TODO list

I wrote:

If nobody is working on this I am prepared to look at it:

. Allow logging of only data definition(DDL), or DDL and modification
statements

Here are some options:

1. change the type of "log_statement" option from boolean to string,
with allowed values of "all, mod, ddl, none" with default "none".
2. same as 1. but make boolean true values synonyms for "all" and
boolean false values synonyms for "none".
3. keep "log_statement" option as now and add a new option
"log_statement_level" with the same options as 1. but default to "all",
which will have no effect unless "log_statement" is true.

Also, I assume "modification statements" means insert/update/delete, or
are we talking about DDL mods (like "alter table")?

Finally, what about functions that have side effects? It would be nice
to be able to detect the statements to be logged at the syntactic level,
but it strikes me that that might not be possible.

cheers

andrew

Andrew Dunstan wrote:

2 things.

I submitted a patch for this 5 months ago, which is still waiting to be
merged (hope it hasn't bitrotted in the meantime):

. Allow log lines to include session-level information, like database
and user

If nobody is working on this I am prepared to look at it:

. Allow logging of only data definition(DDL), or DDL and modification
statements

Yes, sorry I haven't gotten back to that, and yes the DDL idea is a good
one.

For the log idea, I think we need to get a way to merge all the per-line
info into one setup, so pid, timestamp, user, etc would all be
configurable using your setup.

-- 
  Bruce Momjian                        |  http://candle.pha.pa.us
  pgman@candle.pha.pa.us               |  (610) 359-1001
  +  If your life is a hard drive,     |  13 Roberts Road
  +  Christ can be your backup.        |  Newtown Square, Pennsylvania 19073

Bruce Momjian said:

Andrew Dunstan wrote:

2 things.

I submitted a patch for this 5 months ago, which is still waiting to

be merged (hope it hasn't bitrotted in the meantime):

. Allow log lines to include session-level information, like database

and user

If nobody is working on this I am prepared to look at it:

. Allow logging of only data definition(DDL), or DDL and modification

statements

Yes, sorry I haven't gotten back to that, and yes the DDL idea is a good

one.

For the log idea, I think we need to get a way to merge all the
per-line info into one setup, so pid, timestamp, user, etc would all be

configurable using your setup.

I thought we had thrashed this out back in August. Certainly the only
thing I recall seeing after I submitted the patch was some stylistic
criticism from Neil, which I addressed in a revised patch.

Anyway, it is in principle doable. That's partly why I adopted a printf
style format string. There are some wrinkles, though:
. interaction with syslog pid/timestamp logging
. making sure the info is available when you need to log it - I had to
rearrange a few thing to avoid getting SEGVs, IIRC.

Also, the session duration logging part of the patch is orthogonal to the
issue.

cheers

andrew

Andrew Dunstan wrote:

Bruce Momjian said:

Andrew Dunstan wrote:

2 things.

I submitted a patch for this 5 months ago, which is still waiting to

be merged (hope it hasn't bitrotted in the meantime):

. Allow log lines to include session-level information, like database

and user

If nobody is working on this I am prepared to look at it:

. Allow logging of only data definition(DDL), or DDL and modification

statements

Yes, sorry I haven't gotten back to that, and yes the DDL idea is a good

one.

For the log idea, I think we need to get a way to merge all the
per-line info into one setup, so pid, timestamp, user, etc would all be

configurable using your setup.

I thought we had thrashed this out back in August. Certainly the only
thing I recall seeing after I submitted the patch was some stylistic
criticism from Neil, which I addressed in a revised patch.

Anyway, it is in principle doable. That's partly why I adopted a printf
style format string. There are some wrinkles, though:
. interaction with syslog pid/timestamp logging

Yes. If you use syslog, just don't ask for pid/timestamp and let syslog
do it. Of course, right now we are able to send non-pid/timestamp to
syslog _and_ send pid/timestamp to a log file, but that seems like a
rare operation that doesn't justify keeping the various log parameters
separate.

Also, I would like to see some kind of session identifier that is more
unique than pid, which wraps around. Ideally we could have 10{pid},
then then the pid wraps around, 20{pid), or something like that.

. making sure the info is available when you need to log it - I had to
rearrange a few thing to avoid getting SEGVs, IIRC.

Of course some messages, like postmaster status messages, don't have
some of these fields, like username or host. Is that going to cause
problems for tools that read our log files?

Also, the session duration logging part of the patch is orthogonal to the
issue.

You mean query duration? Yes, it is orthoginal.

-- 
  Bruce Momjian                        |  http://candle.pha.pa.us
  pgman@candle.pha.pa.us               |  (610) 359-1001
  +  If your life is a hard drive,     |  13 Roberts Road
  +  Christ can be your backup.        |  Newtown Square, Pennsylvania 19073

Bruce Momjian wrote:

Andrew Dunstan wrote:

I thought we had thrashed this out back in August. Certainly the only
thing I recall seeing after I submitted the patch was some stylistic
criticism from Neil, which I addressed in a revised patch.

Anyway, it is in principle doable. That's partly why I adopted a printf
style format string. There are some wrinkles, though:
. interaction with syslog pid/timestamp logging

Yes. If you use syslog, just don't ask for pid/timestamp and let syslog
do it. Of course, right now we are able to send non-pid/timestamp to
syslog _and_ send pid/timestamp to a log file, but that seems like a
rare operation that doesn't justify keeping the various log parameters
separate.

I'm OK with that as long as it is the consensus view. It does seem a
little odd to remove functionality (however rare) for the sake of
configuration neatness, though.

Also, I would like to see some kind of session identifier that is more
unique than pid, which wraps around. Ideally we could have 10{pid},
then then the pid wraps around, 20{pid), or something like that.

This requires some thought. ISTM it wouldn't buy you much unless you
made it persistent across server restarts, and possibly not even then. I
don't see it as a reason to hold up these features, though. If someone
wants to tackle this it could be plugged in to the loginfo feature very
easily as an extra escape sequence.

. making sure the info is available when you need to log it - I had to
rearrange a few thing to avoid getting SEGVs, IIRC.

Of course some messages, like postmaster status messages, don't have
some of these fields, like username or host. Is that going to cause
problems for tools that read our log files?

If users want a non-empty info string they will have to teach the tools
to handle it anyway. The point was, however, that rolling up PID and
timestamp into the printf-style format will require some significant
work, because we wouldn't want to lose that info if the user/db weren't
known, whereas the patch currently suppresses all log-info output if
these are not present (i.e. when MyProcPort == NULL).

Also, the session duration logging part of the patch is orthogonal to the
issue.

You mean query duration? Yes, it is orthoginal.

No, I meant the logging of the end of a session, including its duration,
which was also in the patch.

cheers

andrew

On Tue, 6 Jan 2004, Andrew Dunstan wrote:

Also, I would like to see some kind of session identifier that is more
unique than pid, which wraps around. Ideally we could have 10{pid},
then then the pid wraps around, 20{pid), or something like that.

This requires some thought. ISTM it wouldn't buy you much unless you
made it persistent across server restarts, and possibly not even then.

And on OpenBSD (though no other platforms that I know of) the PID is a
random number, so there is no "wrapping" to begin with.

Jon

Am Tuesday 06 January 2004 21:30 schrieb Jon Jensen:

On Tue, 6 Jan 2004, Andrew Dunstan wrote:

Also, I would like to see some kind of session identifier that is more
unique than pid, which wraps around. Ideally we could have 10{pid},
then then the pid wraps around, 20{pid), or something like that.

This requires some thought. ISTM it wouldn't buy you much unless you
made it persistent across server restarts, and possibly not even then.

And on OpenBSD (though no other platforms that I know of) the PID is a
random number, so there is no "wrapping" to begin with.

Linux >= 2.6 has random pids too.

Andrew Dunstan wrote:

Bruce Momjian wrote:

Andrew Dunstan wrote:

I thought we had thrashed this out back in August. Certainly the only
thing I recall seeing after I submitted the patch was some stylistic
criticism from Neil, which I addressed in a revised patch.

Anyway, it is in principle doable. That's partly why I adopted a printf
style format string. There are some wrinkles, though:
. interaction with syslog pid/timestamp logging

Yes. If you use syslog, just don't ask for pid/timestamp and let syslog
do it. Of course, right now we are able to send non-pid/timestamp to
syslog _and_ send pid/timestamp to a log file, but that seems like a
rare operation that doesn't justify keeping the various log parameters
separate.

I'm OK with that as long as it is the consensus view. It does seem a
little odd to remove functionality (however rare) for the sake of
configuration neatness, though.

Yes, agreed. Let's explore it. The rare functionality we would be
removing is for:

o User uses syslog
o User wants to log postmaster output to a file too
o User wants timestamp info in the postmaster output file

And the downside is that they get duplicate timestamps in syslog.

Now, if we don't merge timestamp into your new per-line log string, we
end up with a rather illogical and inflexible output format, only to
allow the rare case listed above.

Clearly this isn't a 100% clear decision, but it seems to lean in the
direction of removing the functionality with the goal of providing a
more logical logging API to the users.

Also, I would like to see some kind of session identifier that is more
unique than pid, which wraps around. Ideally we could have 10{pid},
then then the pid wraps around, 20{pid), or something like that.

This requires some thought. ISTM it wouldn't buy you much unless you
made it persistent across server restarts, and possibly not even then. I
don't see it as a reason to hold up these features, though. If someone
wants to tackle this it could be plugged in to the loginfo feature very
easily as an extra escape sequence.

Yes, that was my idea --- let's get this in and we can then add a
session id, and your point about restarts is a good one.

. making sure the info is available when you need to log it - I had to
rearrange a few thing to avoid getting SEGVs, IIRC.

Of course some messages, like postmaster status messages, don't have
some of these fields, like username or host. Is that going to cause
problems for tools that read our log files?

If users want a non-empty info string they will have to teach the tools
to handle it anyway. The point was, however, that rolling up PID and
timestamp into the printf-style format will require some significant
work, because we wouldn't want to lose that info if the user/db weren't
known, whereas the patch currently suppresses all log-info output if
these are not present (i.e. when MyProcPort == NULL).

Oh, good point. That would suggest that maybe we are better off leaving
pid and timestamp as separate options and _not_ merge them into your new
string. I am starting to think having your string print only
session-specific information is the best way.

I wonder if we should rename your GUC log_session_line or something like
that.

Also, the session duration logging part of the patch is orthogonal to the
issue.

You mean query duration? Yes, it is orthoginal.

No, I meant the logging of the end of a session, including its duration,
which was also in the patch.

Oh, I missed that one. It seems like a nice addition. I see you added
it when they ask for log_connections. Good idea.

-- 
  Bruce Momjian                        |  http://candle.pha.pa.us
  pgman@candle.pha.pa.us               |  (610) 359-1001
  +  If your life is a hard drive,     |  13 Roberts Road
  +  Christ can be your backup.        |  Newtown Square, Pennsylvania 19073

Jon Jensen wrote:

On Tue, 6 Jan 2004, Andrew Dunstan wrote:

Also, I would like to see some kind of session identifier that is more
unique than pid, which wraps around. Ideally we could have 10{pid},
then then the pid wraps around, 20{pid), or something like that.

This requires some thought. ISTM it wouldn't buy you much unless you
made it persistent across server restarts, and possibly not even then.

And on OpenBSD (though no other platforms that I know of) the PID is a
random number, so there is no "wrapping" to begin with.

OK, so a sessionid based on prefix+pid won't work portably. If we
*really* want to do it, a cluster-wide sequence generator would probably
be the way to go, but I suspect that with the ability to log session
termination explicitly (which I have already provided) much of the
supposed extra utility would disappear anyway.

cheers

andrew

Bruce Momjian wrote:

Andrew Dunstan wrote:

Bruce Momjian wrote:

Andrew Dunstan wrote:

I thought we had thrashed this out back in August. Certainly the only
thing I recall seeing after I submitted the patch was some stylistic
criticism from Neil, which I addressed in a revised patch.

Anyway, it is in principle doable. That's partly why I adopted a printf
style format string. There are some wrinkles, though:
. interaction with syslog pid/timestamp logging

Yes. If you use syslog, just don't ask for pid/timestamp and let syslog
do it. Of course, right now we are able to send non-pid/timestamp to
syslog _and_ send pid/timestamp to a log file, but that seems like a
rare operation that doesn't justify keeping the various log parameters
separate.

I'm OK with that as long as it is the consensus view. It does seem a
little odd to remove functionality (however rare) for the sake of
configuration neatness, though.

Yes, agreed. Let's explore it. The rare functionality we would be
removing is for:

o User uses syslog
o User wants to log postmaster output to a file too
o User wants timestamp info in the postmaster output file

And the downside is that they get duplicate timestamps in syslog.

Now, if we don't merge timestamp into your new per-line log string, we
end up with a rather illogical and inflexible output format, only to
allow the rare case listed above.

Clearly this isn't a 100% clear decision, but it seems to lean in the
direction of removing the functionality with the goal of providing a
more logical logging API to the users.

Also, I would like to see some kind of session identifier that is more
unique than pid, which wraps around. Ideally we could have 10{pid},
then then the pid wraps around, 20{pid), or something like that.

This requires some thought. ISTM it wouldn't buy you much unless you
made it persistent across server restarts, and possibly not even then. I
don't see it as a reason to hold up these features, though. If someone
wants to tackle this it could be plugged in to the loginfo feature very
easily as an extra escape sequence.

Yes, that was my idea --- let's get this in and we can then add a
session id, and your point about restarts is a good one.

. making sure the info is available when you need to log it - I had to
rearrange a few thing to avoid getting SEGVs, IIRC.

Of course some messages, like postmaster status messages, don't have
some of these fields, like username or host. Is that going to cause
problems for tools that read our log files?

If users want a non-empty info string they will have to teach the tools
to handle it anyway. The point was, however, that rolling up PID and
timestamp into the printf-style format will require some significant
work, because we wouldn't want to lose that info if the user/db weren't
known, whereas the patch currently suppresses all log-info output if
these are not present (i.e. when MyProcPort == NULL).

Oh, good point. That would suggest that maybe we are better off leaving
pid and timestamp as separate options and _not_ merge them into your new
string. I am starting to think having your string print only
session-specific information is the best way.

I wonder if we should rename your GUC log_session_line or something like
that.

Also, the session duration logging part of the patch is orthogonal to the
issue.

You mean query duration? Yes, it is orthoginal.

No, I meant the logging of the end of a session, including its duration,
which was also in the patch.

Oh, I missed that one. It seems like a nice addition. I see you added
it when they ask for log_connections. Good idea.

I think you are looking at the original patch, not the revised patch,
which is here: http://candle.pha.pa.us/mhonarc/patches2/msg00091.html
and provides a separate GUC var called "log_session_end" - Neil wanted
these not to be combined, IIRC.

I am agnostic as to the names of the variables.

cheers

andrew

Andrew Dunstan wrote:

I wrote:

If nobody is working on this I am prepared to look at it:

. Allow logging of only data definition(DDL), or DDL and modification
statements

Here are some options:

1. change the type of "log_statement" option from boolean to string,
with allowed values of "all, mod, ddl, none" with default "none".
2. same as 1. but make boolean true values synonyms for "all" and
boolean false values synonyms for "none".
3. keep "log_statement" option as now and add a new option
"log_statement_level" with the same options as 1. but default to "all",
which will have no effect unless "log_statement" is true.

I like 1.

Also, I assume "modification statements" means insert/update/delete, or

Yes.

are we talking about DDL mods (like "alter table")?

Alter is DDL.

Finally, what about functions that have side effects? It would be nice
to be able to detect the statements to be logged at the syntactic level,
but it strikes me that that might not be possible.

Not possible.

-- 
  Bruce Momjian                        |  http://candle.pha.pa.us
  pgman@candle.pha.pa.us               |  (610) 359-1001
  +  If your life is a hard drive,     |  13 Roberts Road
  +  Christ can be your backup.        |  Newtown Square, Pennsylvania 19073

Bruce Momjian wrote:

Andrew Dunstan wrote:

wow. that was nearly 3 months ago ...

I wrote:

If nobody is working on this I am prepared to look at it:

. Allow logging of only data definition(DDL), or DDL and modification
statements

Here are some options:

1. change the type of "log_statement" option from boolean to string,
with allowed values of "all, mod, ddl, none" with default "none".
2. same as 1. but make boolean true values synonyms for "all" and
boolean false values synonyms for "none".
3. keep "log_statement" option as now and add a new option
"log_statement_level" with the same options as 1. but default to "all",
which will have no effect unless "log_statement" is true.

I like 1.

Also, I assume "modification statements" means insert/update/delete, or

Yes.

are we talking about DDL mods (like "alter table")?

Alter is DDL.

Finally, what about functions that have side effects? It would be nice
to be able to detect the statements to be logged at the syntactic level,
but it strikes me that that might not be possible.

Not possible.

Subsequent discussion suggested we should add "syntax-errors" to the
allowed values (and I would favor making it the default).

The problem is this - in order to make the decision about whether or not
to log, we need to have parsed the statement (unless the level is set
to "all"). My simple approach, which would mean that the entire patch
would amount to around 100 lines, maybe, plus docco, would have the (I
think) trivial side effect of reversing the order in which a logged
statement and the corresponding parse error log entry appeared. You
objected to that effect, so I stopped work on it.

Now I can think of a couple of different approaches which would not have
this effect:
a. embed a time machine in postgres so we can make a decision based on
information we do not yet have, or
b. find some spot where we can trap the parse error log message before
it is emitted and then first log the statement. That spot is probably
somewhere in src/backend/utils/error/elog.c, but I am not quite sure where.

I have rejected as ugly and unmaintainable monkeying with the parser
itself to produce the desired effect, and I regret that idea a is beyond
my humble abilities :-)

cheers

andrew

Andrew Dunstan wrote:

Bruce Momjian wrote:

Andrew Dunstan wrote:

wow. that was nearly 3 months ago ...

Oh, I remember why I kept this email now. I am going to try to code
this.

Subsequent discussion suggested we should add "syntax-errors" to the
allowed values (and I would favor making it the default).

We already have log_min_error_statement. Seems that is what should be
used if someone wants only syntax errors.

The problem is this - in order to make the decision about whether or not
to log, we need to have parsed the statement (unless the level is set
to "all"). My simple approach, which would mean that the entire patch
would amount to around 100 lines, maybe, plus docco, would have the (I
think) trivial side effect of reversing the order in which a logged
statement and the corresponding parse error log entry appeared. You
objected to that effect, so I stopped work on it.

Now I can think of a couple of different approaches which would not have
this effect:
a. embed a time machine in postgres so we can make a decision based on
information we do not yet have, or
b. find some spot where we can trap the parse error log message before
it is emitted and then first log the statement. That spot is probably
somewhere in src/backend/utils/error/elog.c, but I am not quite sure where.

I have rejected as ugly and unmaintainable monkeying with the parser
itself to produce the desired effect, and I regret that idea a is beyond
my humble abilities :-)

I will start on this now. Thanks.

-- 
  Bruce Momjian                        |  http://candle.pha.pa.us
  pgman@candle.pha.pa.us               |  (610) 359-1001
  +  If your life is a hard drive,     |  13 Roberts Road
  +  Christ can be your backup.        |  Newtown Square, Pennsylvania 19073

Andrew Dunstan wrote:

Here are some options:

1. change the type of "log_statement" option from boolean to string,
with allowed values of "all, mod, ddl, none" with default "none".
2. same as 1. but make boolean true values synonyms for "all" and
boolean false values synonyms for "none".
3. keep "log_statement" option as now and add a new option
"log_statement_level" with the same options as 1. but default to "all",
which will have no effect unless "log_statement" is true.

I like 1.

OK, here is a patch that implements #1. Here is sample output:

test=> set client_min_messages = 'log';
SET
test=> set log_statement = 'mod';
SET
test=> select 1;
?column?
----------
1
(1 row)

test=> update test set x=1;
LOG: statement: update test set x=1;
ERROR: relation "test" does not exist
test=> update test set x=1;
LOG: statement: update test set x=1;
ERROR: relation "test" does not exist
test=> copy test from '/tmp/x';
LOG: statement: copy test from '/tmp/x';
ERROR: relation "test" does not exist
test=> copy test to '/tmp/x';
ERROR: relation "test" does not exist
test=> prepare xx as select 1;
PREPARE
test=> prepare xx as update x set y=1;
LOG: statement: prepare xx as update x set y=1;
ERROR: relation "x" does not exist
test=> explain analyze select 1;;
QUERY PLAN
------------------------------------------------------------------------------------
Result (cost=0.00..0.01 rows=1 width=0) (actual time=0.006..0.007 rows=1 loops=1)
Total runtime: 0.046 ms
(2 rows)

test=> explain analyze update test set x=1;
LOG: statement: explain analyze update test set x=1;
ERROR: relation "test" does not exist
test=> explain update test set x=1;
ERROR: relation "test" does not exist

It checks PREPARE and EXECUTE ANALYZE too. The log_statement values are
'none', 'mod', 'ddl', and 'all'. For 'all', it prints before the query
is parsed, and for ddl/mod, it does it right after parsing using the
node tag (or command tag for CREATE/ALTER/DROP), so any non-parse errors
will print after the log line.

-- 
  Bruce Momjian                        |  http://candle.pha.pa.us
  pgman@candle.pha.pa.us               |  (610) 359-1001
  +  If your life is a hard drive,     |  13 Roberts Road
  +  Christ can be your backup.        |  Newtown Square, Pennsylvania 19073

Unless I'm missing something, this patch has the effect that with values
of "ddl" or "mod" for log_statement, a statement with a parse error
will not be logged, which was what I hoped to avoid.

cheers

andrew

Bruce Momjian wrote:

Andrew Dunstan wrote:

Unless I'm missing something, this patch has the effect that with values
of "ddl" or "mod" for log_statement, a statement with a parse error
will not be logged, which was what I hoped to avoid.

Right. The query type can't be determined during a syntax error because
the parser couldn't identify the supplied command. I think that is
fine.

What it does allow is to for 'all' to display the command before the
syntax error.

-- 
  Bruce Momjian                        |  http://candle.pha.pa.us
  pgman@candle.pha.pa.us               |  (610) 359-1001
  +  If your life is a hard drive,     |  13 Roberts Road
  +  Christ can be your backup.        |  Newtown Square, Pennsylvania 19073

Bruce Momjian wrote:

Andrew Dunstan wrote:

Unless I'm missing something, this patch has the effect that with values
of "ddl" or "mod" for log_statement, a statement with a parse error
will not be logged, which was what I hoped to avoid.

Right. The query type can't be determined during a syntax error because
the parser couldn't identify the supplied command. I think that is
fine.

What it does allow is to for 'all' to display the command before the
syntax error.

If I had to make a choice I'd go the other way.

However, I think with a little extra work it might be possible to have both.

I will look into it at some stage.

cheers

andrew

Andrew Dunstan wrote:

Bruce Momjian wrote:

Andrew Dunstan wrote:

Unless I'm missing something, this patch has the effect that with values
of "ddl" or "mod" for log_statement, a statement with a parse error
will not be logged, which was what I hoped to avoid.

Right. The query type can't be determined during a syntax error because
the parser couldn't identify the supplied command. I think that is
fine.

What it does allow is to for 'all' to display the command before the
syntax error.

If I had to make a choice I'd go the other way.

Uh, what other way?

However, I think with a little extra work it might be possible to have both.

Right now, the way it is done, only a real syntax error skips logging.
If you referenced an invalid table or something, it does print the log
just before the invalid table name mention.

How would we test the command type before hitting a syntax error? I
can't think of a way, and I am not sure it would even be meaningful.

-- 
  Bruce Momjian                        |  http://candle.pha.pa.us
  pgman@candle.pha.pa.us               |  (610) 359-1001
  +  If your life is a hard drive,     |  13 Roberts Road
  +  Christ can be your backup.        |  Newtown Square, Pennsylvania 19073

Bruce Momjian wrote:

Andrew Dunstan wrote:

Bruce Momjian wrote:

Andrew Dunstan wrote:

Unless I'm missing something, this patch has the effect that with values
of "ddl" or "mod" for log_statement, a statement with a parse error
will not be logged, which was what I hoped to avoid.

Right. The query type can't be determined during a syntax error because
the parser couldn't identify the supplied command. I think that is
fine.

What it does allow is to for 'all' to display the command before the
syntax error.

If I had to make a choice I'd go the other way.

Uh, what other way?

reverse the order rather than suppress the message.

However, I think with a little extra work it might be possible to have both.

Right now, the way it is done, only a real syntax error skips logging.
If you referenced an invalid table or something, it does print the log
just before the invalid table name mention.

How would we test the command type before hitting a syntax error? I
can't think of a way, and I am not sure it would even be meaningful.

I agree that you can't test the statement type on a parse error. But
that doesn't mean to me that "mod" should suppress logging statements
with syntax errors. In fact, after the discussion surrounding this I
thought the consensus was to have these things as additive rather than
just one level selected.

How to do it in the order you prefer? I would trap the parse error and
log the statement before emitting the error log.

If I find a simple way I'll submit a further patch.

Certainly your patch contains the guts of what needs to be done in any case.

cheers

andrew

cheers

andrew

Andrew Dunstan wrote:

However, I think with a little extra work it might be possible to have both.

Right now, the way it is done, only a real syntax error skips logging.
If you referenced an invalid table or something, it does print the log
just before the invalid table name mention.

How would we test the command type before hitting a syntax error? I
can't think of a way, and I am not sure it would even be meaningful.

I agree that you can't test the statement type on a parse error. But
that doesn't mean to me that "mod" should suppress logging statements
with syntax errors. In fact, after the discussion surrounding this I
thought the consensus was to have these things as additive rather than
just one level selected.

It is additive in that 'mod' also includes 'ddl' queries.

How to do it in the order you prefer? I would trap the parse error and
log the statement before emitting the error log.

If I find a simple way I'll submit a further patch.

Certainly your patch contains the guts of what needs to be done in any case.

Right now we have log_min_error_statement:

#log_min_error_statement = panic # Values in order of increasing severity:
# debug5, debug4, debug3, debug2, debug1,
# info, notice, warning, error, panic(off)

which does allow control of printing only statements generating errors,
which includes syntax errors. I don't see why this functionality should
be mixed in with log_statement.

Did you want a 'syntax error' level to log_statement, that would print
only statements with syntax errors but not other errors? That doesn't
seem very useful to me.

-- 
  Bruce Momjian                        |  http://candle.pha.pa.us
  pgman@candle.pha.pa.us               |  (610) 359-1001
  +  If your life is a hard drive,     |  13 Roberts Road
  +  Christ can be your backup.        |  Newtown Square, Pennsylvania 19073

Bruce Momjian wrote:

Right now we have log_min_error_statement:

#log_min_error_statement = panic # Values in order of increasing severity:
# debug5, debug4, debug3, debug2, debug1,
# info, notice, warning, error, panic(off)

which does allow control of printing only statements generating errors,
which includes syntax errors. I don't see why this functionality should
be mixed in with log_statement.

Did you want a 'syntax error' level to log_statement, that would print
only statements with syntax errors but not other errors? That doesn't
seem very useful to me.

It wasn't my idea, but I thought it was a good one. But it would go
along with the idea of these settings as a list instead of a hierarchy,
e.g.:

log_statement = "syntax-errors, ddl, mod"

In fact, I liked it so much that I thought "syntax-errors" should be the
default instead of "none".

I think I'd prefer that to having it tied to the log_min_error_statement
level. But I don't care that much.

cheers

andrew

Andrew Dunstan wrote:

Bruce Momjian wrote:

Right now we have log_min_error_statement:

#log_min_error_statement = panic # Values in order of increasing severity:
# debug5, debug4, debug3, debug2, debug1,
# info, notice, warning, error, panic(off)

which does allow control of printing only statements generating errors,
which includes syntax errors. I don't see why this functionality should
be mixed in with log_statement.

Did you want a 'syntax error' level to log_statement, that would print
only statements with syntax errors but not other errors? That doesn't
seem very useful to me.

It wasn't my idea, but I thought it was a good one. But it would go
along with the idea of these settings as a list instead of a hierarchy,
e.g.:

log_statement = "syntax-errors, ddl, mod"

In fact, I liked it so much that I thought "syntax-errors" should be the
default instead of "none".

I think I'd prefer that to having it tied to the log_min_error_statement
level. But I don't care that much.

OK, at least we understand each other. Right now we don't have any
special "syntax error" log processing. We have errors logged through
log_min_error_statement, and mod/ddl through the new log_statement.

I can see a use case for having mod/ddl control of logging, and error
control of logging, but why would you want to see syntax error queries
but not other error queries? That's why I think log_min_error_statement
is sufficient. If we add syntax logging, wouldn't that conflict with
log_min_error_statement logging, because those are errors too. Maybe we
need to add a 'synax' mode to log_min_error_statement above error that
logs only syntax errors but not others.

-- 
  Bruce Momjian                        |  http://candle.pha.pa.us
  pgman@candle.pha.pa.us               |  (610) 359-1001
  +  If your life is a hard drive,     |  13 Roberts Road
  +  Christ can be your backup.        |  Newtown Square, Pennsylvania 19073

Bruce Momjian wrote:

Andrew Dunstan wrote:

Bruce Momjian wrote:

Right now we have log_min_error_statement:

#log_min_error_statement = panic # Values in order of increasing severity:
# debug5, debug4, debug3, debug2, debug1,
# info, notice, warning, error, panic(off)

which does allow control of printing only statements generating errors,
which includes syntax errors. I don't see why this functionality should
be mixed in with log_statement.

Did you want a 'syntax error' level to log_statement, that would print
only statements with syntax errors but not other errors? That doesn't
seem very useful to me.

It wasn't my idea, but I thought it was a good one. But it would go
along with the idea of these settings as a list instead of a hierarchy,
e.g.:

log_statement = "syntax-errors, ddl, mod"

In fact, I liked it so much that I thought "syntax-errors" should be the
default instead of "none".

I think I'd prefer that to having it tied to the log_min_error_statement
level. But I don't care that much.

OK, at least we understand each other. Right now we don't have any
special "syntax error" log processing. We have errors logged through
log_min_error_statement, and mod/ddl through the new log_statement.

I can see a use case for having mod/ddl control of logging, and error
control of logging, but why would you want to see syntax error queries
but not other error queries? That's why I think log_min_error_statement
is sufficient. If we add syntax logging,Thinks wouldn't that conflict with
log_min_error_statement logging, because those are errors too. Maybe we
need to add a 'synax' mode to log_min_error_statement above error that
logs only syntax errors but not others.

Thinks .... experiments .... yes, OK, I agree. Please forgive any
denseness. Not sure if we need another level.

Why do we have log_min_error_statement default to PANIC level? Wouldn't
ERROR be a better default?

cheers

andrew

Andrew Dunstan wrote:

I think I'd prefer that to having it tied to the log_min_error_statement
level. But I don't care that much.

OK, at least we understand each other. Right now we don't have any
special "syntax error" log processing. We have errors logged through
log_min_error_statement, and mod/ddl through the new log_statement.

I can see a use case for having mod/ddl control of logging, and error
control of logging, but why would you want to see syntax error queries
but not other error queries? That's why I think log_min_error_statement
is sufficient. If we add syntax logging,Thinks wouldn't that conflict with
log_min_error_statement logging, because those are errors too. Maybe we
need to add a 'synax' mode to log_min_error_statement above error that
logs only syntax errors but not others.

Thinks .... experiments .... yes, OK, I agree. Please forgive any
denseness. Not sure if we need another level.

No problem. It is good to think through these things to make sure we
have everything covered.

Why do we have log_min_error_statement default to PANIC level? Wouldn't
ERROR be a better default?

Panic basically means off, meaning we don't print queries that generate
errors. Should we print them by default?

-- 
  Bruce Momjian                        |  http://candle.pha.pa.us
  pgman@candle.pha.pa.us               |  (610) 359-1001
  +  If your life is a hard drive,     |  13 Roberts Road
  +  Christ can be your backup.        |  Newtown Square, Pennsylvania 19073

[moved to hackers]

Bruce Momjian wrote:

Andrew Dunstan wrote:

Why do we have log_min_error_statement default to PANIC level? Wouldn't
ERROR be a better default?

Panic basically means off, meaning we don't print queries that generate
errors. Should we print them by default?

I would. Some error happens and you don't know what statement caused it,
by which time it is too late to go back and turn up error statement logging.

cheers

andrew

Patch applied.

---------------------------------------------------------------------------

Bruce Momjian wrote:

Andrew Dunstan wrote:

Here are some options:

1. change the type of "log_statement" option from boolean to string,
with allowed values of "all, mod, ddl, none" with default "none".
2. same as 1. but make boolean true values synonyms for "all" and
boolean false values synonyms for "none".
3. keep "log_statement" option as now and add a new option
"log_statement_level" with the same options as 1. but default to "all",
which will have no effect unless "log_statement" is true.

I like 1.

OK, here is a patch that implements #1. Here is sample output:

test=> set client_min_messages = 'log';
SET
test=> set log_statement = 'mod';
SET
test=> select 1;
?column?
----------
1
(1 row)

test=> update test set x=1;
LOG: statement: update test set x=1;
ERROR: relation "test" does not exist
test=> update test set x=1;
LOG: statement: update test set x=1;
ERROR: relation "test" does not exist
test=> copy test from '/tmp/x';
LOG: statement: copy test from '/tmp/x';
ERROR: relation "test" does not exist
test=> copy test to '/tmp/x';
ERROR: relation "test" does not exist
test=> prepare xx as select 1;
PREPARE
test=> prepare xx as update x set y=1;
LOG: statement: prepare xx as update x set y=1;
ERROR: relation "x" does not exist
test=> explain analyze select 1;;
QUERY PLAN
------------------------------------------------------------------------------------
Result (cost=0.00..0.01 rows=1 width=0) (actual time=0.006..0.007 rows=1 loops=1)
Total runtime: 0.046 ms
(2 rows)

test=> explain analyze update test set x=1;
LOG: statement: explain analyze update test set x=1;
ERROR: relation "test" does not exist
test=> explain update test set x=1;
ERROR: relation "test" does not exist

It checks PREPARE and EXECUTE ANALYZE too. The log_statement values are
'none', 'mod', 'ddl', and 'all'. For 'all', it prints before the query
is parsed, and for ddl/mod, it does it right after parsing using the
node tag (or command tag for CREATE/ALTER/DROP), so any non-parse errors
will print after the log line.

-- 
Bruce Momjian                        |  http://candle.pha.pa.us
pgman@candle.pha.pa.us               |  (610) 359-1001
+  If your life is a hard drive,     |  13 Roberts Road
+  Christ can be your backup.        |  Newtown Square, Pennsylvania 19073

Index: doc/src/sgml/runtime.sgml
===================================================================
RCS file: /cvsroot/pgsql-server/doc/src/sgml/runtime.sgml,v
retrieving revision 1.257
diff -c -c -r1.257 runtime.sgml
*** doc/src/sgml/runtime.sgml	5 Apr 2004 03:02:03 -0000	1.257
--- doc/src/sgml/runtime.sgml	6 Apr 2004 03:56:08 -0000
***************
*** 2121,2132 ****
</varlistentry>

<varlistentry id="guc-log-statement" xreflabel="log_statement">
! <term><varname>log_statement</varname> (<type>boolean</type>)</term>
<listitem>
<para>
! Causes each SQL statement to be logged. The default is
! off. Only superusers can disable this option if it has been
! enabled by an administrator.
</para>

<note>
--- 2121,2141 ----
</varlistentry>

<varlistentry id="guc-log-statement" xreflabel="log_statement">
! <term><varname>log_statement</varname> (<type>string</type>)</term>
<listitem>
<para>
! Controls which SQL statement are logged. Valid values are
! <literal>all</>, <literal>ddl</>, <literal>mod</>, and
! <literal>none</>. <literal>ddl</> logs all data definition
! commands like <literal>CREATE</>, <literal>ALTER</>, and
! <literal>DROP</> commands. <literal>mod</> logs all
! <literal>ddl</> statements, plus <literal>INSERT</>,
! <literal>UPDATE</>, <literal>DELETE</>, <literal>TRUNCATE</>,
! and <literal>COPY FROM</>. <literal>PREPARE</> and
! <literal>EXPLAIN ANALYZE</> statements are also considered for
! appropriate commands. The default is <literal>none</>. Only
! superusers can reduce the detail of this option if it has been
! set by an administrator.
</para>

<note>
Index: src/backend/tcop/postgres.c
===================================================================
RCS file: /cvsroot/pgsql-server/src/backend/tcop/postgres.c,v
retrieving revision 1.397
diff -c -c -r1.397 postgres.c
*** src/backend/tcop/postgres.c	24 Mar 2004 22:40:29 -0000	1.397
--- src/backend/tcop/postgres.c	6 Apr 2004 03:56:11 -0000
***************
*** 87,92 ****
--- 87,94 ----
/* flag for logging end of session */
bool        Log_disconnections = false;

+ LogStmtLevel log_statement = LOGSTMT_NONE;
+ 
/*
* Flags for expensive function optimization -- JMH 3/9/92
*/
***************
*** 471,479 ****
List *
pg_parse_query(const char *query_string)
{
! 	List	   *raw_parsetree_list;

! if (log_statement)
ereport(LOG,
(errmsg("statement: %s", query_string)));

--- 473,482 ----
List *
pg_parse_query(const char *query_string)
{
! 	List	   *raw_parsetree_list,
! 			   *parsetree_item;

! if (log_statement == LOGSTMT_ALL)
ereport(LOG,
(errmsg("statement: %s", query_string)));

***************
*** 482,487 ****
--- 485,535 ----

raw_parsetree_list = raw_parser(query_string);

+ 	/* do log_statement tests for mod and ddl */
+ 	if (log_statement == LOGSTMT_MOD ||
+ 		log_statement == LOGSTMT_DDL)
+ 	{
+ 		foreach(parsetree_item, raw_parsetree_list)
+ 		{
+ 			Node	   *parsetree = (Node *) lfirst(parsetree_item);
+ 			const char *commandTag;
+ 	
+ 			if (IsA(parsetree, ExplainStmt) &&
+ 				((ExplainStmt *)parsetree)->analyze)
+ 				parsetree = (Node *)(((ExplainStmt *)parsetree)->query);
+ 			
+ 			if (IsA(parsetree, PrepareStmt))
+ 				parsetree = (Node *)(((PrepareStmt *)parsetree)->query);
+ 			
+ 			if (IsA(parsetree, SelectStmt))
+ 				continue;	/* optimization for frequent command */
+ 				
+ 			if (log_statement == LOGSTMT_MOD &&
+ 				(IsA(parsetree, InsertStmt) ||
+ 				 IsA(parsetree, UpdateStmt) ||
+ 				 IsA(parsetree, DeleteStmt) ||
+ 				 IsA(parsetree, TruncateStmt) ||
+ 				 (IsA(parsetree, CopyStmt) &&
+ 				  ((CopyStmt *)parsetree)->is_from)))	/* COPY FROM */
+ 			{
+ 				ereport(LOG,
+ 						(errmsg("statement: %s", query_string)));
+ 				break;
+ 			}
+ 			commandTag = CreateCommandTag(parsetree);
+ 			if (strncmp(commandTag, "CREATE ", strlen("CREATE ")) == 0 ||
+ 				strncmp(commandTag, "ALTER ", strlen("ALTER ")) == 0 ||
+ 				strncmp(commandTag, "DROP ", strlen("DROP ")) == 0 ||
+ 				IsA(parsetree, GrantStmt) ||	/* GRANT or REVOKE */
+ 				IsA(parsetree, CommentStmt))
+ 			{
+ 				ereport(LOG,
+ 						(errmsg("statement: %s", query_string)));
+ 				break;
+ 			}
+ 		}
+ 	}
+ 
if (log_parser_stats)
ShowUsage("PARSER STATISTICS");

***************
*** 2488,2494 ****
SetConfigOption("log_disconnections", "true", debug_context, gucsource);
}
if (debug_flag >= 2)
! 		SetConfigOption("log_statement", "true", debug_context, gucsource);
if (debug_flag >= 3)
SetConfigOption("debug_print_parse", "true", debug_context, gucsource);
if (debug_flag >= 4)
--- 2536,2542 ----
SetConfigOption("log_disconnections", "true", debug_context, gucsource);
}
if (debug_flag >= 2)
! 		SetConfigOption("log_statement", "all", debug_context, gucsource);
if (debug_flag >= 3)
SetConfigOption("debug_print_parse", "true", debug_context, gucsource);
if (debug_flag >= 4)
Index: src/backend/utils/misc/guc.c
===================================================================
RCS file: /cvsroot/pgsql-server/src/backend/utils/misc/guc.c,v
retrieving revision 1.197
diff -c -c -r1.197 guc.c
*** src/backend/utils/misc/guc.c	5 Apr 2004 03:02:07 -0000	1.197
--- src/backend/utils/misc/guc.c	6 Apr 2004 03:56:14 -0000
***************
*** 86,103 ****
bool doit, GucSource source);
#endif

! static const char *assign_defaultxactisolevel(const char *newval,
! bool doit, GucSource source);
! static const char *assign_log_min_messages(const char *newval,
! bool doit, GucSource source);
static const char *assign_client_min_messages(const char *newval,
bool doit, GucSource source);
static const char *assign_min_error_statement(const char *newval, bool doit,
GucSource source);
! static const char *assign_msglvl(int *var, const char *newval,
! bool doit, GucSource source);
static const char *assign_log_error_verbosity(const char *newval, bool doit,
GucSource source);
static bool assign_phony_autocommit(bool newval, bool doit, GucSource source);

--- 86,107 ----
bool doit, GucSource source);
#endif

! static const char *assign_defaultxactisolevel(const char *newval, bool doit,
! 						   GucSource source);
! static const char *assign_log_min_messages(const char *newval, bool doit,
! 						   GucSource source);
static const char *assign_client_min_messages(const char *newval,
bool doit, GucSource source);
static const char *assign_min_error_statement(const char *newval, bool doit,
GucSource source);
! static const char *assign_msglvl(int *var, const char *newval, bool doit,
! 						   GucSource source);
static const char *assign_log_error_verbosity(const char *newval, bool doit,
GucSource source);
+ static const char *assign_log_statement(const char *newval, bool doit,
+ 						   GucSource source);
+ static const char *assign_log_stmtlvl(int *var, const char *newval,
+ 						   bool doit, GucSource source);
static bool assign_phony_autocommit(bool newval, bool doit, GucSource source);

***************
*** 107,113 ****
#ifdef USE_ASSERT_CHECKING
bool		assert_enabled = true;
#endif
- bool		log_statement = false;
bool		log_duration = false;
bool		Debug_print_plan = false;
bool		Debug_print_parse = false;
--- 111,116 ----
***************
*** 145,150 ****
--- 148,154 ----
static char *client_min_messages_str;
static char *log_min_messages_str;
static char *log_error_verbosity_str;
+ static char *log_statement_str;
static char *log_min_error_statement_str;
static char *log_destination_string;
static bool phony_autocommit;
***************
*** 528,541 ****
false, NULL, NULL
},
{
- 		{"log_statement", PGC_USERLIMIT, LOGGING_WHAT,
- 			gettext_noop("Logs each SQL statement."),
- 			NULL
- 		},
- 		&log_statement,
- 		false, NULL, NULL
- 	},
- 	{
{"log_duration", PGC_USERLIMIT, LOGGING_WHAT,
gettext_noop("Logs the duration each completed SQL statement."),
NULL
--- 532,537 ----
***************
*** 1442,1447 ****
--- 1438,1451 ----
&log_error_verbosity_str,
"default", assign_log_error_verbosity, NULL
},
+ 	{
+ 		{"log_statement", PGC_USERLIMIT, LOGGING_WHAT,
+ 			gettext_noop("Sets the type of statements logged."),
+ 			gettext_noop("Valid values are \"none\", \"mod\", \"ddl\", and \"all\".")
+ 		},
+ 		&log_statement_str,
+ 		"none", assign_log_statement, NULL
+ 	},

{
{"log_min_error_statement", PGC_USERLIMIT, LOGGING_WHEN,
***************
*** 2007,2020 ****
struct config_string *conf = (struct config_string *) gconf;
char *str;

! 					/*
! 					 * Check to make sure we only have valid
! 					 * PGC_USERLIMITs
! 					 */
Assert(conf->gen.context != PGC_USERLIMIT ||
conf->assign_hook == assign_log_min_messages ||
! 					   conf->assign_hook == assign_client_min_messages ||
! 						conf->assign_hook == assign_min_error_statement);
*conf->variable = NULL;
conf->reset_val = NULL;
conf->session_val = NULL;
--- 2011,2021 ----
struct config_string *conf = (struct config_string *) gconf;
char	   *str;

! /* Check to make sure we only have valid PGC_USERLIMITs */
Assert(conf->gen.context != PGC_USERLIMIT ||
conf->assign_hook == assign_log_min_messages ||
! conf->assign_hook == assign_min_error_statement ||
! conf->assign_hook == assign_log_statement);
*conf->variable = NULL;
conf->reset_val = NULL;
conf->session_val = NULL;
***************
*** 3025,3039 ****
if (record->context == PGC_USERLIMIT &&
IsUnderPostmaster && !superuser())
{
! int old_int_value,
! new_int_value;

! 						/* all USERLIMIT strings are message levels */
! 						assign_msglvl(&new_int_value, newval,
! 									  true, source);
! 						assign_msglvl(&old_int_value, conf->reset_val,
! 									  true, source);
! 						if (new_int_value > old_int_value)
{
/* Limit non-superuser changes */
if (source > PGC_S_UNPRIVILEGED)
--- 3026,3048 ----
if (record->context == PGC_USERLIMIT &&
IsUnderPostmaster && !superuser())
{
! 						int		var_value, reset_value, new_value;
! 						const char * (*var_hook) (int *var, const char *newval,
! 									bool doit, GucSource source);
!     
! 						if (conf->assign_hook == assign_log_statement)
! 							var_hook = assign_log_stmtlvl;
! 						else
! 							var_hook = assign_msglvl;
! 
! 						(*var_hook) (&new_value, newval, true, source);
! 						(*var_hook) (&reset_value, conf->reset_val,	true,
! 									 source);
! 						(*var_hook) (&var_value, *conf->variable, true,
! 									 source);

! 						/* Limit non-superuser changes */
! 						if (new_value > reset_value)
{
/* Limit non-superuser changes */
if (source > PGC_S_UNPRIVILEGED)
***************
*** 3046,3055 ****
return false;
}
}
! 							/* Allow change if admin should override */
! 						assign_msglvl(&old_int_value, *conf->variable,
! 									  true, source);
! 						if (new_int_value < old_int_value)
{
if (source < PGC_S_UNPRIVILEGED &&
record->source > PGC_S_UNPRIVILEGED)
--- 3055,3063 ----
return false;
}
}
! 
! 						/* Allow change if admin should override */
! 						if (new_value < var_value)
{
if (source < PGC_S_UNPRIVILEGED &&
record->source > PGC_S_UNPRIVILEGED)
***************
*** 4646,4651 ****
--- 4654,4693 ----
{
if (doit)
Log_error_verbosity = PGERROR_VERBOSE;
+ 	}
+ 	else
+ 		return NULL;			/* fail */
+ 	return newval;				/* OK */
+ }
+ 
+ static const char *
+ assign_log_statement(const char *newval, bool doit, GucSource source)
+ {
+ 	return (assign_log_stmtlvl((int *)&log_statement, newval, doit, source));
+ }
+ 
+ static const char *
+ assign_log_stmtlvl(int *var, const char *newval, bool doit, GucSource source)
+ {
+ 	if (strcasecmp(newval, "none") == 0)
+ 	{
+ 		if (doit)
+ 			(*var) = LOGSTMT_NONE;
+ 	}
+ 	else if (strcasecmp(newval, "mod") == 0)
+ 	{
+ 		if (doit)
+ 			(*var) = LOGSTMT_MOD;
+ 	}
+ 	else if (strcasecmp(newval, "ddl") == 0)
+ 	{
+ 		if (doit)
+ 			(*var) = LOGSTMT_DDL;
+ 	}
+ 	else if (strcasecmp(newval, "all") == 0)
+ 	{
+ 		if (doit)
+ 			(*var) = LOGSTMT_ALL;
}
else
return NULL;			/* fail */
Index: src/backend/utils/misc/postgresql.conf.sample
===================================================================
RCS file: /cvsroot/pgsql-server/src/backend/utils/misc/postgresql.conf.sample,v
retrieving revision 1.112
diff -c -c -r1.112 postgresql.conf.sample
*** src/backend/utils/misc/postgresql.conf.sample	5 Apr 2004 03:02:08 -0000	1.112
--- src/backend/utils/misc/postgresql.conf.sample	6 Apr 2004 03:56:14 -0000
***************
*** 191,197 ****
# %s=session start timestamp
# %x=stop here in non-session processes
# %%='%'
! #log_statement = false
#log_hostname = false

--- 191,197 ----
# %s=session start timestamp
# %x=stop here in non-session processes
# %%='%'
! #log_statement = 'none'		# none, mod, ddl, all
#log_hostname = false

Index: src/include/tcop/tcopprot.h
===================================================================
RCS file: /cvsroot/pgsql-server/src/include/tcop/tcopprot.h,v
retrieving revision 1.63
diff -c -c -r1.63 tcopprot.h
*** src/include/tcop/tcopprot.h	24 Mar 2004 22:40:29 -0000	1.63
--- src/include/tcop/tcopprot.h	6 Apr 2004 03:56:14 -0000
***************
*** 35,40 ****
--- 35,53 ----
extern char *rendezvous_name;
extern int	max_stack_depth;

+ /* GUC-configurable parameters */
+ 
+ typedef enum
+ {
+ 	/* Reverse order so GUC USERLIMIT is easier */
+ 	LOGSTMT_ALL = 0,			/* log all statements, can not be -1 */
+ 	LOGSTMT_DDL,				/* log data definition statements */
+ 	LOGSTMT_MOD,				/* log modification statements, plus DDL */
+ 	LOGSTMT_NONE				/* log no statements */
+ } LogStmtLevel;
+ 
+ extern LogStmtLevel log_statement;
+ 
#ifndef BOOTSTRAP_INCLUDE

extern List *pg_parse_and_rewrite(const char *query_string,
Index: src/include/utils/guc.h
===================================================================
RCS file: /cvsroot/pgsql-server/src/include/utils/guc.h,v
retrieving revision 1.44
diff -c -c -r1.44 guc.h
*** src/include/utils/guc.h	19 Jan 2004 19:04:40 -0000	1.44
--- src/include/utils/guc.h	6 Apr 2004 03:56:15 -0000
***************
*** 103,109 ****
} GucSource;

/* GUC vars that are actually declared in guc.c, rather than elsewhere */
- extern bool log_statement;
extern bool log_duration;
extern bool Debug_print_plan;
extern bool Debug_print_parse;
--- 103,108 ----

---------------------------(end of broadcast)---------------------------
TIP 7: don't forget to increase your free space map settings

-- 
  Bruce Momjian                        |  http://candle.pha.pa.us
  pgman@candle.pha.pa.us               |  (610) 359-1001
  +  If your life is a hard drive,     |  13 Roberts Road
  +  Christ can be your backup.        |  Newtown Square, Pennsylvania 19073