[Fwd: RE: Build farm on Windows]

Andrew Dunstan wrote:

Can one of the Windows buildfarm owners please try building and running
"make check" by hand rather than using the buildfarm script? It looks
like they all stopped reporting around the same time, and this might
give us a better clue about when things fall over.

Also, if you're up for it, please try reversing this patch, which looks
innocuous enough, but is the only thing I can see in the relevant time
period that looks at all suspicious:
http://archives.postgresql.org/pgsql-committers/2006-07/msg00256.php

will see what i can do(it definitly hangs in make check here too) - but
this issue seem to kill my box up to the point where it is impossible to
login(!) and i have to hard-reboot it.
Looks like it is churning CPU like mad when that happens ...

Stefan

Andrew Dunstan wrote:

Can one of the Windows buildfarm owners please try building and running
"make check" by hand rather than using the buildfarm script? It looks
like they all stopped reporting around the same time, and this might
give us a better clue about when things fall over.

Also, if you're up for it, please try reversing this patch, which looks
innocuous enough, but is the only thing I can see in the relevant time
period that looks at all suspicious:
http://archives.postgresql.org/pgsql-committers/2006-07/msg00256.php

looks like the postmaster fails to startup up:

./pg_regress --temp-install=./tmp_check --top-builddir=../../..
--temp-port=55678 --schedule=./parallel_schedule --multibyte=SQL_ASCII
--load-language=plpgsql
============== removing existing temp installation ==============
============== creating temporary installation ==============
============== initializing database system ==============
============== starting postmaster ==============

pg_regress: postmaster did not start within 60 seconds
Examine ./log/postmaster.log for the reason
make[2]: *** [check] Error 2
make[2]: Leaving directory
`/home/pgbuild/pgfarmbuild/HEAD/pgsql.1436/src/test/regress'
make[1]: *** [check] Error 2
make[1]: Leaving directory
`/home/pgbuild/pgfarmbuild/HEAD/pgsql.1436/src/test'
make: *** [check] Error 2

and the logfile is full of:

FATAL: failed to initialize timezone_abbreviations to "Default"
FATAL: failed to initialize timezone_abbreviations to "Default"
LOG: background writer process (PID 1568) exited with exit code 0
LOG: terminating any other active server processes
FATAL: failed to initialize timezone_abbreviations to "Default"
LOG: all server processes terminated; reinitializing
FATAL: failed to initialize timezone_abbreviations to "Default"
FATAL: failed to initialize timezone_abbreviations to "Default"
LOG: background writer process (PID 244) exited with exit code 0
LOG: terminating any other active server processes
LOG: all server processes terminated; reinitializing
FATAL: failed to initialize timezone_abbreviations to "Default"
FATAL: failed to initialize timezone_abbreviations to "Default"
LOG: background writer process (PID 468) exited with exit code 0
LOG: terminating any other active server processes

...

Stefan

Stefan Kaltenbrunner wrote:

Andrew Dunstan wrote:

Can one of the Windows buildfarm owners please try building and running
"make check" by hand rather than using the buildfarm script? It looks
like they all stopped reporting around the same time, and this might
give us a better clue about when things fall over.

Also, if you're up for it, please try reversing this patch, which looks
innocuous enough, but is the only thing I can see in the relevant time
period that looks at all suspicious:
http://archives.postgresql.org/pgsql-committers/2006-07/msg00256.php

will see what i can do(it definitly hangs in make check here too) - but
this issue seem to kill my box up to the point where it is impossible to
login(!) and i have to hard-reboot it.
Looks like it is churning CPU like mad when that happens ...

Does it get past the initdb stage? Past db startup? Past creating the
regression db? Run any tests and report results?

cheers

andrew

The TimeZone changes are looking might suspicious ...

cheers

andrew

Stefan Kaltenbrunner wrote:

Andrew Dunstan <andrew@dunslane.net> writes:

The TimeZone changes are looking might suspicious ...

FATAL: failed to initialize timezone_abbreviations to "Default"

Hm. It looks like this is working in the postmaster but failing
in subprocesses. I'll see if I can duplicate it using EXEC_BACKEND.

regards, tom lane

I wrote:

Andrew Dunstan <andrew@dunslane.net> writes:

The TimeZone changes are looking might suspicious ...

FATAL: failed to initialize timezone_abbreviations to "Default"

Hm. It looks like this is working in the postmaster but failing
in subprocesses. I'll see if I can duplicate it using EXEC_BACKEND.

Nope, works fine with EXEC_BACKEND, so it's something Windows-specific.
I'm not sure why you're not getting any more specific messages ---
they should be coming out at WARNING level AFAICS. You'll need to trace
through load_tzoffsets() and see why it's failing in the subprocess.

regards, tom lane

Tom Lane wrote:

I wrote:

Andrew Dunstan <andrew@dunslane.net> writes:

The TimeZone changes are looking might suspicious ...

FATAL: failed to initialize timezone_abbreviations to "Default"

Hm. It looks like this is working in the postmaster but failing
in subprocesses. I'll see if I can duplicate it using EXEC_BACKEND.

Nope, works fine with EXEC_BACKEND, so it's something Windows-specific.
I'm not sure why you're not getting any more specific messages ---
they should be coming out at WARNING level AFAICS. You'll need to trace
through load_tzoffsets() and see why it's failing in the subprocess.

that was a bit painful but we failed to see a useful error message due
to the fact that we have been activly suppressing it - with a quick hack
like:

---
/home/pgbuild/pgfarmbuild/HEAD/pgsql/src/backend/utils/misc/tzparser.c
    Tue Jul 25 05:51:21 2006
+++ src/backend/utils/misc/tzparser.c   Fri Jul 28 19:33:24 2006
@@ -326,7 +326,6 @@
        if (!tzFile)
        {
                /* at level 0, if file doesn't exist, guc.c's complaint
is enough */
-               if (errno != ENOENT || depth > 0)
                        ereport(tz_elevel,
                                        (errcode_for_file_access(),
                                         errmsg("could not read time
zone file \"%s\": %m",

(will probably get mangled by my mailer)

I get a much more useful:

WARNING: could not read time zone file "Default": No such file or directory
FATAL: failed to initialize timezone_abbreviations to "Default"
WARNING: could not read time zone file "Default": No such file or directory
FATAL: failed to initialize timezone_abbreviations to "Default"
LOG: background writer process (PID 3776) exited with exit code 0
LOG: terminating any other active server processes
WARNING: could not read time zone file "Default": No such file or directory
FATAL: failed to initialize timezone_abbreviations to "Default"
LOG: all server processes terminated; reinitializing
WARNING: could not read time zone file "Default": No such file or directory

which gives a strong further hint at the underlying issue.

Stefan

Stefan Kaltenbrunner <stefan@kaltenbrunner.cc> writes:

I get a much more useful:

WARNING: could not read time zone file "Default": No such file or directory
FATAL: failed to initialize timezone_abbreviations to "Default"

Hm, but why would the file not be there? Try hacking it to print the
whole path it's trying to open, maybe that will help.

regards, tom lane

Tom Lane wrote:

Stefan Kaltenbrunner <stefan@kaltenbrunner.cc> writes:

I get a much more useful:

WARNING: could not read time zone file "Default": No such file or directory
FATAL: failed to initialize timezone_abbreviations to "Default"

Hm, but why would the file not be there? Try hacking it to print the
whole path it's trying to open, maybe that will help.

WARNING: could not read time zone file
"/home/pgbuild/devel/pginst/share/postgresql/timezonesets/Default": No
such file or directory
FATAL: failed to initialize timezone_abbreviations to "Default"
WARNING: could not read time zone file
"/home/pgbuild/devel/pginst/share/postgresql/timezonesets/Default": No
such file or directory
FATAL: failed to initialize timezone_abbreviations to "Default"
LOG: background writer process (PID 1460) exited with exit code 0
LOG: terminating any other active server processes
WARNING: could not read time zone file
"/home/pgbuild/devel/pginst/share/postgresql/timezonesets/Default": No
such file or directory

$ ls -l /home/pgbuild/devel/pginst/share/postgresql/timezonesets/Default
-rw-r--r-- 1 pgbuild Administ 28630 Jul 28 20:03
/home/pgbuild/devel/pginst/share/postgresql/timezonesets/Default

so it's there but as a msys-virtual path - is that get passed to some
win32 function expecting a windows-style path ?

Stefan

Stefan Kaltenbrunner <stefan@kaltenbrunner.cc> writes:

WARNING: could not read time zone file
"/home/pgbuild/devel/pginst/share/postgresql/timezonesets/Default": No
such file or directory

$ ls -l /home/pgbuild/devel/pginst/share/postgresql/timezonesets/Default
-rw-r--r-- 1 pgbuild Administ 28630 Jul 28 20:03
/home/pgbuild/devel/pginst/share/postgresql/timezonesets/Default

so it's there but as a msys-virtual path - is that get passed to some
win32 function expecting a windows-style path ?

Hm. We pass it to fopen(). The equivalent code in pgtz.c generates the
path to /timezone files exactly the same way, but uses open() ... is
there a difference in what they'll take?

regards, tom lane

Stefan Kaltenbrunner wrote:

WARNING: could not read time zone file
"/home/pgbuild/devel/pginst/share/postgresql/timezonesets/Default": No
such file or directory

This is an MSys virtual path, of which postgres naturally knows
nothing.We should have made the appropriate calls to turn it into a
genuine Windows path. (Darn, not having a Windows box to test on is
annoying).

cheers

andrew

Stefan Kaltenbrunner <stefan@kaltenbrunner.cc> writes:

WARNING: could not read time zone file
"/home/pgbuild/devel/pginst/share/postgresql/timezonesets/Default": No
such file or directory

so it's there but as a msys-virtual path - is that get passed to some
win32 function expecting a windows-style path ?

Doh, I see what's the problem: we calculate the sharedir path using
my_exec_path, and falling back to the hardwired PGSHAREDIR path if
my_exec_path isn't correct. The problem is that in a Windows
subprocess, my_exec_path isn't correct until read_backend_variables
has been done, and *that happens after InitializeGUCOptions* in
SubPostmasterMain(). So we're trying to set up the tz name data
before we have the path we need.

The reason I didn't notice this in testing with EXEC_BACKEND is that
I wasn't testing in a relocated installation, and so the fallback
get_share_path calculation got the right answer anyway.

Not sure about a clean fix. Probably we'll have to do something
similar to the way TimeZone is handled, where we don't try to read
in the data until later on in the initialization sequence.

regards, tom lane

Tom Lane wrote:

Stefan Kaltenbrunner <stefan@kaltenbrunner.cc> writes:

WARNING: could not read time zone file
"/home/pgbuild/devel/pginst/share/postgresql/timezonesets/Default": No
such file or directory

so it's there but as a msys-virtual path - is that get passed to some
win32 function expecting a windows-style path ?

Doh, I see what's the problem: we calculate the sharedir path using
my_exec_path, and falling back to the hardwired PGSHAREDIR path if
my_exec_path isn't correct. The problem is that in a Windows
subprocess, my_exec_path isn't correct until read_backend_variables
has been done, and *that happens after InitializeGUCOptions* in
SubPostmasterMain(). So we're trying to set up the tz name data
before we have the path we need.

Is there a reason we have to do things in this order? Could we just
postpone the call to InitializeGUCOptions() for a couple of lines?

If not, then ...

The reason I didn't notice this in testing with EXEC_BACKEND is that
I wasn't testing in a relocated installation, and so the fallback
get_share_path calculation got the right answer anyway.

Not sure about a clean fix. Probably we'll have to do something
similar to the way TimeZone is handled, where we don't try to read
in the data until later on in the initialization sequence.

I guess we'd need to set a flag that would postpone reading the data
just during the startup phase, but have it called immediately in all
other cases.

cheers

andrew

Andrew Dunstan <andrew@dunslane.net> writes:

Is there a reason we have to do things in this order? Could we just
postpone the call to InitializeGUCOptions() for a couple of lines?

Maybe, but I'm disinclined to mess with that. I have a patch that
makes it work like TimeZone, but am having difficulty committing
... looks like that Polish script kiddie is at it again ...

regards, tom lane