I have a serious problem with hanging backends.

Started by Nonameover 24 years ago3 messagesbugs
Jump to latest
#1Noname
johan27@advalvas.be

Hi,

I have a serious problem with hanging backends.
This is wat I get from ps ->
postgres 23507 22794 0 13:33 ? 00:00:09
/usr/local/pgsql/bin/postgres 1.1.3.123 postgres bts01 SELECT waiting

This blocks all the other clients. (ODBC call failed)

The situation

- PostgreSQL 6.4.2 on i686-pc-linux-gnu, compiled by gcc 2.7.2.
(SlackWare)
- 30 Access 2000 clients connecting through 6.40.00.06 ODBC driver

- ODBC connections error's

-> CONN ERROR: func=SQLGetConnectOption,
desc='fOption=30002', errnum=205, errmsg='Unknown connect
option (Get)'

------------------------------------------------------------
henv=144902176, conn=144863832, status=0,
num_stmts=16
so ck=144902128, stmts=144902048, lobj_type=-999

----------------SocketInfo-------------------------------
socket=-1, reverse=0, errornumber=0, errormsg='(NULL)'
buffer_in=144870304, buffer_out=144874408
buffer_filled_in=0, buffer_filled_out=0, buffer_read_in=0

CONN ERROR: func=SQLSetConnectOption,
desc='fOption=30002, vParam=143791720', errnum=205,
errmsg='Unknown connect option (Set)'

------------------------------------------------------------
henv=144902176, conn=144863832, status=0, num_stmts=16
sock=144902128, stmts=144902048, lobj_type=-999

----------------SocketInfo-------------------------------
socket=-1, reverse=0, errornumber=0, errormsg='(NULL)'
buffer_in=144870304, buffer_out=144874408
buffer_filled_in=0, buffer_filled_out=0,
buffer_read_in=0

- In de server-log I found that the access-application wants to
connect with the user Admin then we get
FATAL 1 : user Admin not in pg_shadow

But in the connection properties the user postgres is set.
This (connection with user Admin) seems to have started with
the conversion 97 -> 2000

I've added the user, this error was gone, but some of the jobs

are admin owned other postgres

- I had problems before with the MaxBackendId
I've rebuild with MaxBackendId set to 128 and started the
postmaster with -B 256 parameter.
This was because of the message "cache invalidation
initialization failed"
So this was solved, or is

- In the loggings the SELECT waiting jobs hang on ProcessQuery and
the other InitPostgres fail.

- vacuum verbose analyze doen't seem to report errors

- debug higher then 4 is not usable (2 gig in 10 minutes)

In which direction do I have to search?
Can there be data corruption without being reported?
Can it be solved with a pg_dump file, destroydb, createdb,
psql -e file ? (Not done already -> production machine)
Is our version to old -> should an upgrade help?
.... ?

Thanks,

Jeroen Ryckeboer

#2Colin 't Hart
cthart@yahoo.com
In reply to: Noname (#1)
Re: I have a serious problem with hanging backends.

Jeroen asks:

[much snipped]

Is our version to old -> should an upgrade help?
.... ?

Yes. :-(

It's quite old; 7.1.3 is the latest and recommended version.
Any version < 7.0 has serious bugs.

Thanks,

Jeroen Ryckeboer

Cheers,

Colin

#3Tom Lane
tgl@sss.pgh.pa.us
In reply to: Noname (#1)
Re: I have a serious problem with hanging backends.

johan27@advalvas.be writes:

- PostgreSQL 6.4.2 on i686-pc-linux-gnu, compiled by gcc 2.7.2.

In which direction do I have to search?

Try a more recent release, for starters. PG 7.1.3 is current.

(You might want to find a less ancient gcc version, too...)

regards, tom lane