Re: log_min_messages per backend type

Re: log_min_messages per backend type

--
Andrew Dunstan
EDB: https://www.enterprisedb.com

Re: log_min_messages per backend type

> On Dec 10, 2025, at 02:24, Alvaro Herrera  wrote:
> 
> On 2025-Dec-09, Alvaro Herrera wrote:
> 
>> So here's your v6 again with those fixes as 0003 -- let's see what CI
>> thinks of this.  I haven't looked at your doc changes yet.
> 
> This passed CI, so I have marked it as Ready for Committer.  Further
> comments are still welcome, of course, but if there are none, I intend
> to get it committed in a few days.
> 
> 
> I'm not really happy with our usage of the translatable description
> field for things such as %b in log_line_prefix or the ps display; I
> think we should use the shorter names there instead.  Otherwise, you end
> up with log lines that look something like this (visible in a server
> with %b in log_line_prefix, which the TAP tests as well as pg_regress
> have):
> 
>  2025-12-08 21:38:04.304 CET autovacuum launcher[2452437] DEBUG: autovacuum launcher started
> 
> where the bit before the PID is marked for translation.  I think it
> should rather be
> 
>  2025-12-08 21:38:04.304 CET autovacuum[2452437] DEBUG: autovacuum launcher started
> 
> where that name (the same we'll use in log_min_messages) is not
> translated.
> 
> However, this issue is rather independent of the patch in this thread,
> so I'm going to discuss it in another thread; the name string though is
> added by this patch.
> 
> -- 
> Álvaro Herrera        Breisgau, Deutschland  —  https://www.EnterpriseDB.com/
> "Java is clearly an example of money oriented programming"  (A. Stepanov)

Overall the patch is solid, I just have a couple of suggestions:

1 - user experience
```
evantest=# show log_min_messages;
  log_min_messages
---------------------
 warning,backend:log
(1 row)
```

Now “show log_min_messages” prints the raw string the user set, in above example, there is not a white-space between the two log levels, and “show” result doesn’t have a white-space between the two log levels either. IMO, “SHOW log_min_messages” should display a stable result, in other words, say “fatal, backend:log” and “backend:log, fatal” should show the same result as they are actually meaning the same. So, I would suggest normalize the raw string: put the general level in the first place and sort others by process type, then SHOW returns the normalized string.

2 - refactoring for 0001
```
+		sep = strchr(tok, ':');
+		if (sep == NULL)
+		{
+			bool		found = false;
+
+			/* Reject duplicates for generic log level. */
+			if (genericlevel != -1)
+			{
+				GUC_check_errdetail("Generic log level was already assigned.");
+				guc_free(rawstring);
+				list_free(elemlist);
+				return false;
+			}
+
+			/* Is the log level valid? */
+			for (entry = server_message_level_options; entry && entry->name; entry++)
+			{
+				if (pg_strcasecmp(entry->name, tok) == 0)
+				{
+					genericlevel = entry->val;
+					found = true;
+					break;
+				}
+			}
+
+			if (!found)
+			{
+				GUC_check_errdetail("Unrecognized log level: \"%s\".", tok);
+				guc_free(rawstring);
+				list_free(elemlist);
+				return false;
+			}
+		}
+		else
+		{
+			char	   *loglevel;
+			char	   *ptype;
+			bool		found = false;
+
+			ptype = guc_malloc(LOG, (sep - tok) + 1);
+			if (!ptype)
+			{
+				guc_free(rawstring);
+				list_free(elemlist);
+				return false;
+			}
+			memcpy(ptype, tok, sep - tok);
+			ptype[sep - tok] = '\0';
+			loglevel = sep + 1;
+
+			/* Is the log level valid? */
+			for (entry = server_message_level_options; entry && entry->name; entry++)
+			{
+				if (pg_strcasecmp(entry->name, loglevel) == 0)
+				{
+					found = true;
+					break;
+				}
+			}
+
+			if (!found)
+			{
+				GUC_check_errdetail("Unrecognized log level: \"%s\".", loglevel);
+				guc_free(ptype);
+				guc_free(rawstring);
+				list_free(elemlist);
+				return false;
+			}
```

In the “if” and “else” clauses, there are duplicate code to valid log levels. We should refactor the code to avoid the duplication. For example, pull up “loglevel” to the “for” loop level, then we can valid it after the “if-else”.

Best regards,
--
Chao Li (Evan)
HighGo Software Co., Ltd.
https://www.highgo.com/

Re: log_min_messages per backend type

On Fri, Nov 21, 2025, at 11:13 AM, Euler Taveira wrote:
> I think you missed my point here. As I said if your proposal will let us with 2
> terminologies (backend type and process type) as shown above.  We can debate if
> we (1) keep the current terminology (backend type), (2) use the proposed one
> (process type) or (3) use a mixed variation (keep the SQL interface --
> backend_type column but change the description to "process type"). I wouldn't
> like to break compatibility (pg_stat_activity changes tend to break a lot of
> stuff) so I'm fine with (1) and (3).
>

After digesting it for a couple of days, I decided to adopt the terminology in
commit dbf8cfb4f02e (process type) just for this patch. This discussion about
changing "backend type" terminology in other parts belongs to another patch.

>>> Good question. The current patch uses "backend" to B_INVALID (that's exactly the
>>> MyBackendType for postmaster -- see below). I think it is reasonable to create a
>>> new category "postmaster" and assign it to B_INVALID.
>>
>> I guess that would work, but I think it's inadequate.  Maybe we could
>> add a new value B_POSTMASTER and have postmaster switch to that as early
>> as possible.  Then anything that still has B_INVALID must necessarily be
>> an improperly identified process.  Users wouldn't assign a value to that
>> one (the GUC wouldn't let you); instead those would always use the
>> default value.  Hopefully nobody would see that very often, or at all.
>>
>
> I will create a separate patch for this suggestion.
>

After reflection, add B_POSTMASTER into BackendType / proctypelist.h is not a
good idea. These data structures were designed to represent postmaster
children. The child_process_kinds array (launch_backend.c) is assembled with
proctypelist.h which means a function like postmaster_child_launch() could use
one of its elements (e.g. B_POSTMASTER) to start a new child; that's awkward. Do
you envision any issues to use B_INVALID for postmaster? (I wrote 0002 to
minimize the windows that each child process has B_INVALID.)

This new version contains the following changes:

- fix variable assignment (Chao Li)
- fix memory release order (Chao Li)
- change category for B_INVALID (backend -> postmaster) (Alvaro)
- rewrite documentation (Chao Li, Alvaro)
- rename backend type to process type (Alvaro)
- new patch (0002) that assigns MyBackendType as earlier as possible


-- 
Euler Taveira
EDB   https://www.enterprisedb.com/

Re: log_min_messages per backend type

On 2025/03/05 9:33, Euler Taveira wrote:
>> > +        Valid BACKENDTYPE values are ARCHIVER,
>> > +        AUTOVACUUM, BACKEND,
>> > +        BGWORKER, BGWRITER,
>> > +        CHECKPOINTER, LOGGER,
>> > +        SLOTSYNCWORKER, WALRECEIVER,
>> > +        WALSENDER, WALSUMMARIZER, and
>> > +        WALWRITER.

What about postmaster?

For parallel workers launched for parallel queries, should they follow
the backend's log level or the background worker's? Since they operate
as part of a parallel query executed by a backend, it seems more logical
for them to follow the backend's setting.

+	[B_CHECKPOINTER] = "checkpointer",
+	[B_STARTUP] = "backend",	/* XXX same as backend? */

I like the idea of allowing log levels to be set per process.
There were times I wanted to use debug5 specifically for
the startup process when troubleshooting WAL replay. It would be
helpful to distinguish the startup process from a regular backend,
so we can set its log level independently.

Regards,

-- 
Fujii Masao
Advanced Computing Technology Center
Research and Development Headquarters
NTT DATA CORPORATION

Re: log_min_messages per backend type

Hello Euler,

On 2024-Dec-17, Euler Taveira wrote:

> Sometimes you need to inspect some debug messages from autovacuum worker but
> you cannot apply the same setting for backends (that could rapidly fill the log
> file). This proposal aims to change log_min_messages to have different log
> levels depending on which backend type the message comes from.
> 
> The syntax was changed from enum to string and it accepts a list of elements.
> 
> Instead of enum, it now accepts a comma-separated list of elements (string).
> Each element is LOGLEVEL:BACKENDTYPE.

This format seems unintuitive.  I would have thought you do it the other
way around, "backendtype:loglevel" ... that seems more natural because
it's like assigning the 'loglevel' value to the 'backendtype' element.
  SET log_min_messages TO 'checkpointer:debug2, autovacuum:debug1';


I dislike the array of names in variable.c.  We already have an array in
launch_backend.c (child_process_kinds), plus GetBackendTypeDesc in
miscinit.c.  Maybe not for this patch to clean up though.


I think it should be acceptable to configure one global setting with
exceptions for particular backend types:

log_min_messages = WARNING, autovacuum:DEBUG1

Right now I think the code only accepts the unadorned log level if there
are no other items in the list.  I think the proposal downthread is to
use the keyword ALL for this,

  log_min_messages = all:WARNING, autovacuum:DEBUG1   # I don't like this

but I think it's inferior, because then "all" is not really "all", and I
think it would be different if I had said

  log_min_messages = autovacuum:DEBUG1, all:WARNING   # I don't like this

because it looks like the "all" entry should override the one I set for
autovacuum before, which frankly would not make sense to me.

So I think these two lines,

log_min_messages = WARNING, autovacuum:DEBUG1
log_min_messages = autovacuum:DEBUG1, WARNING

should behave identically and mean "set the level for autovacuum to
DEBUG1, and to any other backend type to WARNING.


Also, I think it'd be better to reject duplicates in the list.  Right
now it looks like the last entry for one backend type overrides prior
ones.  I mean

log_min_messages = autovacuum:DEBUG1, autovacuum:ERROR

would set autovacuum to error, but that might be mistake prone if your
string is long.  So implementation-wise I suggest to initialize the
whole newlogminmsgs array to -1, then scan the list of entries (saving
an entry without backend type as the one to use later and) setting every
backend type to the number specified; if we see trying to set a value
that's already different from -1, throw error.  After scanning the whole
log_min_messages array, we scan the newlogminmsgs and set any entries
that are still -1 to the value that we saved before.


The new code in variable.c should be before the /* DATESTYLE */ comment
rather than at the end of the file.


You still have many XXX comments.  Also, postgresql.conf should list the
valid values for backendtype, as well as show an example of a valid
setting.  Please don't use ALLCAPS backend types in the docs, this looks
ugly:

> +        Valid BACKENDTYPE values are ARCHIVER,
> +        AUTOVACUUM, BACKEND,
> +        BGWORKER, BGWRITER,
> +        CHECKPOINTER, LOGGER,
> +        SLOTSYNCWORKER, WALRECEIVER,
> +        WALSENDER, WALSUMMARIZER, and
> +        WALWRITER.

-- 
Álvaro Herrera        Breisgau, Deutschland  —  https://www.EnterpriseDB.com/
"No necesitamos banderas
 No reconocemos fronteras"                  (Jorge González)

Re: log_min_messages per backend type

On 2025-Aug-01, Japin Li wrote:

> If we set the log level for all backend types, I don't think a generic log
> level is necessary.

I don't understand what you mean by this.  Can you elaborate?

-- 
Álvaro Herrera               48°01'N 7°57'E  —  https://www.EnterpriseDB.com/
"If it is not right, do not do it.
If it is not true, do not say it." (Marcus Aurelius, Meditations)

Re: log_min_messages per backend type

On 2025-Nov-06, Euler Taveira wrote:

> +        Valid values are a comma-separated list of backendtype:level
> +        and a single level. The list allows it to use
> +        different levels per backend type. Only the single level
> +        is mandatory (order does not matter) and it is assigned to the backend
> +        types that are not specified in the list.
> +        Valid backendtype values are archiver,
> +        autovacuum, backend,
> +        bgworker, bgwriter,
> +        checkpointer, ioworker,
> +        syslogger, slotsyncworker,
> +        startup, walreceiver,
> +        walsender, walsummarizer, and
> +        walwriter.
> +        Valid level values are DEBUG5,
> +        DEBUG4, DEBUG3, DEBUG2,
> +        DEBUG1, INFO, NOTICE,
> +        WARNING, ERROR, LOG,
> +        FATAL, and PANIC.  Each level includes
> +        all the levels that follow it.  The later the level, the fewer messages are sent
> +        to the log.  The default is WARNING for all backend types.

I would use type:level rather than "backendtype".
Also, the glossary says we have "auxiliary processes" and "backends", so
from the user perspective, these types are not all backends, but instead
process types.

I think the list of backend types is pretty hard to read.  What do you
think about using 

to create a friendlier representation?  

So you would say something like "Valid types are listed in the table
below, each corresponding to either postmaster, an auxiliary process
type or a backend".  (Eh, wait, you seem not to have "postmaster" in
your list, what's up with that?)

(I'm not sure about making the log levels also a , because
for that list the order matters, and if we have more than one column
then the order is going to be ambiguous, and if we have just one column
it's going to be too tall.  But maybe it's not all that bad?)

-- 
Álvaro Herrera               48°01'N 7°57'E  —  https://www.EnterpriseDB.com/

Re: log_min_messages per backend type

On 2025-Nov-18, Euler Taveira wrote:

> On Thu, Nov 6, 2025, at 1:01 PM, Alvaro Herrera wrote:

> See miscadmin.h.
> 
> /*
>  * MyBackendType indicates what kind of a backend this is.
>  *
>  * If you add entries, please also update the child_process_kinds array in
>  * launch_backend.c.
>  */
> typedef enum BackendType
> 
> The "backend type" terminology is exposed.

Something appearing in the source code does not equate it being exposed
to end users.  Things are exposed when they are in the documentation, or
when the SQL interface requires you to use the term.  So I don't think
the fact that BackendType appears in the code forces us to use that name
in the user-visible interface now.

In the glossary, we talk about "processes"; in the definitions there,
"backend" is one type of process.  So in this list of "backend types"
that you want to introduce, what you really should be talking about is
"process types", where "backend" is one of the options.

> It is even available in the
> pg_stat_activity view. I agree that "backend" is not a good name to define a
> Postgres process. I don't think we should be inconsistent only in this patch.
> Even if the proposal is to rename enum BackendType (I won't propose it as part
> of this patch), it would make some extension authors unhappy according to
> codesearch.debian.net.

I don't think we should rename the BackendType enum.  That's in the
source code, not in the documentation or the SQL interface, so we don't
need to care.

> > I think the list of backend types is pretty hard to read.  What do you
> > think about using 
> > 
> > to create a friendlier representation?  
> 
> I tried but didn't like the result. See the attached image.

Well, I like it very much.  The original is a solid wall of text, very
hard to read, where you have to squint hunting commas in order to
distinguish one item from the next.  (Maybe you are young and have good
eyesight, but you won't be young forever.)  In the screenshot you show,
the list of possible process types to use is nicely separated, which
makes it very easy to catch at a glance.  Maybe remove "the table",
which is obviously inappropriate, and just say "Valid process types are
listed below, each corresponding to either postmaster, an auxiliary
process type or a backend".

(Note your original wording says "a backend type, corresponding to [blah
blah] or a backend" which makes no sense -- how is a backend a type of
backend?  It isn't.  It's a type of process.)

> You said table but it is a multi-column list; it doesn't contain a
> header or borders.

Right.  You don't need a full-blown table here: this simple list is
perfectly adequate.

> Good question. The current patch uses "backend" to B_INVALID (that's exactly the
> MyBackendType for postmaster -- see below). I think it is reasonable to create a
> new category "postmaster" and assign it to B_INVALID.

I guess that would work, but I think it's inadequate.  Maybe we could
add a new value B_POSTMASTER and have postmaster switch to that as early
as possible.  Then anything that still has B_INVALID must necessarily be
an improperly identified process.  Users wouldn't assign a value to that
one (the GUC wouldn't let you); instead those would always use the
default value.  Hopefully nobody would see that very often, or at all.

-- 
Álvaro Herrera        Breisgau, Deutschland  —  https://www.EnterpriseDB.com/

Re: log_min_messages per backend type

Hello

I noticed failures under Windows, and realized that the EXEC_BACKEND
case was busted.  That's easy to fix -- just assign MyBackendType in
SubPostmasterMain() as well.  With that fix I was still getting some
crashes in three test modules, and after some debugging it turned out
that if you have shared_preload_library with a background worker and use
%b in the log_line_prefix, you get a crash trying to expand a name that
hasn't been set up yet.  I figured we can use a constant string in that
case.  Better ideas for that string welcome.

So here's your v6 again with those fixes as 0003 -- let's see what CI
thinks of this.  I haven't looked at your doc changes yet.


BTW with %b in log_line_prefix, the log file looks ... interesting.

-- 
Álvaro Herrera        Breisgau, Deutschland  —  https://www.EnterpriseDB.com/
"Before you were born your parents weren't as boring as they are now. They
got that way paying your bills, cleaning up your room and listening to you
tell them how idealistic you are."  -- Charles J. Sykes' advice to teenagers

Re: log_min_messages per backend type

BTW another thing I realized while looking this over, is that we quite
uselessly transform the integer backend type to a string, pass it as a
string using the --forkchild= argument to the child process, then parse
the string back to an int to use as an array index.  It would be much
easier to just use the integer value everywhere, as the attached shows.

-- 
Álvaro Herrera               48°01'N 7°57'E  —  https://www.EnterpriseDB.com/
"I love the Postgres community. It's all about doing things _properly_. :-)"
(David Garamond)

Re: log_min_messages per backend type

On 2025-Dec-09, Alvaro Herrera wrote:

> So here's your v6 again with those fixes as 0003 -- let's see what CI
> thinks of this.  I haven't looked at your doc changes yet.

This passed CI, so I have marked it as Ready for Committer.  Further
comments are still welcome, of course, but if there are none, I intend
to get it committed in a few days.


I'm not really happy with our usage of the translatable description
field for things such as %b in log_line_prefix or the ps display; I
think we should use the shorter names there instead.  Otherwise, you end
up with log lines that look something like this (visible in a server
with %b in log_line_prefix, which the TAP tests as well as pg_regress
have):

  2025-12-08 21:38:04.304 CET autovacuum launcher[2452437] DEBUG: autovacuum launcher started

where the bit before the PID is marked for translation.  I think it
should rather be

  2025-12-08 21:38:04.304 CET autovacuum[2452437] DEBUG: autovacuum launcher started

where that name (the same we'll use in log_min_messages) is not
translated.

However, this issue is rather independent of the patch in this thread,
so I'm going to discuss it in another thread; the name string though is
added by this patch.

-- 
Álvaro Herrera        Breisgau, Deutschland  —  https://www.EnterpriseDB.com/
"Java is clearly an example of money oriented programming"  (A. Stepanov)

Re: log_min_messages per backend type

On 2026-Jan-15, Euler Taveira wrote:

> +	/*
> +	 * Use a stable representation of log_min_messages. The generic level is
> +	 * always the first element and the other elements (type:level) are sorted
> +	 * by process type.
> +	 */
> +	list_sort(elemlist, string_cmp);
> +	foreach(l, elemlist)
> +	{
> +		char	   *tok = (char *) lfirst(l);
> +
> +		if (strchr(tok, ':') == NULL)
> +		{
> +			elemlist = list_delete(elemlist, tok);
> +			elemlist = lcons(tok, elemlist);
> +			break;
> +		}
> +	}

> +static int
> +string_cmp(const ListCell *a, const ListCell *b)
> +{
> +	const char *s = lfirst(a);
> +	const char *t = lfirst(b);
> +
> +	return strcmp(s, t);
> +}

I'm not opposed to this idea, but I think it should be implemented not
by sorting and then moving the first element to the top of the list, but
instead by modifying the cmp function so that the desired order is
achieved directly.  So the cmp() should return -1 if element a has no
colon, or 1 if element b has no colon, otherwise return per strcmp.
That way you can remove the foreach() block above, which is icky.

-- 
Álvaro Herrera               48°01'N 7°57'E  —  https://www.EnterpriseDB.com/
"Computing is too important to be left to men." (Karen Spärck Jones)

Re: log_min_messages per backend type

On 2026-Jan-15, Euler Taveira wrote:

> On Tue, Dec 9, 2025, at 1:30 PM, Alvaro Herrera wrote:
> > BTW another thing I realized while looking this over, is that we quite
> > uselessly transform the integer backend type to a string, pass it as a
> > string using the --forkchild= argument to the child process, then parse
> > the string back to an int to use as an array index.  It would be much
> > easier to just use the integer value everywhere, as the attached shows.
> >
> 
> It is a good simplification.

Thanks!  Pushed.

> There is just one oversight.
> 
> +   child_type = (BackendType) atoi(child_kind);
> +   if (child_type <= B_INVALID || child_type > BACKEND_NUM_TYPES)
>         elog(ERROR, "unknown child kind %s", child_kind);
> 
> It should be BACKEND_NUM_TYPES - 1.

Good catch, thanks.

-- 
Álvaro Herrera               48°01'N 7°57'E  —  https://www.EnterpriseDB.com/
"Digital and video cameras have this adjustment and film cameras don't for the
same reason dogs and cats lick themselves: because they can."   (Ken Rockwell)

Re: log_min_messages per backend type

On 2026-Jan-15, Euler Taveira wrote:

> @@ -1275,10 +1279,16 @@ check_log_min_messages(char **newval, void **extra, GucSource source)
>  	char	   *rawstring;
>  	List	   *elemlist;
>  	ListCell   *l;
> +	char	   *result;
> +	bool		first = true;
>  	int			newlevel[BACKEND_NUM_TYPES];
>  	bool		assigned[BACKEND_NUM_TYPES];
>  	int			genericlevel = -1;	/* -1 means not assigned */
>  
> +	result = (char *) guc_malloc(LOG, MAX_LMM_STR_LEN);
> +	if (!result)
> +		return false;
     ...
> +	foreach(l, elemlist)
> +	{
> +		char	   *tok = (char *) lfirst(l);
> +
> +		if (first)
> +		{
> +			strncpy(result, tok, MAX_LMM_STR_LEN);
> +			first = false;
> +		}
> +		else
> +		{
> +			strlcat(result, ", ", MAX_LMM_STR_LEN);
> +			strlcat(result, tok, MAX_LMM_STR_LEN);
> +		}
> +	}

BTW I think this whole MAX_LMM_STR_LEN thing is unnecessary.  It is
probably easy enough to use a StringInfo to create the value
iteratively, then once you know its final length do a guc_malloc() and
strcpy the value there.

-- 
Álvaro Herrera         PostgreSQL Developer  —  https://www.EnterpriseDB.com/
"The Gord often wonders why people threaten never to come back after they've
been told never to return" (www.actsofgord.com)

Re: log_min_messages per backend type

On 2026-Jan-15, Euler Taveira wrote:

> @@ -586,6 +588,8 @@ SubPostmasterMain(int argc, char *argv[])
>  	IsPostmasterEnvironment = true;
>  	whereToSendOutput = DestNone;
>  
> +	MyBackendType = B_BACKEND;
> +
>  	/*
>  	 * Capture the end of process creation for logging. We don't include the
>  	 * time spent copying data from shared memory and setting up the backend.

BTW I think SubPostmasterMain() shouldn't do this.

-- 
Álvaro Herrera               48°01'N 7°57'E  —  https://www.EnterpriseDB.com/
"Curiosity is the hope that something wondrous waits just out of sight,
for we are surrounded by a garden of miracles"                       (Namron)
                           (Marjorie Liu and Sana Takeda, Monstress: Inferno)

Re: log_min_messages per backend type

Here are some fixups for the main patch.  I include yours so that CI
will test this one.

* I found some errdetail messages unclear, so I reworded them.
Specifically you had
 ERROR:  invalid value for parameter "log_min_messages": "backend:error, debug1, backend:warning"
 DETAIL:  Process type "backend" was already assigned.
which is not completely incorrect but looks weird from the user
perspective.  I changed it to
 DETAIL:  Redundant log level specification for process type "backend".

Also,
 ERROR:  invalid value for parameter "log_min_messages": "debug1, backend:error, fatal"
 DETAIL:  Generic log level was already assigned.
I changed to
 DETAIL:  Redundant specification of default log level.

and added process type in this one:
 ERROR:  invalid value for parameter "log_min_messages": "backend:error, checkpointer:bar, archiver:debug1"
 DETAIL:  Unrecognized log level: "bar".
changed to
 DETAIL:  Unrecognized log level for process type "checkpointer": "bar".
(This one can show an invalid process type if the level is also bogus.
I don't think that's a terrible problem.)

* I'm not fond of the term "generic", so I changed it to "default",
which IMO makes more sense from the user point of view.

* There seemed to be too much guc_mallocking[*] going on; for instance for
ptype.  I simplified that by just overwriting the : to a '\0', then we
can use the original string for pg_strcasecmp().  At the bottom of the
loop we restore the ':', so that the code below can reuse the original
values.  This is okay because we have already mallocked a copy for
SplitGUCList.

* There's no reason to add a WARNING to every process type in the
proctypelist.h table.  We can just set the values to WARNING in the
log_min_messages initializer in guc_tables.c.

* There's no reason AFAICS to have log_min_messages_process_types in
guc_tables.c; nobody else uses it.  I made it a local array of char*
inside the check_log_min_messages function.

* There's no need to initialize the newlevel[] array, since we're going
to assign values to every individual element at one point or another.
And for the assigned[] array, we don't need to set the value of each to
'false' in a loop; we can just initialize to '{0}' and the compiler will
do the needful.  With this we can remove the loop entirely.

* "guc_free()+list_free()+return false" the whole time in the various
failure places across the loop was tiresome.  I put them all in a single
place and jump there with a goto.

* in foreach() loops, tracking of a 'bool first' is unnecessary.  You
can just compare foreach_current_index() with 0.

* Rewrote docs and comments

-- 
Álvaro Herrera         PostgreSQL Developer  —  https://www.EnterpriseDB.com/

Re: log_min_messages per backend type

On 2026-Feb-09, Chao Li wrote:

> While it’s reasonable to configure a full comma-separated list in
> postgresql.conf, it seems there’s currently no convenient way for a
> superuser to change the log level of a single process type. For
> example:
> 
> ```
> evantest=# alter system set log_min_messages='backend:debug4';
> ERROR:  invalid value for parameter "log_min_messages": "backend:debug4"
> DETAIL:  Default log level was not defined.
> ```
> 
> This feels a bit inconvenient in practice. Intuitively, it would be
> nice if setting a single type:level could be merged into the existing
> log_min_messages value, rather than requiring users to restate the
> entire configuration.

This has been discussed before and no satisfactory solution has been
found.  You're welcome to spend time thinking on this; see e.g.
https://postgr.es/m/20200928033924.GA998042@toroid.org where not only
the details were discussed, but also some links to other threads about
the same idea were given.

-- 
Álvaro Herrera               48°01'N 7°57'E  —  https://www.EnterpriseDB.com/

Re: log_min_messages per backend type

On 2026-Feb-09, Chao Li wrote:

> There’s also a related regression in the interactive case. Before this
> patch, when debugging via psql, a superuser could easily do:
>
> ```
> # set log_min_messages = debug5
> ```
> 
> to temporarily increase verbosity for the current backend. With the
> new syntax, doing the equivalent now requires copying the entire
> log_min_messages string and modifying just one part, which feels
> noticeably heavier for quick debugging.

Nonsense.  You can still do the above, and it has the same behavior as
before.

-- 
Álvaro Herrera         PostgreSQL Developer  —  https://www.EnterpriseDB.com/
"Curiosity is the hope that something wondrous waits just out of sight,
for we are surrounded by a garden of miracles"                       (Namron)
                           (Marjorie Liu and Sana Takeda, Monstress: Inferno)

Re: log_min_messages per backend type

On 2026-Feb-06, Euler Taveira wrote:

> On Fri, Feb 6, 2026, at 8:05 AM, Alvaro Herrera wrote:

> > * I'm not fond of the term "generic", so I changed it to "default",
> > which IMO makes more sense from the user point of view.
> 
> I'm fine with "default" instead of "generic". However, you forgot to replace
> "generic" in a few places. The attached patch fixes them all.

I just pushed this, and somehow I forgot to squash this into the commit.
I don't think it matters terribly much though, so I'm going to hang onto
this for a while in case some bug is found.

I made some more changes though.  Notably, I reworded some things in the
docs that appeared ungrammatical and added a simple example; added the
list of process types to postgresql.conf.sample; and moved the GUC
check/assign hooks to elog.c together with the other elog-affecting
GUCs, as I didn't think they made much sense in variable.c.

Another thing, which is actually quite significant, is that I fixed a
bug I had introduced on Friday without realizing: I had added a "break"
in the loop that searches for process types.  But we mustn't break
there!  Otherwise we fail to set the level for process types that share
the same process "category" with others, such as "backend" and
"autovacuum".  So I removed the break and added a comment about it.
It's sad that no tests broke when I introduced that bug, but I think it
would be difficult to cover this.

Thanks

-- 
Álvaro Herrera         PostgreSQL Developer  —  https://www.EnterpriseDB.com/
"That sort of implies that there are Emacs keystrokes which aren't obscure.
I've been using it daily for 2 years now and have yet to discover any key
sequence which makes any sense."                        (Paul Thomas)

Re: log_min_messages per backend type

On 2026-Feb-09, Alvaro Herrera wrote:

> I just pushed this, and somehow I forgot to squash this into the commit.
> I don't think it matters terribly much though, so I'm going to hang onto
> this for a while in case some bug is found.

Euler mentioned offlist that I also forgot to remove
log_min_messages_process_types from guc.h.  So that gives us this patch
for now.


diff --git a/src/backend/utils/error/elog.c b/src/backend/utils/error/elog.c
index 129906e2daa..59315e94e3e 100644
--- a/src/backend/utils/error/elog.c
+++ b/src/backend/utils/error/elog.c
@@ -2190,7 +2190,7 @@ check_log_min_messages(char **newval, void **extra, GucSource source)
 	char	   *result;
 	int			newlevel[BACKEND_NUM_TYPES];
 	bool		assigned[BACKEND_NUM_TYPES] = {0};
-	int			genericlevel = -1;	/* -1 means not assigned */
+	int			defaultlevel = -1;	/* -1 means not assigned */
 
 	const char *const process_types[] = {
 #define PG_PROCTYPE(bktype, bkcategory, description, main_func, shmem_attach) \
@@ -2228,8 +2228,8 @@ check_log_min_messages(char **newval, void **extra, GucSource source)
 			const struct config_enum_entry *entry;
 			bool		found;
 
-			/* Reject duplicates for generic log level. */
-			if (genericlevel != -1)
+			/* Reject duplicates for default log level. */
+			if (defaultlevel != -1)
 			{
 				GUC_check_errdetail("Redundant specification of default log level.");
 				goto lmm_fail;
@@ -2241,7 +2241,7 @@ check_log_min_messages(char **newval, void **extra, GucSource source)
 			{
 				if (pg_strcasecmp(entry->name, elem) == 0)
 				{
-					genericlevel = entry->val;
+					defaultlevel = entry->val;
 					found = true;
 					break;
 				}
@@ -2331,9 +2331,9 @@ lmm_fail:
 	}
 
 	/*
-	 * The generic log level must be specified. It is the fallback value.
+	 * The default log level must be specified. It is the fallback value.
 	 */
-	if (genericlevel == -1)
+	if (defaultlevel == -1)
 	{
 		GUC_check_errdetail("Default log level was not defined.");
 		guc_free(rawstring);
@@ -2345,7 +2345,7 @@ lmm_fail:
 	for (int i = 0; i < BACKEND_NUM_TYPES; i++)
 	{
 		if (!assigned[i])
-			newlevel[i] = genericlevel;
+			newlevel[i] = defaultlevel;
 	}
 
 	/*
diff --git a/src/include/utils/guc.h b/src/include/utils/guc.h
index 8acbdba7ff5..c46203fabfe 100644
--- a/src/include/utils/guc.h
+++ b/src/include/utils/guc.h
@@ -329,8 +329,6 @@ extern PGDLLIMPORT bool trace_sort;
 extern PGDLLIMPORT bool optimize_bounded_sort;
 #endif
 
-extern PGDLLIMPORT const char *const log_min_messages_process_types[];
-
 /*
  * Declarations for options for enum values
  *

-- 
Álvaro Herrera               48°01'N 7°57'E  —  https://www.EnterpriseDB.com/
"I suspect most samba developers are already technically insane...
Of course, since many of them are Australians, you can't tell." (L. Torvalds)

Re: log_min_messages per backend type

On Thu, Jul 31, 2025, at 8:34 PM, Japin Li wrote:
> On Thu, 31 Jul 2025 at 18:51, Alvaro Herrera  wrote:
>> On 2025-Aug-01, Japin Li wrote:
>>
>>> If we set the log level for all backend types, I don't think a generic log
>>> level is necessary.
>>
>> I don't understand what you mean by this.  Can you elaborate?
>>
> For example:
>
> ALTER SYSTEM SET log_min_messages TO
> 'archiver:info, autovacuum:info, backend:info, bgworker:info, 
> bgwriter:info, checkpointer:info, ioworker:info, logger:info, 
> slotsyncworker:info, startup:info, walreceiver:info, walsender:info, 
> walsummarizer:info, walwriter:info';
>
> Given that we've set a log level for every backend type and
> assigned[BACKEND_NUM_TYPES] is true, a generic level seems redundant.
>

The main reason I didn't consider this idea was that this GUC assignment will
fail as soon as a new backend type is added. Restore a dump should work
across major versions.

> I think we can avoid memory allocation by using pg_strncasecmp().
>

loglevel does not required a memory allocation but I didn't include the btype
part because it complicates the error message that uses btype a few lines
above.

This new patch contains the following changes:

- patch was rebased
- use commit dbf8cfb4f02
- use some GUC memory allocation functions
- avoid one memory allocation (suggested by Japin Li)
- rename backend type: logger -> syslogger
- adjust tests to increase test coverage
- improve documentation and comments to reflect the current state


-- 
Euler Taveira
EDB   https://www.enterprisedb.com/

Re: log_min_messages per backend type

Re: log_min_messages per backend type

> On Feb 11, 2026, at 06:05, Euler Taveira  wrote:
> 
> On Mon, Feb 9, 2026, at 10:03 AM, Alvaro Herrera wrote:
>> On 2026-Feb-09, Alvaro Herrera wrote:
>> 
>>> I just pushed this, and somehow I forgot to squash this into the commit.
>>> I don't think it matters terribly much though, so I'm going to hang onto
>>> this for a while in case some bug is found.
>> 
>> Euler mentioned offlist that I also forgot to remove
>> log_min_messages_process_types from guc.h.  So that gives us this patch
>> for now.
>> 
> 
> I expanded this patch a bit with fixes proposed by Man Zeng and Noriyoshi
> Shinoda. Besides these changes, I noticed that
> 
> * default value in uppercase: since we changed the vartype from enum to string
>  and enum maps DEBUG1 to "debug1" (see server_message_level_options), it
>  should use the lowercase string to keep the same output as previous versions.
>  There is no bug here. It is just a matter of keeping the same output.
> * use single quote: since we changed the vartype from enum to string, we should
>  add single quotes. If you don't add single quotes, it works (for a single
>  default log level) -- as exercised in the tests. However, as soon as the user
>  adds an extra element it fails. This avoids confusion.
> 
> $ psql
> psql (19devel)
> Type "help" for help.
> 
> postgres=# show log_min_messages;
> log_min_messages 
> ------------------
> WARNING
> (1 row)
> 
> postgres=# set log_min_messages to debug1;
> SET
> postgres=# show log_min_messages;
> log_min_messages 
> ------------------
> debug1
> (1 row)
> 
> postgres=# set log_min_messages to 'debug1';
> SET
> postgres=# show log_min_messages;
> log_min_messages 
> ------------------
> debug1
> (1 row)
> 
> postgres=# set log_min_messages to warning, backend:error;
> ERROR:  syntax error at or near ":"
> LINE 1: set log_min_messages to warning, backend:error;
>                                                ^
> postgres=# set log_min_messages to 'warning, backend:error';
> SET
> postgres=# show log_min_messages;
>    log_min_messages    
> ------------------------
> warning, backend:error
> (1 row)
> 
> 
> -- 
> Euler Taveira
> EDB   https://www.enterprisedb.com/<0001-Minor-cleanups-in-log_min_messages.patch>

This cleanup patch looks good to me.

Best regards,
--
Chao Li (Evan)
HighGo Software Co., Ltd.
https://www.highgo.com/

Re: log_min_messages per backend type

On Mon, Feb 9, 2026, at 10:03 AM, Alvaro Herrera wrote:
> On 2026-Feb-09, Alvaro Herrera wrote:
>
>> I just pushed this, and somehow I forgot to squash this into the commit.
>> I don't think it matters terribly much though, so I'm going to hang onto
>> this for a while in case some bug is found.
>
> Euler mentioned offlist that I also forgot to remove
> log_min_messages_process_types from guc.h.  So that gives us this patch
> for now.
>

I expanded this patch a bit with fixes proposed by Man Zeng and Noriyoshi
Shinoda. Besides these changes, I noticed that

* default value in uppercase: since we changed the vartype from enum to string
  and enum maps DEBUG1 to "debug1" (see server_message_level_options), it
  should use the lowercase string to keep the same output as previous versions.
  There is no bug here. It is just a matter of keeping the same output.
* use single quote: since we changed the vartype from enum to string, we should
  add single quotes. If you don't add single quotes, it works (for a single
  default log level) -- as exercised in the tests. However, as soon as the user
  adds an extra element it fails. This avoids confusion.

$ psql
psql (19devel)
Type "help" for help.

postgres=# show log_min_messages;
 log_min_messages 
------------------
 WARNING
(1 row)

postgres=# set log_min_messages to debug1;
SET
postgres=# show log_min_messages;
 log_min_messages 
------------------
 debug1
(1 row)

postgres=# set log_min_messages to 'debug1';
SET
postgres=# show log_min_messages;
 log_min_messages 
------------------
 debug1
(1 row)

postgres=# set log_min_messages to warning, backend:error;
ERROR:  syntax error at or near ":"
LINE 1: set log_min_messages to warning, backend:error;
                                                ^
postgres=# set log_min_messages to 'warning, backend:error';
SET
postgres=# show log_min_messages;
    log_min_messages    
------------------------
 warning, backend:error
(1 row)


-- 
Euler Taveira
EDB   https://www.enterprisedb.com/

Re: log_min_messages per backend type

On Tue, Dec 9, 2025, at 11:00 PM, Chao Li wrote:
> Now “show log_min_messages” prints the raw string the user set, in 
> above example, there is not a white-space between the two log levels, 
> and “show” result doesn’t have a white-space between the two log levels 
> either. IMO, “SHOW log_min_messages” should display a stable result, in 
> other words, say “fatal, backend:log” and “backend:log, fatal” should 
> show the same result as they are actually meaning the same. So, I would 
> suggest normalize the raw string: put the general level in the first 
> place and sort others by process type, then SHOW returns the normalized 
> string.
>

I thought about it but leave it alone because (a) it would increase this patch
footprint and (b) the input might be different from the output. I could also be
done in another patch but under reflection an unstable output can break tests
or whatever uses the SHOW log_min_messages output. I thought this change would
require a new show_log_min_messages to manipulate the input again but we can
reassign the GUC value after sorting the existing list and creating a new string
list.

> In the “if” and “else” clauses, there are duplicate code to valid log 
> levels. We should refactor the code to avoid the duplication. For 
> example, pull up “loglevel” to the “for” loop level, then we can valid 
> it after the “if-else”.
>

The for loop is duplicate but if you create a separate function for it but the
result is:

 src/backend/commands/variable.c | 43 ++++++++++++++++++++++---------------------
 1 file changed, 22 insertions(+), 21 deletions(-)


I'll post a patch in a couple of hours after spend more time in it.


-- 
Euler Taveira
EDB   https://www.enterprisedb.com/

Re: log_min_messages per backend type

On Sat, Jan 17, 2026, at 7:02 PM, Alvaro Herrera wrote:
>
> I'm not opposed to this idea, but I think it should be implemented not
> by sorting and then moving the first element to the top of the list, but
> instead by modifying the cmp function so that the desired order is
> achieved directly.  So the cmp() should return -1 if element a has no
> colon, or 1 if element b has no colon, otherwise return per strcmp.
> That way you can remove the foreach() block above, which is icky.
>

Good idea. The v9 contains this implementation. I also use the StringInfo as
you suggested in the other email. The 0003 removed the wrong B_BACKEND
assignment pointed in a previous email. This new patchset contains some changes
suggested by Japin Li [1] (I didn't change the error message because it adds an
extra message to translate). There was a minor issue with a null-terminated
string that I fixed too.

PS> I didn't change the background worker case that I mentioned in [2].

[1] https://postgr.es/m/MEAPR01MB3031FA1986F3FC91481E28CCB68DA@MEAPR01MB3031.ausprd01.prod.outlook.com
[2] https://postgr.es/m/f20103f2-d960-4206-9969-c4a936524ab1@app.fastmail.com


-- 
Euler Taveira
EDB   https://www.enterprisedb.com/

Re: log_min_messages per backend type

On Tue, Dec 9, 2025, at 1:30 PM, Alvaro Herrera wrote:
> BTW another thing I realized while looking this over, is that we quite
> uselessly transform the integer backend type to a string, pass it as a
> string using the --forkchild= argument to the child process, then parse
> the string back to an int to use as an array index.  It would be much
> easier to just use the integer value everywhere, as the attached shows.
>

It is a good simplification. There is just one oversight.

+   child_type = (BackendType) atoi(child_kind);
+   if (child_type <= B_INVALID || child_type > BACKEND_NUM_TYPES)
        elog(ERROR, "unknown child kind %s", child_kind);

It should be BACKEND_NUM_TYPES - 1.


-- 
Euler Taveira
EDB   https://www.enterprisedb.com/

Re: log_min_messages per backend type

On Thu, Nov 6, 2025, at 1:01 PM, Alvaro Herrera wrote:
>
> I would use type:level rather than "backendtype".
> Also, the glossary says we have "auxiliary processes" and "backends", so
> from the user perspective, these types are not all backends, but instead
> process types.
>

Hasn't that ship already sailed? If your suggestion is limited to "type:level",
that's ok. I wouldn't like to use 2 terminologies ("process type" and "backend
type") in the documentation.

See miscadmin.h.

/*
 * MyBackendType indicates what kind of a backend this is.
 *
 * If you add entries, please also update the child_process_kinds array in
 * launch_backend.c.
 */
typedef enum BackendType

The "backend type" terminology is exposed. It is even available in the
pg_stat_activity view. I agree that "backend" is not a good name to define a
Postgres process. I don't think we should be inconsistent only in this patch.
Even if the proposal is to rename enum BackendType (I won't propose it as part
of this patch), it would make some extension authors unhappy according to
codesearch.debian.net.

> I think the list of backend types is pretty hard to read.  What do you
> think about using 
> 
> to create a friendlier representation?  
>

I tried but didn't like the result. See the attached image. You said table but
it is a multi-column list; it doesn't contain a header or borders. Reading this
message and Chao Li message, I realized the description is not good enough yet.

> So you would say something like "Valid types are listed in the table
> below, each corresponding to either postmaster, an auxiliary process
> type or a backend".  (Eh, wait, you seem not to have "postmaster" in
> your list, what's up with that?)
>

Good question. The current patch uses "backend" to B_INVALID (that's exactly the
MyBackendType for postmaster -- see below). I think it is reasonable to create a
new category "postmaster" and assign it to B_INVALID. If, for some reason, a
process temporarily has MyBackendType equals to B_INVALID, it is ok to still
consider it a postmaster process.

$ ps aux | grep postgres
euler      56968  0.0  0.0 217144 29016 ?        Ss   19:28   0:00 /c/pg/bin/postgres -D /c/pg/pgdata
euler      56969  0.0  0.0 217276  9948 ?        Ss   19:28   0:00 postgres: io worker 0
euler      56970  0.0  0.0 217276  7620 ?        Ss   19:28   0:00 postgres: io worker 1
euler      56971  0.0  0.0 217144  6408 ?        Ss   19:28   0:00 postgres: io worker 2
euler      56972  0.0  0.0 217276  8388 ?        Ss   19:28   0:00 postgres: checkpointer
euler      56973  0.0  0.0 217300  8744 ?        Ss   19:28   0:00 postgres: background writer
euler      56975  0.0  0.0 217276 11292 ?        Ss   19:28   0:00 postgres: walwriter
euler      56976  0.0  0.0 218724  9140 ?        Ss   19:28   0:00 postgres: autovacuum launcher
euler      56977  0.0  0.0 218724  9232 ?        Ss   19:28   0:00 postgres: logical replication launcher
euler      58717  0.0  0.0   6676  2232 pts/1    S+   19:39   0:00 grep --color=auto postgres
$ for p in $(pgrep postgres); do echo "pid: $p" && gdb -q -batch -ex 'set pagination off' -ex "attach $p" -ex 'p MyBackendType' -ex 'quit' 2> /dev/null; done | grep -E '^\$1|pid'
pid: 56968
$1 = B_INVALID
pid: 56969
$1 = B_IO_WORKER
pid: 56970
$1 = B_IO_WORKER
pid: 56971
$1 = B_IO_WORKER
pid: 56972
$1 = B_CHECKPOINTER
pid: 56973
$1 = B_BG_WRITER
pid: 56975
$1 = B_WAL_WRITER
pid: 56976
$1 = B_AUTOVAC_LAUNCHER
pid: 56977
$1 = B_BG_WORKER

> (I'm not sure about making the log levels also a , because
> for that list the order matters, and if we have more than one column
> then the order is going to be ambiguous, and if we have just one column
> it's going to be too tall.  But maybe it's not all that bad?)
>

+1.


-- 
Euler Taveira
EDB   https://www.enterprisedb.com/

Re: log_min_messages per backend type

> On Feb 9, 2026, at 20:50, Alvaro Herrera  wrote:
> 
> On 2026-Feb-09, Chao Li wrote:
> 
>> There’s also a related regression in the interactive case. Before this
>> patch, when debugging via psql, a superuser could easily do:
>> 
>> ```
>> # set log_min_messages = debug5
>> ```
>> 
>> to temporarily increase verbosity for the current backend. With the
>> new syntax, doing the equivalent now requires copying the entire
>> log_min_messages string and modifying just one part, which feels
>> noticeably heavier for quick debugging.
> 
> Nonsense.  You can still do the above, and it has the same behavior as
> before.
> 

Ah, yes, for the SET scenario, it sets the log level for all types, so the backend type gets the value, which ends up with the same behavior as before.

Best regards,
--
Chao Li (Evan)
HighGo Software Co., Ltd.
https://www.highgo.com/

Re: log_min_messages per backend type

On Wed, Nov 19, 2025, at 7:44 AM, Alvaro Herrera wrote:
> On 2025-Nov-18, Euler Taveira wrote:
>> The "backend type" terminology is exposed.
>
> Something appearing in the source code does not equate it being exposed
> to end users.  Things are exposed when they are in the documentation, or
> when the SQL interface requires you to use the term.  So I don't think
> the fact that BackendType appears in the code forces us to use that name
> in the user-visible interface now.
>

It is *already* exposed.

$ grep -B 3 -A 3 'backend type' doc/src/sgml/config.sgml 
         

         
          The backend type corresponds to the column
          backend_type in the view
          
          pg_stat_activity,
--
        character count of the error position therein,
        location of the error in the PostgreSQL source code
        (if log_error_verbosity is set to verbose),
        application name, backend type, process ID of parallel group leader,
        and query id.
        Here is a sample table definition for storing CSV-format log output:

$ grep -B 3 -A 3 'backend_type' doc/src/sgml/config.sgml 

         
          The backend type corresponds to the column
          backend_type in the view
          
          pg_stat_activity,
          but additional types can appear
--
  query_pos integer,
  location text,
  application_name text,
  backend_type text,
  leader_pid integer,
  query_id bigint,
  PRIMARY KEY (session_id, session_line_num)
--
         Client application name
        
        
         backend_type
         string
         Type of backend
        

$ psql -c "SELECT backend_type FROM pg_stat_activity" -d postgres
         backend_type         
------------------------------
 client backend
 autovacuum launcher
 logical replication launcher
 io worker
 io worker
 io worker
 checkpointer
 background writer
 walwriter
(9 rows)

> In the glossary, we talk about "processes"; in the definitions there,
> "backend" is one type of process.  So in this list of "backend types"
> that you want to introduce, what you really should be talking about is
> "process types", where "backend" is one of the options.
>
>> It is even available in the
>> pg_stat_activity view. I agree that "backend" is not a good name to define a
>> Postgres process. I don't think we should be inconsistent only in this patch.
>> Even if the proposal is to rename enum BackendType (I won't propose it as part
>> of this patch), it would make some extension authors unhappy according to
>> codesearch.debian.net.
>
> I don't think we should rename the BackendType enum.  That's in the
> source code, not in the documentation or the SQL interface, so we don't
> need to care.
>

I think you missed my point here. As I said if your proposal will let us with 2
terminologies (backend type and process type) as shown above.  We can debate if
we (1) keep the current terminology (backend type), (2) use the proposed one
(process type) or (3) use a mixed variation (keep the SQL interface --
backend_type column but change the description to "process type"). I wouldn't
like to break compatibility (pg_stat_activity changes tend to break a lot of
stuff) so I'm fine with (1) and (3).

>> > I think the list of backend types is pretty hard to read.  What do you
>> > think about using 
>> > 
>> > to create a friendlier representation?  
>> 
>> I tried but didn't like the result. See the attached image.
>
> Well, I like it very much.  The original is a solid wall of text, very
> hard to read, where you have to squint hunting commas in order to
> distinguish one item from the next.  (Maybe you are young and have good
> eyesight, but you won't be young forever.)  In the screenshot you show,
> the list of possible process types to use is nicely separated, which
> makes it very easy to catch at a glance.  Maybe remove "the table",
> which is obviously inappropriate, and just say "Valid process types are
> listed below, each corresponding to either postmaster, an auxiliary
> process type or a backend".
>

I don't have a strong preference. One advantage of this suggestion is that it 
is visually easier to identify the types.

> (Note your original wording says "a backend type, corresponding to [blah
> blah] or a backend" which makes no sense -- how is a backend a type of
> backend?  It isn't.  It's a type of process.)
>

I'm rewriting that paragraph.

>> Good question. The current patch uses "backend" to B_INVALID (that's exactly the
>> MyBackendType for postmaster -- see below). I think it is reasonable to create a
>> new category "postmaster" and assign it to B_INVALID.
>
> I guess that would work, but I think it's inadequate.  Maybe we could
> add a new value B_POSTMASTER and have postmaster switch to that as early
> as possible.  Then anything that still has B_INVALID must necessarily be
> an improperly identified process.  Users wouldn't assign a value to that
> one (the GUC wouldn't let you); instead those would always use the
> default value.  Hopefully nobody would see that very often, or at all.
>

I will create a separate patch for this suggestion.


-- 
Euler Taveira
EDB   https://www.enterprisedb.com/

Re: log_min_messages per backend type

On Sun, Oct 5, 2025, at 11:18 AM, Euler Taveira wrote:
> This new patch contains the following changes:
>
> - patch was rebased
> - use commit dbf8cfb4f02
> - use some GUC memory allocation functions
> - avoid one memory allocation (suggested by Japin Li)
> - rename backend type: logger -> syslogger
> - adjust tests to increase test coverage
> - improve documentation and comments to reflect the current state
>

Patch was rebased since commit fce7c73fba4 broke it. No modifications.


-- 
Euler Taveira
EDB   https://www.enterprisedb.com/

Re: log_min_messages per backend type

> On Feb 6, 2026, at 23:45, Euler Taveira  wrote:
> 
> On Fri, Feb 6, 2026, at 8:05 AM, Alvaro Herrera wrote:
>> Here are some fixups for the main patch.  I include yours so that CI
>> will test this one.
>> 
> 
> All suggestions seem good to me.
> 
>> * I'm not fond of the term "generic", so I changed it to "default",
>> which IMO makes more sense from the user point of view.
>> 
> 
> I'm fine with "default" instead of "generic". However, you forgot to replace
> "generic" in a few places. The attached patch fixes them all.
> 
>> * There seemed to be too much guc_mallocking[*] going on; for instance for
>> ptype.  I simplified that by just overwriting the : to a '\0', then we
>> can use the original string for pg_strcasecmp().  At the bottom of the
>> loop we restore the ':', so that the code below can reuse the original
>> values.  This is okay because we have already mallocked a copy for
>> SplitGUCList.
>> 
> 
> Nice trick.
> 
>> * There's no reason to add a WARNING to every process type in the
>> proctypelist.h table.  We can just set the values to WARNING in the
>> log_min_messages initializer in guc_tables.c.
>> 
> 
> Ok.
> 
>> * There's no reason AFAICS to have log_min_messages_process_types in
>> guc_tables.c; nobody else uses it.  I made it a local array of char*
>> inside the check_log_min_messages function.
>> 
> 
> That's true.
> 
>> * There's no need to initialize the newlevel[] array, since we're going
>> to assign values to every individual element at one point or another.
>> And for the assigned[] array, we don't need to set the value of each to
>> 'false' in a loop; we can just initialize to '{0}' and the compiler will
>> do the needful.  With this we can remove the loop entirely.
>> 
> 
> Ok.
> 
>> * "guc_free()+list_free()+return false" the whole time in the various
>> failure places across the loop was tiresome.  I put them all in a single
>> place and jump there with a goto.
>> 
> 
> I tend to avoid goto. However, in this case, it seems it improves readability.
> 
>> * in foreach() loops, tracking of a 'bool first' is unnecessary.  You
>> can just compare foreach_current_index() with 0.
>> 
> 
> Didn't know about foreach_current_index.
> 
>> * Rewrote docs and comments
>> 
> 
> Ok.
> 
> I attached the same patches that you shared plus the s/generic/default/ that I
> said earlier for the sake of CI.
> 
> 
> -- 
> Euler Taveira
> EDB   https://www.enterprisedb.com/

Hi Euler,

Thanks for updating the patch. I briefly played with v11 and overall it looks solid, but I ran into one usability concern.

While it’s reasonable to configure a full comma-separated list in postgresql.conf, it seems there’s currently no convenient way for a superuser to change the log level of a single process type. For example:

```
evantest=# alter system set log_min_messages='backend:debug4';
ERROR:  invalid value for parameter "log_min_messages": "backend:debug4"
DETAIL:  Default log level was not defined.
```

This feels a bit inconvenient in practice. Intuitively, it would be nice if setting a single type:level could be merged into the existing log_min_messages value, rather than requiring users to restate the entire configuration.

I dug into this a bit and noticed that, while check_log_min_messages() already normalizes and sorts the value (and could in theory merge with the existing setting), the current ALTER SYSTEM path doesn’t preserve the transformed value returned by the check hook. In AlterSystemSetConfigFile(), the newval produced by parse_and_validate_value() is effectively discarded for string GUCs:

```
	if (!parse_and_validate_value(record, value,
								  PGC_S_FILE, ERROR,
								  &newval, &newextra))
		ereport(ERROR,
				(errcode(ERRCODE_INVALID_PARAMETER_VALUE),
				 errmsg("invalid value for parameter \"%s\": \"%s\"",
						name, value)));

	if (record->vartype == PGC_STRING && newval.stringval != NULL)
		guc_free(newval.stringval);
```

So even if the check hook tried to be smarter, the infrastructure doesn’t currently allow it.

There’s also a related regression in the interactive case. Before this patch, when debugging via psql, a superuser could easily do:
```
# set log_min_messages = debug5
```

to temporarily increase verbosity for the current backend. With the new syntax, doing the equivalent now requires copying the entire log_min_messages string and modifying just one part, which feels noticeably heavier for quick debugging.

Do you think it would make sense to support setting a specific process type’s log level more directly (either via merging semantics or some other mechanism)?

Best regards,
--
Chao Li (Evan)
HighGo Software Co., Ltd.
https://www.highgo.com/

Re: log_min_messages per backend type

> On Dec 11, 2025, at 09:57, Euler Taveira  wrote:
> 
> 
>> In the “if” and “else” clauses, there are duplicate code to valid log 
>> levels. We should refactor the code to avoid the duplication. For 
>> example, pull up “loglevel” to the “for” loop level, then we can valid 
>> it after the “if-else”.
>> 
> 
> The for loop is duplicate but if you create a separate function for it but the
> result is:
> 
> src/backend/commands/variable.c | 43 ++++++++++++++++++++++---------------------
> 1 file changed, 22 insertions(+), 21 deletions(-)
> 
> 

I don’t think we need to add a separate function. We can use ‘if-else” to parse log level, then verify it after “if-else”.

Best regards,
--
Chao Li (Evan)
HighGo Software Co., Ltd.
https://www.highgo.com/

Re: log_min_messages per backend type

Re: log_min_messages per backend type

On Fri, Feb 6, 2026, at 8:05 AM, Alvaro Herrera wrote:
> Here are some fixups for the main patch.  I include yours so that CI
> will test this one.
>

All suggestions seem good to me.

> * I'm not fond of the term "generic", so I changed it to "default",
> which IMO makes more sense from the user point of view.
>

I'm fine with "default" instead of "generic". However, you forgot to replace
"generic" in a few places. The attached patch fixes them all.

> * There seemed to be too much guc_mallocking[*] going on; for instance for
> ptype.  I simplified that by just overwriting the : to a '\0', then we
> can use the original string for pg_strcasecmp().  At the bottom of the
> loop we restore the ':', so that the code below can reuse the original
> values.  This is okay because we have already mallocked a copy for
> SplitGUCList.
>

Nice trick.

> * There's no reason to add a WARNING to every process type in the
> proctypelist.h table.  We can just set the values to WARNING in the
> log_min_messages initializer in guc_tables.c.
>

Ok.

> * There's no reason AFAICS to have log_min_messages_process_types in
> guc_tables.c; nobody else uses it.  I made it a local array of char*
> inside the check_log_min_messages function.
>

That's true.

> * There's no need to initialize the newlevel[] array, since we're going
> to assign values to every individual element at one point or another.
> And for the assigned[] array, we don't need to set the value of each to
> 'false' in a loop; we can just initialize to '{0}' and the compiler will
> do the needful.  With this we can remove the loop entirely.
>

Ok.

> * "guc_free()+list_free()+return false" the whole time in the various
> failure places across the loop was tiresome.  I put them all in a single
> place and jump there with a goto.
>

I tend to avoid goto. However, in this case, it seems it improves readability.

> * in foreach() loops, tracking of a 'bool first' is unnecessary.  You
> can just compare foreach_current_index() with 0.
>

Didn't know about foreach_current_index.

> * Rewrote docs and comments
>

Ok.

I attached the same patches that you shared plus the s/generic/default/ that I
said earlier for the sake of CI.


-- 
Euler Taveira
EDB   https://www.enterprisedb.com/

Re: log_min_messages per backend type

Hi Euler,

I just reviewed the patch and got a few comments.

> On Nov 6, 2025, at 21:09, Euler Taveira  wrote:
> 
> On Sun, Oct 5, 2025, at 11:18 AM, Euler Taveira wrote:
>> This new patch contains the following changes:
>> 
>> - patch was rebased
>> - use commit dbf8cfb4f02
>> - use some GUC memory allocation functions
>> - avoid one memory allocation (suggested by Japin Li)
>> - rename backend type: logger -> syslogger
>> - adjust tests to increase test coverage
>> - improve documentation and comments to reflect the current state
>> 
> 
> Patch was rebased since commit fce7c73fba4 broke it. No modifications.
> 
> 
> -- 
> Euler Taveira
> EDB   https://www.enterprisedb.com/

1 - variable.c
```
+	/* Initialize the array. */
+	memset(newlevel, WARNING, BACKEND_NUM_TYPES * sizeof(int));
```

I think this statement is wrong, because memset() writes bytes but integers, so this statement will set very byte to WARNING, which should not be the intention. You will need to use a loop to initialize every element of newlevel.

2 - variable.c
```
+	/* Parse string into list of identifiers. */
+	if (!SplitGUCList(rawstring, ',', &elemlist))
+	{
+		/* syntax error in list */
+		GUC_check_errdetail("List syntax is invalid.");
+		guc_free(rawstring);
+		list_free(elemlist);
+		return false;
+	}
```

Every element of elemlist points to a position of rawstring, so it’s better to list_free(elemlist) first then guc_free(rawstring).

3 - launch_backend.c
```
 static inline bool
 should_output_to_server(int elevel)
 {
-	return is_log_level_output(elevel, log_min_messages);
+	return is_log_level_output(elevel, log_min_messages[MyBackendType]);
 }
```

Is it possible that when this function is called, MyBackendType has not been initialized? To be safe, maybe we can check if MyBackendType is 0 (B_INVALID), then use the generic log_min_message.

4 - config.sgml
```
+        Valid values are a comma-separated list of backendtype:level
+        and a single level. The list allows it to use
+        different levels per backend type. Only the single level
+        is mandatory (order does not matter) and it is assigned to the backend
+        types that are not specified in the list.
+        Valid backendtype values are archiver,
+        autovacuum, backend,
+        bgworker, bgwriter,
+        checkpointer, ioworker,
+        syslogger, slotsyncworker,
+        startup, walreceiver,
+        walsender, walsummarizer, and
+        walwriter.
+        Valid level values are DEBUG5,
+        DEBUG4, DEBUG3, DEBUG2,
+        DEBUG1, INFO, NOTICE,
+        WARNING, ERROR, LOG,
+        FATAL, and PANIC.  Each level includes
+        all the levels that follow it.  The later the level, the fewer messages are sent
+        to the log.  The default is WARNING for all backend types.
+        Note that LOG has a different rank here than in
```

* “Valid values are …”, here “are” usually mean both are needed, so maybe change “are” to “can be”.
* It says “the single level is mandatory”, then why there is still a default value?

Best regards,
--
Chao Li (Evan)
HighGo Software Co., Ltd.
https://www.highgo.com/

Re: log_min_messages per backend type

Hello,

Here's a couple of trivial cleanups of this code.  0001 simplifies code
and removes a duplicate by creating a small subroutine; 0002 also
simplifies memory freeing a tad, by freeing allocations as soon as
no longer needed instead of waiting and being forced to duplicate the
frees in several places.

This is two commits here just for ease of review, but I'd probably push
as one commit (and without backpatch).

-- 
Álvaro Herrera        Breisgau, Deutschland  —  https://www.EnterpriseDB.com/
"World domination is proceeding according to plan"        (Andrew Morton)

Re: log_min_messages per backend type

On Thu, Nov 6, 2025, at 11:53 AM, Chao Li wrote:
> I just reviewed the patch and got a few comments.
>

Thanks for your review.

> +	/* Initialize the array. */
> +	memset(newlevel, WARNING, BACKEND_NUM_TYPES * sizeof(int));
> ```
>
> I think this statement is wrong, because memset() writes bytes but 
> integers, so this statement will set very byte to WARNING, which should 
> not be the intention. You will need to use a loop to initialize every 
> element of newlevel.
>

Facepalm. Good catch.

> 2 - variable.c
> ```
> +	/* Parse string into list of identifiers. */
> +	if (!SplitGUCList(rawstring, ',', &elemlist))
> +	{
> +		/* syntax error in list */
> +		GUC_check_errdetail("List syntax is invalid.");
> +		guc_free(rawstring);
> +		list_free(elemlist);
> +		return false;
> +	}
> ```
>
> Every element of elemlist points to a position of rawstring, so it’s 
> better to list_free(elemlist) first then guc_free(rawstring).
>

Fixed.

> 3 - launch_backend.c
> ```
>  static inline bool
>  should_output_to_server(int elevel)
>  {
> -	return is_log_level_output(elevel, log_min_messages);
> +	return is_log_level_output(elevel, log_min_messages[MyBackendType]);
>  }
> ```
>
> Is it possible that when this function is called, MyBackendType has not 
> been initialized? To be safe, maybe we can check if MyBackendType is 0 
> (B_INVALID), then use the generic log_min_message.
>

It uses the generic log level if you don't modify backend type "backend". If
you do specify "backend", that level is used instead. After Alvaro's question
in the other email, I added "postmaster" backend type and assigned it to
B_INVALID. That's exactly the log level that will be used. As I said in that
email, it is ok to consider a process whose backend type is B_INVALID as a
postmaster process.

> * “Valid values are …”, here “are” usually mean both are needed, so 
> maybe change “are” to “can be”.
> * It says “the single level is mandatory”, then why there is still a 
> default value?
>

It seems the current sentence that describe the comma-separated list is
ambiguous. The sentence doesn't make it clear that the list contains 2 elements
(type:level and level) and the "level" element is mandatory. What about the new
sentence?


-- 
Euler Taveira
EDB   https://www.enterprisedb.com/

Re: log_min_messages per backend type

On Thu, Mar 6, 2025, at 10:33 AM, Andres Freund wrote:
> Huh, the startup process is among the most crucial things to monitor?
>

Good point. Fixed.

After collecting some suggestions, I'm attaching a new patch contains the
following changes:

- patch was rebased
- include Alvaro's patch (v2-0001) [1] as a basis for this patch
- add ioworker as new backend type
- add startup as new backend type per Andres suggestion
- small changes into documentation

> I don't know what I think about the whole patch, but I do want to voice
> *strong* opposition to duplicating a list of all backend types into multiple
> places. It's already painfull enough to add a new backend type, without having
> to pointlessly go around and manually add a new backend type to mulltiple
> arrays that have completely predictable content.
>

I'm including Alvaro's patch as is just to make the CF bot happy and to
illustrate how it would be if we adopt his solution to centralize the list of
backend types. I think Alvaro's proposal overcomes the objection [2], right?


[1] https://www.postgresql.org/message-id/202507282113.vdp4axehoppi@alvherre.pgsql
[2] https://www.postgresql.org/message-id/y5tgui75jrcj6mm5nmoq4yqwage2432akx4kp2ogtcnim3wskx@2ipmtfi4qvpi


-- 
Euler Taveira
EDB   https://www.enterprisedb.com/

Re: log_min_messages per backend type

Hi,

>
> On 2025-03-04 Tu 7:33 PM, Euler Taveira wrote:
>>> I think it should be acceptable to configure one global setting with
>>> exceptions for particular backend types:
>>>
>>> log_min_messages = WARNING, autovacuum:DEBUG1
>>>
>>> Right now I think the code only accepts the unadorned log level if
>>> there
>>> are no other items in the list.  I think the proposal downthread is to
>>> use the keyword ALL for this,
>>>
>>>   log_min_messages = all:WARNING, autovacuum:DEBUG1   # I don't
>>> like this
>>>
>>> but I think it's inferior, because then "all" is not really "all",
>>> and I
>>> think it would be different if I had said
>>>
>>>   log_min_messages = autovacuum:DEBUG1, all:WARNING   # I don't
>>> like this
>>>
>>> because it looks like the "all" entry should override the one I set
>>> for
>>> autovacuum before, which frankly would not make sense to me.
>>
>> Good point. After reflection, I agree that "all" is not a good keyword.
>> This patch turns backend type as optional so WARNING means apply this
>> log level as a final step to the backend types that are not
>> specified in
>> the list.
>>
>>> So I think these two lines,
>>>
>>> log_min_messages = WARNING, autovacuum:DEBUG1
>>> log_min_messages = autovacuum:DEBUG1, WARNING
>>>
>>> should behave identically and mean "set the level for autovacuum to
>>> DEBUG1, and to any other backend type to WARNING.
>>
>> Done.
>
>
> Just bikeshedding a bit ...
>
> I'm not mad keen on this design. I think the value should be either a
> single setting like "WARNING" or a set of type:setting pairs. I agree
> that "all" is a bad name, but I think "default" would make sense.
>
> I can live with it but I think this just looks a bit odd.
>

Just bringing some thoughts about it...

How about using something like *:WARNING? I'm not sure if it could also be
confusing as the "all" keyword, but I think it could also be interpreted as
"anything else use WARNING".

--
Matheus Alcantara

Re: log_min_messages per backend type

On Mon, Jul 20, 2026 at 9:15 PM Alvaro Herrera  wrote:
>
> Hello,
>
> Here's a couple of trivial cleanups of this code.  0001 simplifies code
> and removes a duplicate by creating a small subroutine; 0002 also
> simplifies memory freeing a tad, by freeing allocations as soon as
> no longer needed instead of waiting and being forced to duplicate the
> frees in several places.

+ rawstring_len = strlen(rawstring);
+ guc_free(rawstring);

SplitGUCList() stores pointers into rawstring in elemlist, so
after guc_free(rawstring), the subsequent foreach over elemlist,etc
still seem to dereference freed memory. So rawstring should remain
alive until we're done using all the elements of elemlist?

BTW, with the patches, I saw the following initdb failure:

    running bootstrap script ... LOG:  invalid value for parameter
"log_min_messages": "warning"
    DETAIL:  Unrecognized log level: "".
    FATAL:  failed to initialize log_min_messages to "warning"
    child process exited with exit code 1
    initdb: removing data directory "data"

Regards,

-- 
Fujii Masao

Re: log_min_messages per backend type

Re: log_min_messages per backend type

On Tue, Dec 9, 2025, at 3:24 PM, Alvaro Herrera wrote:
> On 2025-Dec-09, Alvaro Herrera wrote:
>
>> So here's your v6 again with those fixes as 0003 -- let's see what CI
>> thinks of this.  I haven't looked at your doc changes yet.
>
> This passed CI, so I have marked it as Ready for Committer.  Further
> comments are still welcome, of course, but if there are none, I intend
> to get it committed in a few days.
>

I took another look after Chao Li comments [1]. I created the 0003 patch
that does the sort as suggested. I think it is good to be consistent but
I'm fine if we decided the additional code is not worth. The 32 in the
MAX_LMM_STR_LEN is arbitrary but it is based on the size of the largest
element in the list ("dead-end client backend:warning"). I didn't take
into account the comma and space between elements but it is not
necessary since other elements are smaller than the largest one.
I didn't implement the 2nd suggestion.

I also merged Alvaro's fix to 0002. The v8 is attached.

I didn't change the commit message but if 0003 is merged into 0001 then
it should mention that

8<--------------------------------------------------------------------8<
The SHOW command presents well-formatted list sorted by process type and
the generic log level is the first element list. It improves readability
and has a clear indentation.
8<--------------------------------------------------------------------8<

Do we really need a different backend type in this case? For background
workers the description is "background worker". Shoundn't it use the
same description for this edge case too?

-       backend_type_str = MyBgworkerEntry->bgw_type;
+   {
+       if (MyBgworkerEntry && MyBgworkerEntry->bgw_type[0] != '\0')
+           backend_type_str = MyBgworkerEntry->bgw_type;
+       else
+           backend_type_str = "early bgworker";
+   }

I also noticed that commit 18d67a8d7d30 forgot to add gettext_noop to
the get_backend_type_for_log function. It should be consistent with
GetBackendTypeDesc() return.


[1] https://postgr.es/m/1130A10B-E7AE-49F3-9E13-2CD69B3F9DFD@gmail.com


-- 
Euler Taveira
EDB   https://www.enterprisedb.com/

RE: log_min_messages per backend type

Hi, 
> I just pushed this, and somehow I forgot to squash this into the commit.
Thanks to the developers and committers.

The attached small patch changes the process names in config.sgml to alphabetical order. 
postgresql.conf.sample already has that.

Regards,
Noriyoshi Shinoda

-----Original Message-----
From: zengman  
Sent: Monday, February 9, 2026 10:52 PM
To: Alvaro Herrera 
Cc: pgsql-hackers 
Subject: Re: log_min_messages per backend type

> Euler mentioned offlist that I also forgot to remove 
> log_min_messages_process_types from guc.h.  So that gives us this 
> patch for now.

Hi, 

I just noticed that `log_min_messages_process_types` is not used in the latest commit and should be removed.
I also noticed that `message level` in the comments below should be deleted as well.

```
diff --git a/src/include/postmaster/proctypelist.h b/src/include/postmaster/proctypelist.h
index 4e259e84c2d..229308ec137 100644
--- a/src/include/postmaster/proctypelist.h
+++ b/src/include/postmaster/proctypelist.h
@@ -25,8 +25,7 @@
  */
 
 /*
- * List of process types (symbol, category, description, Main function,
- * shmem_attach, message level) entries.
+ * List of process types (symbol, category, description, Main function, shmem_attach) entries.
  */
```

--
regards,
Man Zeng

Re: log_min_messages per backend type

On Thu, Jul 31, 2025 at 11:19:48AM -0300, Euler Taveira wrote:
> On Thu, Mar 6, 2025, at 10:33 AM, Andres Freund wrote:
> > Huh, the startup process is among the most crucial things to monitor?
> >
> 
> Good point. Fixed.
> 
> After collecting some suggestions, I'm attaching a new patch contains the
> following changes:
> 
> - patch was rebased
> - include Alvaro's patch (v2-0001) [1] as a basis for this patch
> - add ioworker as new backend type
> - add startup as new backend type per Andres suggestion
> - small changes into documentation
> 
> > I don't know what I think about the whole patch, but I do want to voice
> > *strong* opposition to duplicating a list of all backend types into multiple
> > places. It's already painfull enough to add a new backend type, without having
> > to pointlessly go around and manually add a new backend type to mulltiple
> > arrays that have completely predictable content.
> >
> 
> I'm including Alvaro's patch as is just to make the CF bot happy and to
> illustrate how it would be if we adopt his solution to centralize the list of
> backend types. I think Alvaro's proposal overcomes the objection [2], right?
> 

I think we can avoid memory allocation by using pg_strncasecmp().

diff --git a/src/backend/commands/variable.c b/src/backend/commands/variable.c
index 5c769dd7bcc..f854b2fac93 100644
--- a/src/backend/commands/variable.c
+++ b/src/backend/commands/variable.c
@@ -1343,14 +1343,10 @@ check_log_min_messages(char **newval, void **extra, GucSource source)
 		}
 		else
 		{
-			char	   *loglevel;
-			char	   *btype;
-			bool		found = false;

-			btype = palloc((sep - tok) + 1);
-			memcpy(btype, tok, sep - tok);
-			btype[sep - tok] = '\0';
-			loglevel = pstrdup(sep + 1);
+			char	   *btype = tok;
+			char	   *loglevel = sep + 1;
+			bool		found = false;

 			/* Is the log level valid? */
 			for (entry = server_message_level_options; entry && entry->name; entry++)
@@ -1377,7 +1373,7 @@ check_log_min_messages(char **newval, void **extra, GucSource source)
 			found = false;
 			for (int i = 0; i < BACKEND_NUM_TYPES; i++)
 			{
-				if (pg_strcasecmp(log_min_messages_backend_types[i], btype) == 0)
+				if (pg_strncasecmp(log_min_messages_backend_types[i], btype, sep - tok) == 0)
 				{
 					/* Reject duplicates for a backend type. */
 					if (assigned[i])

-- 
Best regards,
Japin Li
ChengDu WenWu Information Technology Co., LTD.

Re: log_min_messages per backend type

On Thu, 31 Jul 2025 at 11:19, "Euler Taveira"  wrote:
> On Thu, Mar 6, 2025, at 10:33 AM, Andres Freund wrote:
>> Huh, the startup process is among the most crucial things to monitor?
>>
>
> Good point. Fixed.
>
> After collecting some suggestions, I'm attaching a new patch contains the
> following changes:
>
> - patch was rebased
> - include Alvaro's patch (v2-0001) [1] as a basis for this patch
> - add ioworker as new backend type
> - add startup as new backend type per Andres suggestion
> - small changes into documentation
>
>> I don't know what I think about the whole patch, but I do want to voice
>> *strong* opposition to duplicating a list of all backend types into multiple
>> places. It's already painfull enough to add a new backend type, without having
>> to pointlessly go around and manually add a new backend type to mulltiple
>> arrays that have completely predictable content.
>>
>
> I'm including Alvaro's patch as is just to make the CF bot happy and to
> illustrate how it would be if we adopt his solution to centralize the list of
> backend types. I think Alvaro's proposal overcomes the objection [2], right?
>

If we set the log level for all backend types, I don't think a generic log
level is necessary.

-- 
Regards,
Japin Li
ChengDu WenWu Information Technology Co., Ltd.

Re: log_min_messages per backend type

On Thu, 31 Jul 2025 at 18:51, Alvaro Herrera  wrote:
> On 2025-Aug-01, Japin Li wrote:
>
>> If we set the log level for all backend types, I don't think a generic log
>> level is necessary.
>
> I don't understand what you mean by this.  Can you elaborate?
>
For example:

ALTER SYSTEM SET log_min_messages TO
'archiver:info, autovacuum:info, backend:info, bgworker:info, bgwriter:info, checkpointer:info, ioworker:info, logger:info, slotsyncworker:info, startup:info, walreceiver:info, walsender:info, walsummarizer:info, walwriter:info';

Given that we've set a log level for every backend type and
assigned[BACKEND_NUM_TYPES] is true, a generic level seems redundant.

-- 
Regards,
Japin Li
ChengDu WenWu Information Technology Co., Ltd.

Re: log_min_messages per backend type

Hi Euler,

On Thu, 15 Jan 2026 at 13:07, "Euler Taveira"  wrote:
> On Tue, Dec 9, 2025, at 3:24 PM, Alvaro Herrera wrote:
>> On 2025-Dec-09, Alvaro Herrera wrote:
>>
>>> So here's your v6 again with those fixes as 0003 -- let's see what CI
>>> thinks of this.  I haven't looked at your doc changes yet.
>>
>> This passed CI, so I have marked it as Ready for Committer.  Further
>> comments are still welcome, of course, but if there are none, I intend
>> to get it committed in a few days.
>>
>
> I took another look after Chao Li comments [1]. I created the 0003 patch
> that does the sort as suggested. I think it is good to be consistent but
> I'm fine if we decided the additional code is not worth. The 32 in the
> MAX_LMM_STR_LEN is arbitrary but it is based on the size of the largest
> element in the list ("dead-end client backend:warning"). I didn't take
> into account the comma and space between elements but it is not
> necessary since other elements are smaller than the largest one.
> I didn't implement the 2nd suggestion.
>
> I also merged Alvaro's fix to 0002. The v8 is attached.
>
> I didn't change the commit message but if 0003 is merged into 0001 then
> it should mention that
>
> 8<--------------------------------------------------------------------8<
> The SHOW command presents well-formatted list sorted by process type and
> the generic log level is the first element list. It improves readability
> and has a clear indentation.
> 8<--------------------------------------------------------------------8<
>
> Do we really need a different backend type in this case? For background
> workers the description is "background worker". Shoundn't it use the
> same description for this edge case too?
>
> -       backend_type_str = MyBgworkerEntry->bgw_type;
> +   {
> +       if (MyBgworkerEntry && MyBgworkerEntry->bgw_type[0] != '\0')
> +           backend_type_str = MyBgworkerEntry->bgw_type;
> +       else
> +           backend_type_str = "early bgworker";
> +   }
>
> I also noticed that commit 18d67a8d7d30 forgot to add gettext_noop to
> the get_backend_type_for_log function. It should be consistent with
> GetBackendTypeDesc() return.
>
>
> [1] https://postgr.es/m/1130A10B-E7AE-49F3-9E13-2CD69B3F9DFD@gmail.com
>
>

Thanks for updating the patch set.

Here are some comments.

1.
We can replace foreach with foreach_ptr in both v8-0001 and v8-0003.

2.
+/* log_min_messages */
+extern PGDLLIMPORT const char *const log_min_messages_process_types[];
+

Comment looks wrong.

3.
For cases where the process type is valid but the log level is unrecognized,
I suggest improving the error message for better clarity, e.g.:

	Unrecognized log level "bar" for process type "backend"

4.
The function name string_cmp feels too generic.
Could we consider a more specific name, for example list_log_min_message_cmp
or another more appropriate one?

> -- 
> Euler Taveira
> EDB   https://www.enterprisedb.com/
>
> [2. text/x-patch; v8-0001-log_min_messages-per-process-type.patch]...
>
> [3. text/x-patch; v8-0002-Assign-backend-type-earlier.patch]...
>
> [4. text/x-patch; v8-0003-fixup-log_min_messages-per-process-type.patch]...

-- 
Regards,
Japin Li
ChengDu WenWu Information Technology Co., Ltd.

Re: log_min_messages per backend type

> Euler mentioned offlist that I also forgot to remove
> log_min_messages_process_types from guc.h.  So that gives us this patch
> for now.

Hi, 

I just noticed that `log_min_messages_process_types` is not used in the latest commit and should be removed.
I also noticed that `message level` in the comments below should be deleted as well.

```
diff --git a/src/include/postmaster/proctypelist.h b/src/include/postmaster/proctypelist.h
index 4e259e84c2d..229308ec137 100644
--- a/src/include/postmaster/proctypelist.h
+++ b/src/include/postmaster/proctypelist.h
@@ -25,8 +25,7 @@
  */
 
 /*
- * List of process types (symbol, category, description, Main function,
- * shmem_attach, message level) entries.
+ * List of process types (symbol, category, description, Main function, shmem_attach) entries.
  */
```

--
regards,
Man Zeng

Re: log_min_messages per backend type

Hi,

On 2025-03-04 21:33:39 -0300, Euler Taveira wrote:
> +/*
> + * This must match enum BackendType! It should be static, but
> + * commands/variable.c needs to get at this.
> + */
> +int			log_min_messages[] = {
> +	[B_INVALID] = WARNING,
> +	[B_BACKEND] = WARNING,
> +	[B_DEAD_END_BACKEND] = WARNING,
> +	[B_AUTOVAC_LAUNCHER] = WARNING,
> +	[B_AUTOVAC_WORKER] = WARNING,
> +	[B_BG_WORKER] = WARNING,
> +	[B_WAL_SENDER] = WARNING,
> +	[B_SLOTSYNC_WORKER] = WARNING,
> +	[B_STANDALONE_BACKEND] = WARNING,
> +	[B_ARCHIVER] = WARNING,
> +	[B_BG_WRITER] = WARNING,
> +	[B_CHECKPOINTER] = WARNING,
> +	[B_STARTUP] = WARNING,
> +	[B_WAL_RECEIVER] = WARNING,
> +	[B_WAL_SUMMARIZER] = WARNING,
> +	[B_WAL_WRITER] = WARNING,
> +	[B_LOGGER] = WARNING,
> +};

> +StaticAssertDecl(lengthof(log_min_messages) == BACKEND_NUM_TYPES,
> +				 "array length mismatch");
> +
> +/*
> + * This must match enum BackendType! It might be in commands/variable.c but for
> + * convenience it is near log_min_messages.
> + */
> +const char *const log_min_messages_backend_types[] = {
> +	[B_INVALID] = "backend",	/* XXX same as backend? */
> +	[B_BACKEND] = "backend",
> +	[B_DEAD_END_BACKEND] = "backend",	/* XXX same as backend? */
> +	[B_AUTOVAC_LAUNCHER] = "autovacuum",
> +	[B_AUTOVAC_WORKER] = "autovacuum",
> +	[B_BG_WORKER] = "bgworker",
> +	[B_WAL_SENDER] = "walsender",
> +	[B_SLOTSYNC_WORKER] = "slotsyncworker",
> +	[B_STANDALONE_BACKEND] = "backend", /* XXX same as backend? */
> +	[B_ARCHIVER] = "archiver",
> +	[B_BG_WRITER] = "bgwriter",
> +	[B_CHECKPOINTER] = "checkpointer",
> +	[B_STARTUP] = "backend",	/* XXX same as backend? */

Huh, the startup process is among the most crucial things to monitor?

> +	[B_WAL_RECEIVER] = "walreceiver",
> +	[B_WAL_SUMMARIZER] = "walsummarizer",
> +	[B_WAL_WRITER] = "walwriter",
> +	[B_LOGGER] = "logger",
> +};
> +
> +StaticAssertDecl(lengthof(log_min_messages_backend_types) == BACKEND_NUM_TYPES,
> +				 "array length mismatch");
> +

I don't know what I think about the whole patch, but I do want to voice
*strong* opposition to duplicating a list of all backend types into multiple
places. It's already painfull enough to add a new backend type, without having
to pointlessly go around and manually add a new backend type to mulltiple
arrays that have completely predictable content.

Greetings,

Andres Freund