Обсуждение: Add socket dir to pg_config..?
All,
Was just wondering if we might want to include the default socket directory that was compiled in as part of the
pg_configoutput..? Thanks,
Stephen
Stephen Frost <sfrost@snowman.net> writes:
> Was just wondering if we might want to include the default socket
> directory that was compiled in as part of the pg_config output..?
[ shrug... ] We don't report the compiled-in port number, which is
considerably more critical. And we don't report changes in any of the
other stuff in pg_config_manual.h.
MHO is that changing the socket directory is only marginally supported,
and we shouldn't encourage it unless we're prepared to fully support it
(which we can't really).
regards, tom lane
2011/10/28 Tom Lane <tgl@sss.pgh.pa.us>:
> Stephen Frost <sfrost@snowman.net> writes:
>> Was just wondering if we might want to include the default socket
>> directory that was compiled in as part of the pg_config output..?
>
> [ shrug... ] We don't report the compiled-in port number, which is
> considerably more critical. And we don't report changes in any of the
> other stuff in pg_config_manual.h.
>
> MHO is that changing the socket directory is only marginally supported,
> and we shouldn't encourage it unless we're prepared to fully support it
> (which we can't really).
There is a TODO about that.
Allow simpler reporting of the unix domain socket directory and allow easier configuration of its default
location
* http://archives.postgresql.org/pgsql-hackers/2010-10/msg01555.php
Last time the subject came in, the result was that pg_config may
output it, but the solution that most people seems to agree at this
time was to add a configure option. Except that we didn't want to
encourage people to change the default_socket_dir, so a documentation
update in this direction was suggested to be done at the same time the
switch is added to PostgreSQL.
--
Cédric Villemain +33 (0)6 20 30 22 52
http://2ndQuadrant.fr/
PostgreSQL: Support 24x7 - Développement, Expertise et Formation
* Tom Lane (tgl@sss.pgh.pa.us) wrote: > [ shrug... ] We don't report the compiled-in port number, which is > considerably more critical. And we don't report changes in any of the > other stuff in pg_config_manual.h. True. > MHO is that changing the socket directory is only marginally supported, > and we shouldn't encourage it unless we're prepared to fully support it > (which we can't really). This concerns me a bit, as most distros change it.. What would you expect to break when the socket dir is changed from the default? Thanks, Stephen
On 10/28/2011 08:01 AM, Stephen Frost wrote: > > >> MHO is that changing the socket directory is only marginally supported, >> and we shouldn't encourage it unless we're prepared to fully support it >> (which we can't really). > This concerns me a bit, as most distros change it.. What would you > expect to break when the socket dir is changed from the default? > > Er, which distros other than debian/ubuntu? cheers andrew
Andrew Dunstan <andrew@dunslane.net> writes: > Er, which distros other than debian/ubuntu? Well, any and all derivatives I guess, to begin with. http://distrowatch.com/dwres.php?resource=independence#debian Based on Debian GNU/Linux: 129 Distributions More seriously, I'm not sure how to understand why some people will both frown upon distribution allowing themselves to patch the version of PostgreSQL they are packaging, and vote against making their life easier. If /tmp is the only decent place where to put the socket file on Unix when security and other concerns are considered, then sure, making distro life difficult is a good thing to do. But then let's take it to the FHS that debian and ubuntu are implementing, AFAIUI. I'm puzzled, maybe I'm not understanding a key point here though. Regards, -- Dimitri Fontaine http://2ndQuadrant.fr PostgreSQL : Expertise, Formation et Support
On Fri, Oct 28, 2011 at 5:33 PM, Dimitri Fontaine <dimitri@2ndquadrant.fr> wrote: > I'm puzzled, maybe I'm not understanding a key point here though. > I think the only thing you're missing here is that, despite appearances some days, Postgres is not run by a single hive mind. Tom thinks there needs to be a single location for all builds as it's part of the protocol, and others agree with the approach Debian took and we haven't really resolved that discrepancy. So Postgres's treatment of this configuration is naturally a bit schizophrenic. -- greg
On Fri, Oct 28, 2011 at 06:33:39PM +0200, Dimitri Fontaine wrote: > Andrew Dunstan <andrew@dunslane.net> writes: > > Er, which distros other than debian/ubuntu? > > Well, any and all derivatives I guess, to begin with. > > http://distrowatch.com/dwres.php?resource=independence#debian > Based on Debian GNU/Linux: 129 Distributions > > More seriously, I'm not sure how to understand why some people will both > frown upon distribution allowing themselves to patch the version of > PostgreSQL they are packaging, and vote against making their life > easier. > > If /tmp is the only decent place where to put the socket file on Unix > when security and other concerns are considered, then sure, making > distro life difficult is a good thing to do. But then let's take it to > the FHS that debian and ubuntu are implementing, AFAIUI. > > I'm puzzled, maybe I'm not understanding a key point here though. > > Regards, > -- > Dimitri Fontaine > http://2ndQuadrant.fr PostgreSQL : Expertise, Formation et Support In Gentoo, we change the socket directory to /var/run/postgresql via pg_config_manual.h. However, I'm not too terribly interested in pg_config outputting the directory location. We inform users at the end of every install where the default location is. Further, all of the packages we maintain build against the sources so the packages automatically know where the socket directory is located. -- Mr. Aaron W. Swenson Gentoo Linux Developer Email : titanofold@gentoo.org GnuPG FP : 2C00 7719 4F85 FB07 A49C 0E31 5713 AA03 D1BB FDA0 GnuPG ID : D1BBFDA0
On Sat, Oct 29, 2011 at 08:28:57PM +0000, Mr. Aaron W. Swenson wrote: > > If /tmp is the only decent place where to put the socket file on Unix > > when security and other concerns are considered, then sure, making > > distro life difficult is a good thing to do. But then let's take it to > > the FHS that debian and ubuntu are implementing, AFAIUI. > In Gentoo, we change the socket directory to /var/run/postgresql via > pg_config_manual.h. However, I'm not too terribly interested in pg_config > outputting the directory location. Frankly, I'm not seeing the difference between the socket directory and the "listen_addresses" option. When connecting you can specify the socket directory to use via the "host" option. It might even be more logical to be able to specify multiple directories. Given we support multiple listen sockets I can't imagine it would require much code. (And yes, just today I ran into the issue of hardcoded paths. If the directory it points to is not world writable then you've limited the users who can run the postgres server. Which is an unnecessary restriction imho). Have a nice day, -- Martijn van Oosterhout <kleptog@svana.org> http://svana.org/kleptog/ > He who writes carelessly confesses thereby at the very outset that he does > not attach much importance to his own thoughts. -- Arthur Schopenhauer
2011/10/30 Martijn van Oosterhout <kleptog@svana.org>: > On Sat, Oct 29, 2011 at 08:28:57PM +0000, Mr. Aaron W. Swenson wrote: >> > If /tmp is the only decent place where to put the socket file on Unix >> > when security and other concerns are considered, then sure, making >> > distro life difficult is a good thing to do. But then let's take it to >> > the FHS that debian and ubuntu are implementing, AFAIUI. > >> In Gentoo, we change the socket directory to /var/run/postgresql via >> pg_config_manual.h. However, I'm not too terribly interested in pg_config >> outputting the directory location. > > Frankly, I'm not seeing the difference between the socket directory and > the "listen_addresses" option. When connecting you can specify the > socket directory to use via the "host" option. > > It might even be more logical to be able to specify multiple > directories. Given we support multiple listen sockets I can't imagine > it would require much code. > > (And yes, just today I ran into the issue of hardcoded paths. If the > directory it points to is not world writable then you've limited the > users who can run the postgres server. Which is an unnecessary > restriction imho). > For Debian, the reason is : Description: Put server Unix sockets into /var/run/postgresql/ by default Forwarded: No, Debian specific configuration with postgresql-common Using /tmp for sockets allows everyone to spoof a PostgreSQL server. Thus use /var/run/postgresql/ for "system" clusters which run as 'postgres' (user clusters will still use /tmp). Since system cluster are by far the common case, set it as default. -- Cédric Villemain +33 (0)6 20 30 22 52 http://2ndQuadrant.fr/ PostgreSQL: Support 24x7 - Développement, Expertise et Formation