if doing a drop-in for ExecStop
> will override the existing value, or simply cause systemd to run both
> commands (which means that whichever runs first will control how it
> stops).

It's the same as with ExecStart=; it gets added to a list, and all of
them are executed in the same order as they appear in the unit file.
In a drop-in, you can reset the list like this:

ExecStop=
ExecStop=new_and_only_command_to_be_executed

> Systemd shouldn't begin killing processes without mercy until the
> process invoked in ExecStop terminates, so it should not terminate
> until the process has finished graceful shutdown.

If ExecStop= wasn't specified, systemd would immediaely kill all the
service processes; from [1]: "If this option [ExecStop] is not
specified, the process is terminated immediately when service stop is
requested", which kinda sounds like "jumping straight to murder."

That the upstream unit file is configured otherwise, shows that
PostgreSQL upstream itself thinks that's a bad idea. No fault on
systemd's part.

Regards.

[1] http://www.freedesktop.org/software/systemd/man/systemd.service.html
-- 
Canek Peláez Valdés
Profesor de asignatura, Facultad de Ciencias
Universidad Nacional Autónoma de México

Re: [gentoo-dev] systemd + postgresql is non-obvious to me

[Mike Gilbert]

On Fri, Aug 22, 2014 at 2:07 PM, Rich Freeman <rich0@gentoo.org> wrote:
> On Fri, Aug 22, 2014 at 1:23 PM, Aaron W. Swenson <titanofold@gentoo.org> wrote:
>> On the whole, I'm displeased with the systemd alternative for
>> controlling PostgreSQL. It's significantly hampered and doesn't allow
>> as much flexibility as the initscript. The major issue being trying to
>> nicely shut down the server instead of jumping straight to murder.
>>
>
> It looks like the package installs a service file provided by
> upstream, so you might want to direct your complaint there unless it
> really is a systemd limitation.
>
> Checking the latest version in portage it calls:
> ExecStop=/usr/@LIBDIR@/postgresql-@SLOT@/bin/pg_ctl stop -D
> ${DATA_DIR} -s -m fast
>
> I'm no postresql expert, but according to the manpage that should
> bring the server to a screeching, but still controlled, halt.  Perhaps
> you would prefer setting it to -m smart instead of -m fast.
>

Personally, I prefer the thing to stop within a reasonable time period
when I tell it to stop. Waiting around for the remaining connections
to magically die (smart mode) is just tedious.

Re: [gentoo-dev] systemd + postgresql is non-obvious to me

[Aaron W. Swenson]

[-- Attachment #1: Type: text/plain, Size: 989 bytes --]

On 2014-08-22 14:07, Rich Freeman wrote:
> On Fri, Aug 22, 2014 at 1:23 PM, Aaron W. Swenson <titanofold@gentoo.org> wrote:
> > On the whole, I'm displeased with the systemd alternative for
> > controlling PostgreSQL. It's significantly hampered and doesn't allow
> > as much flexibility as the initscript. The major issue being trying to
> > nicely shut down the server instead of jumping straight to murder.
> >
> 
> It looks like the package installs a service file provided by
> upstream, so you might want to direct your complaint there unless it
> really is a systemd limitation.

Delayed response, I've been looking into this as I've been working on
unifying the ebuilds, but I can't find where upstream is providing the
systemd unit. The only one I have is the one we've made.

-- 
Mr. Aaron W. Swenson
Gentoo Linux Developer
PostgreSQL Herd Bull
Email : titanofold@gentoo.org
GnuPG FP : 2C00 7719 4F85 FB07 A49C 0E31 5713 AA03 D1BB FDA0
GnuPG ID : D1BBFDA0

[-- Attachment #2: Digital signature --]
[-- Type: application/pgp-signature, Size: 213 bytes --]

Re: [gentoo-dev] systemd + postgresql is non-obvious to me

[Canek Peláez Valdés]

On Mon, Sep 1, 2014 at 10:46 AM, Aaron W. Swenson <titanofold@gentoo.org> wrote:
> On 2014-08-22 14:07, Rich Freeman wrote:
>> On Fri, Aug 22, 2014 at 1:23 PM, Aaron W. Swenson <titanofold@gentoo.org> wrote:
>> > On the whole, I'm displeased with the systemd alternative for
>> > controlling PostgreSQL. It's significantly hampered and doesn't allow
>> > as much flexibility as the initscript. The major issue being trying to
>> > nicely shut down the server instead of jumping straight to murder.
>> >
>>
>> It looks like the package installs a service file provided by
>> upstream, so you might want to direct your complaint there unless it
>> really is a systemd limitation.
>
> Delayed response, I've been looking into this as I've been working on
> unifying the ebuilds, but I can't find where upstream is providing the
> systemd unit. The only one I have is the one we've made.

It is included in:

http://dev.gentoo.org/~titanofold/postgresql-initscript-2.6.tbz2

It doesn't says who the author is; but he or she was the one deciding
to wait five minutes (TimeoutSec=300) for the server to stop (and also
to disable the OOM killer: OOMScoreAdjust=-1000).

Regards.
-- 
Canek Peláez Valdés
Profesor de asignatura, Facultad de Ciencias
Universidad Nacional Autónoma de México

Re: [gentoo-dev] systemd + postgresql is non-obvious to me

[Aaron W. Swenson]

[-- Attachment #1: Type: text/plain, Size: 1818 bytes --]

On 2014-09-01 11:09, Canek Peláez Valdés wrote:
> On Mon, Sep 1, 2014 at 10:46 AM, Aaron W. Swenson <titanofold@gentoo.org> wrote:
> > On 2014-08-22 14:07, Rich Freeman wrote:
> >> On Fri, Aug 22, 2014 at 1:23 PM, Aaron W. Swenson <titanofold@gentoo.org> wrote:
> >> > On the whole, I'm displeased with the systemd alternative for
> >> > controlling PostgreSQL. It's significantly hampered and doesn't allow
> >> > as much flexibility as the initscript. The major issue being trying to
> >> > nicely shut down the server instead of jumping straight to murder.
> >> >
> >>
> >> It looks like the package installs a service file provided by
> >> upstream, so you might want to direct your complaint there unless it
> >> really is a systemd limitation.
> >
> > Delayed response, I've been looking into this as I've been working on
> > unifying the ebuilds, but I can't find where upstream is providing the
> > systemd unit. The only one I have is the one we've made.
> 
> It is included in:
> 
> http://dev.gentoo.org/~titanofold/postgresql-initscript-2.6.tbz2
> 
> It doesn't says who the author is; but he or she was the one deciding
> to wait five minutes (TimeoutSec=300) for the server to stop (and also
> to disable the OOM killer: OOMScoreAdjust=-1000).

Well, I'm the one who committed it to the pgsql-patches repo and
Patrick committed it to gentoo-x86, but neither of us wrote it. There was
some collaboration on the bug [1].

You can also read through the comments to see my other complaints. I
don't know if they're addressed or not.

[1] https://bugs.gentoo.org/show_bug.cgi?id=468868

-- 
Mr. Aaron W. Swenson
Gentoo Linux Developer
PostgreSQL Herd Bull
Email : titanofold@gentoo.org
GnuPG FP : 2C00 7719 4F85 FB07 A49C 0E31 5713 AA03 D1BB FDA0
GnuPG ID : D1BBFDA0

[-- Attachment #2: Digital signature --]
[-- Type: application/pgp-signature, Size: 213 bytes --]

Re: [gentoo-dev] systemd + postgresql is non-obvious to me

[Rich Freeman]

On Mon, Sep 1, 2014 at 11:46 AM, Aaron W. Swenson <titanofold@gentoo.org> wrote:
> On 2014-08-22 14:07, Rich Freeman wrote:
>> On Fri, Aug 22, 2014 at 1:23 PM, Aaron W. Swenson <titanofold@gentoo.org> wrote:
>> > On the whole, I'm displeased with the systemd alternative for
>> > controlling PostgreSQL. It's significantly hampered and doesn't allow
>> > as much flexibility as the initscript. The major issue being trying to
>> > nicely shut down the server instead of jumping straight to murder.
>> >
>>
>> It looks like the package installs a service file provided by
>> upstream, so you might want to direct your complaint there unless it
>> really is a systemd limitation.
>
> Delayed response, I've been looking into this as I've been working on
> unifying the ebuilds, but I can't find where upstream is providing the
> systemd unit. The only one I have is the one we've made.

My bad - I didn't spot it in the SRC_URI and had just assumed it came
from upstream...

Rich