BUG #3948: date/time functions returning wrong value

Jorge Campins wrote:

The following bug has been logged online:

Bug reference: 3948
Logged by: Jorge Campins
Email address: jrcampins@cantv.net
PostgreSQL version: 8.1
Operating system: Windows XP
Description: date/time functions returning wrong value
Details:

As you might be aware, recently Venezuela has changed his timezone from GMT
-4 to GMT -4:30. Windows implemented a new timezone for us, "Caracas", to be
used instead of "Caracas/La Paz".

As soon as I changed my computer's timezone to "Caracas", PostgreSQL
date/time functions such as current_timestamp and localtimestamp started to
return a wrong value (I guess they don't recognize the new timezone so they
are returning GMT).

Then I change it to "La Paz", and once again they returned my computer's
time, but of course it has half an hour difference.

I realize this is not really a bug. We will really appreciate your effort to
solve this problem at your earliest convenience.

Try setting your timezone in postgresql.conf - that should override
whatever Windows has for you.

//Magnus

Magnus Hagander <magnus@hagander.net> writes:

Try setting your timezone in postgresql.conf - that should override
whatever Windows has for you.

Unless his "8.1" is really 8.1.something-pretty-darn-recent, it's
not going to know about the latest Venezuela DST law change anyway.

This does suggest that we'll need to revisit the win32_tzmap[] list
every so often?

regards, tom lane

How can I find which version I'm using? All I could see in pgAdmin is 8.1.

Which value should I use for timezone in postgresql.conf? I tried select *
from pg_timezone_names to get a list of valid time zone names but it failed
with "relation pg_timezone_names does not exist."

Thanks and regards,

Jorge

----- Original Message -----
From: "Tom Lane" <tgl@sss.pgh.pa.us>
To: "Magnus Hagander" <magnus@hagander.net>
Cc: "Jorge Campins" <jrcampins@cantv.net>; <pgsql-bugs@postgresql.org>
Sent: Saturday, February 09, 2008 8:25 PM
Subject: Re: [BUGS] BUG #3948: date/time functions returning wrong value

Tom Lane wrote:

Magnus Hagander <magnus@hagander.net> writes:

Try setting your timezone in postgresql.conf - that should override
whatever Windows has for you.

Unless his "8.1" is really 8.1.something-pretty-darn-recent, it's
not going to know about the latest Venezuela DST law change anyway.

Yes, good point.

Or really, since 8.1 is unsupported, to 8.2.most-recent-available if
possible.

This does suggest that we'll need to revisit the win32_tzmap[] list
every so often?

Seems so. It's the first time I've heard of a timezone being *added* and
not just changed, but obviously it does happen :-(

//Magnus

Jorge Campins wrote:

How can I find which version I'm using? All I could see in pgAdmin is 8.1.

SELECT version();

Which value should I use for timezone in postgresql.conf? I tried select
* from pg_timezone_names to get a list of valid time zone names but it
failed with "relation pg_timezone_names does not exist."

pg_timezone_names was added in 8.2. (which you really should upgrade to
anyway, since 8.1 isn't supported on Windows anymore)

To find it in 8.1, check the files in share/timezone in your PostgreSQL
installation.

//Magnus

I'll upgrade to 8.2 ASAP. Once in 8.2, which value should I use for timezone
in postgresql.conf?

Thanks and regards,

Jorge

----- Original Message -----
From: "Magnus Hagander" <magnus@hagander.net>
To: "Jorge Campins" <jrcampins@cantv.net>
Cc: "Tom Lane" <tgl@sss.pgh.pa.us>; <pgsql-bugs@postgresql.org>
Sent: Sunday, February 10, 2008 5:39 AM
Subject: Re: [BUGS] BUG #3948: date/time functions returning wrong value

In 8.2, use pg_timezone_names to find the proper one for you - it'll
exist once you've upgraded :-)

//Magnus

Jorge Campins wrote:

Magnus Hagander wrote:

Tom Lane wrote:

This does suggest that we'll need to revisit the win32_tzmap[] list
every so often?

Seems so. It's the first time I've heard of a timezone being *added* and
not just changed, but obviously it does happen :-(

Hmm, was this table manually built? I think I see a mistake. It has
two entries for "Mexico Standard Time", one of which (the one at GMT-6)
is mapped to America/Mexico_City (which I think would be correct), and
the other at America/La_Paz (the one at -7). This latter one I think is
mistaken -- firstly because the Windows name is probably something else
and not "Mexico Standard Time" (perhaps "Western Mexico Std Time" or
something like that?), and secondly because America/La_Paz refers to
Bolivia's La Paz and not Mexico's. Mexico's La Paz should probably be
mapped to America/Chihuahua or America/Mazatlan.

It's hard to tell though -- I am not sure how does Windows define
timezones. I have always been annoyed by the fact that Chilean timezone
is nowhere near it's database (I think the closest is Bogota, but it's
really bogus because it's not even in the same hemisphere). Fortunately
this means there's no bogus entry for Chile in this struct ...

I think what I conclude from this is that Windows TZ database is so
bogus that we should avoid trying to rely on it -- I say if the user
does not set "timezone" in postgresql.conf, refuse to start.

--
Alvaro Herrera http://www.CommandPrompt.com/
PostgreSQL Replication, Consulting, Custom Development, 24x7 support

Alvaro Herrera <alvherre@commandprompt.com> writes:

I think what I conclude from this is that Windows TZ database is so
bogus that we should avoid trying to rely on it -- I say if the user
does not set "timezone" in postgresql.conf, refuse to start.

Remember we're also relying on the OS for the time of day :-(.
I'm not sure there's any point in trying to be a lot better than
it is about timekeeping. In particular, even if the user has set
the displayed local time correctly, what Windows will tell us the
UTC time is depends entirely on its idea of the timezone offset.
If we have a different idea of the timezone offset, all it will buy
us is complaints from users that our time is wrong.

(Now, this pessimistic view of things might be obsolete if Windows
systems commonly get their UTC time from NTP, as is standard on
Linux and Mac these days. I hadn't heard that Microsoft was up
to speed on that, though --- if they did, I think it'd force them
to be a whole lot more careful keeping their local timezone knowledge
up to date ...)

regards, tom lane

On Sun, Feb 10, 2008 at 09:33:46PM -0500, Tom Lane wrote:

Alvaro Herrera <alvherre@commandprompt.com> writes:

I think what I conclude from this is that Windows TZ database is so
bogus that we should avoid trying to rely on it -- I say if the user
does not set "timezone" in postgresql.conf, refuse to start.

Remember we're also relying on the OS for the time of day :-(.
I'm not sure there's any point in trying to be a lot better than
it is about timekeeping. In particular, even if the user has set
the displayed local time correctly, what Windows will tell us the
UTC time is depends entirely on its idea of the timezone offset.
If we have a different idea of the timezone offset, all it will buy
us is complaints from users that our time is wrong.

(Now, this pessimistic view of things might be obsolete if Windows
systems commonly get their UTC time from NTP, as is standard on
Linux and Mac these days. I hadn't heard that Microsoft was up
to speed on that, though --- if they did, I think it'd force them
to be a whole lot more careful keeping their local timezone knowledge
up to date ...)

Since Windows 2000, all domain members in a Windows domain member machines
will synchronize their time with NTP. It's a requirement of Kerberos that
the clocks don't drift. Recommended best practice is to have your domain
controller sync to either an external source on the net or to a GPS.

Since Windows 2000, there has been an NTP client included in Windows.

Since Windows XP, I think (could be 2000), all non-domain machines will
ntp-synchronize with servers at microsoft by default.

Since Windows NT, it's been possible and recommended to synchronize all
members of a domain to the domain controllers with the "net time" tools
available.

//Magnus

On Sun, Feb 10, 2008 at 11:22:53PM -0300, Alvaro Herrera wrote:

Magnus Hagander wrote:

Tom Lane wrote:

This does suggest that we'll need to revisit the win32_tzmap[] list
every so often?

Seems so. It's the first time I've heard of a timezone being *added* and
not just changed, but obviously it does happen :-(

Hmm, was this table manually built? I think I see a mistake. It has

Yes. And it's obviously in need of updating.

two entries for "Mexico Standard Time", one of which (the one at GMT-6)
is mapped to America/Mexico_City (which I think would be correct), and
the other at America/La_Paz (the one at -7). This latter one I think is

Yes, that's clearly a mistake :-( It'll only ever use the first one
though...

What's in the database is one called "Mexico Stanadrd Time" and another one
called "Mexico Standard Time 2". We're missing the "2" there.

It's hard to tell though -- I am not sure how does Windows define
timezones. I have always been annoyed by the fact that Chilean timezone
is nowhere near it's database (I think the closest is Bogota, but it's
really bogus because it's not even in the same hemisphere). Fortunately
this means there's no bogus entry for Chile in this struct ...

My TZ database has an entry for Santiago... AFAIK, that's in Chile? -04?
Took me less than 30 seconds to find in the GUI for the Timezone settings.

The internal entry name is "Pacific SA Standad Time", which we map to
America/Santiago.

Is that actually bogus?

I think what I conclude from this is that Windows TZ database is so
bogus that we should avoid trying to rely on it -- I say if the user
does not set "timezone" in postgresql.conf, refuse to start.

While there are a lot of bogus things about the Windows TZ database, that's
not one of them. (the bogusness mostly deals with them not properly
tracking changes in DST rules over time - they are only interesetd in rules
that are in force *today*)

//Magnus

Dear All,

I installed 8.2.6.2 and solved my problem. I set Windows time zone to
"Caracas" and PostgreSQL "America/Caracas" and now they both give me the
right time.

WARNING: not only people in Venezuela but also people in Bolivia is affected
by this problem. As you might recall, Windows used to have a single time
zone for both countries, namely "Caracas/La Paz". If now you set Windows
time zone to "La Paz", PostgreSQL 8.2.6.2 will use "America/Caracas" unless
you specify something else in postgresql.conf. In this case, Windows has
Bolivia's time and PostgreSQL has Venezuela's time. So people in Bolivia
should also specify timezone in postgresql.conf.

Thank you all very much for your kind support.

Regards,

Jorge

----- Original Message -----
From: "Magnus Hagander" <magnus@hagander.net>
To: "Alvaro Herrera" <alvherre@commandprompt.com>
Cc: "Tom Lane" <tgl@sss.pgh.pa.us>; "Jorge Campins" <jrcampins@cantv.net>;
<pgsql-bugs@postgresql.org>
Sent: Monday, February 11, 2008 4:05 AM
Subject: Re: [BUGS] BUG #3948: date/time functions returning wrong value

Jorge Campins wrote:

Dear All,

I installed 8.2.6.2 and solved my problem. I set Windows time zone to
"Caracas" and PostgreSQL "America/Caracas" and now they both give me the
right time.

WARNING: not only people in Venezuela but also people in Bolivia is
affected by this problem. As you might recall, Windows used to have a
single time zone for both countries, namely "Caracas/La Paz". If now you
set Windows time zone to "La Paz", PostgreSQL 8.2.6.2 will use
"America/Caracas" unless you specify something else in postgresql.conf.
In this case, Windows has Bolivia's time and PostgreSQL has Venezuela's
time. So people in Bolivia should also specify timezone in postgresql.conf.

Yeah, I've got an update that has some more timezones around here
available coming up, I think that's going to help it. But always - the
pg database of timezones is a lot more extensive than the Windows one,
so it's always a good idea to check it out.

//Magnus