public inbox for [email protected]  
help / color / mirror / Atom feed
From: Jim Jones <[email protected]>
To: Fujii Masao <[email protected]>
Cc: Maxym Kharchenko <[email protected]>
Cc: Kirill Reshke <[email protected]>
Cc: Álvaro Herrera <[email protected]>
Cc: Fujii Masao <[email protected]>
Cc: Kirill Gavrilov <[email protected]>
Cc: Andrey M. Borodin <[email protected]>
Cc: Euler Taveira <[email protected]>
Cc: [email protected]
Subject: Re: Truncate logs by max_log_size
Date: Thu, 9 Jul 2026 09:41:57 +0200
Message-ID: <[email protected]> (raw)
In-Reply-To: <CAHGQGwEQgFn9UF3wG+LXszM=HMLm6nkEv-Jg2uX7M9M5VLKYkQ@mail.gmail.com>
References: <[email protected]>
	<CAHGQGwEc5Dn5KFU04sDp0Kyai2Te5yTuni3QW7pc2a3=343tJQ@mail.gmail.com>
	<[email protected]>
	<CAHGQGwGVxV1YbtLCT+q1yHejBKjLz7L+8r3WJMW=72SDJFW27g@mail.gmail.com>
	<CACsxBjB5+HLw6jJ17qR5vSg+kBb2G5ACSbd_ytZ0HvvD22JqkQ@mail.gmail.com>
	<CAHGQGwFYxCVmL4Bvm61bUW+WUi+YXFHzXo9hFEoSgYT1=fU-+w@mail.gmail.com>
	<[email protected]>
	<[email protected]>
	<[email protected]>
	<CAHGQGwHJHCi=4zEyMc4GJWPC9-oQZHj+9kncxS2442V9PXGgoQ@mail.gmail.com>
	<[email protected]>
	<CAHGQGwFOV+7nOdfoO=kfVH=-fRA9aQE1YcHHLYty3nfQ9rQ4RA@mail.gmail.com>
	<[email protected]>
	<CAHGQGwHhtcMH2YY9bo=qkZWLDkN7DBiHJ0g2Xja8eb6OfVkVUA@mail.gmail.com>
	<[email protected]>
	<CAHGQGwEQgFn9UF3wG+LXszM=HMLm6nkEv-Jg2uX7M9M5VLKYkQ@mail.gmail.com>


Thanks for the thorough review!

On 09/07/2026 06:03, Fujii Masao wrote:
> - errdetail("prepare: %s", pstmt->plansource->query_string);
> + char *truncated_stmt = truncate_query_log(pstmt->plansource->query_string);
> +
> + errdetail("prepare: %s", truncated_stmt ? truncated_stmt :
> pstmt->plansource->query_string);
> 
> Shouldn't truncated_stmt be freed with pfree() (when non-NULL)
> immediately after errdetail()?


Yes, I missed that one.
Fixed.


> -        is truncated to at most this many bytes.
> +        is truncated to at most this many bytes. When a statement is
> +        truncated, an ellipsis (<literal>...</literal>) is appended to
> +        indicate that truncation has occurred.
> 
> Since an ellipsis is appended, the logged output can exceed the configured
> limit by three bytes. Wouldn't it be clearer to say something like
> "has its statement text truncated to at most this many bytes"?
> That makes it clear the limit applies to the statement text, not the entire
> logged message.

Agreed.
I added your suggestion.

> -ok($node->log_contains(qr/statement: SELECT '123456789ABC$/m, $log_offset),
> +ok($node->log_contains(qr/statement: SELECT '123456789ABC...$/m, $log_offset),
> 
> Shouldn't the dots be escaped? As written, "..." matches any three characters.
> I think it should be "\.\.\.".


At first I also thought so, as in this context ... could mean any three
arbitrary characters, but the tests results showed a different picture,
so I thought it was not necessary. Anyway, I escaped them now in v2.


>     # a 24-byte query should end at the 20th byte ('C').
> 
> Since an ellipsis is now appended, isn't it better to replace this
> with "a 24-byte query should be clipped at the 20th byte ('C')
> and followed by an ellipsis."?
> 
> 
>     # With log_statement_max_length = 20, a 24-byte query should end
>     # at the 20th byte ('C').
> 
> Similarly, isn't it better to replace this with
> "With log_statement_max_length = 20, a 24-byte query should be clipped
> at the 20th byte ('C'). and followed by an ellipsis."?
>> * appending ellipsis to the query to indicate truncation (docs and tests
>> also updated accordingly). A side effect of this change is that setting
>> the parameter to 0 logs an ellipsis, which is not zero in length, but is
>> arguably correct, as it indicates that the whole query has been
>> truncated - just noting.
> So, should we update the docs and comments accordingly? For example:
> 
>  --------------------------------------------
> -        A value of zero causes statements to be logged with an empty body.
> +        A value of zero causes statements to be logged as only an ellipsis.
>          <literal>-1</literal> (the default) logs statements in full.
>          If this value is specified without units, it is taken as bytes.
> 
> { name => 'log_statement_max_length', type => 'int', context =>
> 'PGC_SUSET', group => 'LOGGING_WHAT',
> -  short_desc => 'Sets the maximum length in bytes of logged statements.',
> -  long_desc => '-1 means log statement in full; 0 means log an empty
> statement body.',
> +  short_desc => 'Sets the maximum length in bytes of logged statement text.',
> +  long_desc => '-1 means log statement in full; 0 means log only an ellipsis.',
> 
> - #log_statement_max_length = -1          # max bytes of logged statements;
> + #log_statement_max_length = -1          # max bytes of logged statement text;
>                                          # -1 means log statement in full,
> -                                        # 0 means log an empty body
> +                                        # 0 means log only an ellipsis
> 
> -# Verify 0 logs an empty statement body.
> +# Verify 0 logs only an ellipsis.
>  note "Zero length truncation";
> 
> ok($node->log_contains(qr/statement: \.\.\.\s*$/m, $log_offset),
> -       "0 logs statement body with ellipsis");
> +       "0 logs only an ellipsis");
>  --------------------------------------------


Comments and docs updated.


>> BTW, should I open a new CF entry for this?
> Yes, I think that's a good idea so these follow-up patches don't get forgotten.

Done: https://commitfest.postgresql.org/patch/6998/

Best, Jim


Attachments:

  [text/x-patch] v2-0001-Improve-log_statement_max_length-truncation.patch (9.1K, ../[email protected]/2-v2-0001-Improve-log_statement_max_length-truncation.patch)
  download | inline diff:
From 39072fa5bff8e23d4c56604d45635f835a0c78ad Mon Sep 17 00:00:00 2001
From: Jim Jones <[email protected]>
Date: Thu, 9 Jul 2026 09:25:32 +0200
Subject: [PATCH v2 1/2] Improve log_statement_max_length truncation

Apply log_statement_max_length consistently when logging prepared
statements. Previously, the prepared query emitted in the DETAIL
message for EXECUTE was not truncated, even though the corresponding
statement logged by log_statement was.

Also append an ellipsis to truncated statements, so it is immediately
apparent when a statement has been truncated.

This also avoid scanning the entire query string when determining
if truncation is needed by using strnlen() instead of strlen() in
truncate_query_log().

Author: Jim Jones <[email protected]>
Co-authored-by: Fujii Masao <[email protected]>
---
 doc/src/sgml/config.sgml                      |  6 ++-
 src/backend/tcop/postgres.c                   | 20 +++++---
 src/backend/utils/misc/postgresql.conf.sample |  4 +-
 .../t/014_log_statement_max_length.pl         | 46 ++++++++++++++-----
 4 files changed, 54 insertions(+), 22 deletions(-)

diff --git a/doc/src/sgml/config.sgml b/doc/src/sgml/config.sgml
index c67130c620e..0848c18d329 100644
--- a/doc/src/sgml/config.sgml
+++ b/doc/src/sgml/config.sgml
@@ -8539,8 +8539,10 @@ log_line_prefix = '%m [%p] %q%u@%d/%a '
         <xref linkend="guc-log-min-duration-statement"/>,
         <xref linkend="guc-log-min-duration-sample"/>, or
         <xref linkend="guc-log-transaction-sample-rate"/>
-        is truncated to at most this many bytes.
-        A value of zero causes statements to be logged with an empty body.
+        has its statement text truncated to at most this many bytes.
+        When a statement is truncated, an ellipsis (<literal>...</literal>)
+        is appended to indicate that truncation has occurred.
+        A value of zero causes statements to be logged as only an ellipsis.
         <literal>-1</literal> (the default) logs statements in full.
         If this value is specified without units, it is taken as bytes.
         This setting does not affect statements logged because of
diff --git a/src/backend/tcop/postgres.c b/src/backend/tcop/postgres.c
index ce18df820cd..019e4ab42fe 100644
--- a/src/backend/tcop/postgres.c
+++ b/src/backend/tcop/postgres.c
@@ -2553,8 +2553,9 @@ check_log_duration(char *msec_str, bool was_logged)
  * truncate_query_log
  *		Truncate query string if needed for logging
  *
- * Returns a palloc'd truncated copy if truncation is needed,
- * or NULL if no truncation is required.
+ * Returns a palloc'd copy of the query truncated for logging, with an
+ * ellipsis appended if truncation occurs, or NULL if no truncation is
+ * required.
  */
 static char *
 truncate_query_log(const char *query)
@@ -2567,7 +2568,7 @@ truncate_query_log(const char *query)
 	if (!query || log_statement_max_length < 0)
 		return NULL;
 
-	query_len = strlen(query);
+	query_len = strnlen(query, (size_t) log_statement_max_length + MAX_MULTIBYTE_CHAR_LEN);
 
 	/*
 	 * No need to allocate a truncated copy if the query is shorter than
@@ -2578,9 +2579,10 @@ truncate_query_log(const char *query)
 
 	/* Truncate at a multibyte character boundary */
 	truncated_len = pg_mbcliplen(query, query_len, log_statement_max_length);
-	truncated_query = (char *) palloc(truncated_len + 1);
+	truncated_query = (char *) palloc(truncated_len + 4);
 	memcpy(truncated_query, query, truncated_len);
-	truncated_query[truncated_len] = '\0';
+	memcpy(truncated_query + truncated_len, "...", 3);
+	truncated_query[truncated_len + 3] = '\0';
 
 	return truncated_query;
 }
@@ -2608,7 +2610,13 @@ errdetail_execute(List *raw_parsetree_list)
 			pstmt = FetchPreparedStatement(stmt->name, false);
 			if (pstmt)
 			{
-				errdetail("prepare: %s", pstmt->plansource->query_string);
+				char *truncated_stmt = truncate_query_log(pstmt->plansource->query_string);
+
+				errdetail("prepare: %s", truncated_stmt ? truncated_stmt : pstmt->plansource->query_string);
+
+				if (truncated_stmt != NULL)
+					pfree(truncated_stmt);
+
 				return 0;
 			}
 		}
diff --git a/src/backend/utils/misc/postgresql.conf.sample b/src/backend/utils/misc/postgresql.conf.sample
index 47e1221f3c3..7958653077b 100644
--- a/src/backend/utils/misc/postgresql.conf.sample
+++ b/src/backend/utils/misc/postgresql.conf.sample
@@ -676,9 +676,9 @@
                                         # bind-parameter values to N bytes;
                                         # -1 means print in full, 0 disables
 #log_statement = 'none'                 # none, ddl, mod, all
-#log_statement_max_length = -1          # max bytes of logged statements;
+#log_statement_max_length = -1          # max bytes of logged statement text;
                                         # -1 means log statement in full,
-                                        # 0 means log an empty body
+                                        # 0 means log only an ellipsis
 #log_replication_commands = off
 #log_temp_files = -1                    # log temporary files equal or larger
                                         # than the specified size in kilobytes;
diff --git a/src/test/modules/test_misc/t/014_log_statement_max_length.pl b/src/test/modules/test_misc/t/014_log_statement_max_length.pl
index b1ce6068f5d..b879566ef54 100644
--- a/src/test/modules/test_misc/t/014_log_statement_max_length.pl
+++ b/src/test/modules/test_misc/t/014_log_statement_max_length.pl
@@ -15,14 +15,15 @@ $node->init();
 $node->start;
 
 # Verify ASCII truncation. With log_statement_max_length = 20,
-# a 24-byte query should end at the 20th byte ('C').
+# a 24-byte query should be clipped at the 20th byte ('C') and
+# followed by an ellipsis.
 note "ASCII truncation via log_statement";
 my $log_offset = -s $node->logfile;
 $node->psql(
 	'postgres', "
 	SET log_statement_max_length TO 20;
 	SELECT '123456789ABCDEF';");
-ok($node->log_contains(qr/statement: SELECT '123456789ABC$/m, $log_offset),
+ok($node->log_contains(qr/statement: SELECT '123456789ABC\.\.\.$/m, $log_offset),
 	"ASCII query truncated at 20 bytes");
 
 # Verify -1 logs statement in full (closing quote must be present).
@@ -51,23 +52,23 @@ SKIP:
 		SET client_encoding TO 'UTF8';
 		SET log_statement_max_length TO 11;
 		$mbquery");
-	ok($node->log_contains(qr/statement: SELECT 'AA$/m, $log_offset),
+	ok($node->log_contains(qr/statement: SELECT 'AA\.\.\.$/m, $log_offset),
 		"multibyte truncation at character boundary");
 }
 
-# Verify 0 logs an empty statement body.
+# Verify 0 logs only an ellipsis.
 note "Zero length truncation";
 $log_offset = -s $node->logfile;
 $node->psql(
 	'postgres', "
 	SET log_statement_max_length TO 0;
 	SELECT '123456789ABCDEF';");
-ok($node->log_contains(qr/statement:\s*$/m, $log_offset),
-	"0 logs an empty statement body");
+ok($node->log_contains(qr/statement: \.\.\.\s*$/m, $log_offset),
+	"0 logs statement body with only an ellipsis");
 
 # Verify truncation via the extended query protocol (execute message).
-# With log_statement_max_length = 20, a 24-byte query should end
-# at the 20th byte ('C').
+# With log_statement_max_length = 20, a 24-byte query should be clipped
+# at the 20th byte ('C') and followed by an ellipsis.
 note "Extended query protocol (execute) truncation";
 $log_offset = -s $node->logfile;
 $node->psql(
@@ -75,7 +76,7 @@ $node->psql(
 	SET log_statement_max_length TO 20;
 	SELECT '123456789ABCDEF' \\bind \\g");
 ok( $node->log_contains(
-		qr/execute <unnamed>: SELECT '123456789ABC$/m, $log_offset),
+		qr/execute <unnamed>: SELECT '123456789ABC\.\.\.$/m, $log_offset),
 	"extended protocol execute truncated at 20 bytes");
 
 # Verify extended protocol also respects -1 (no truncation; closing quote
@@ -102,14 +103,35 @@ $node->psql(
 	SET log_statement_max_length TO 20;
 	SELECT '123456789ABCDEF' \\bind \\g");
 ok( $node->log_contains(
-		qr/parse <unnamed>: SELECT '123456789ABC$/m, $log_offset),
+		qr/parse <unnamed>: SELECT '123456789ABC\.\.\.$/m, $log_offset),
 	"parse duration entry truncated");
 ok( $node->log_contains(
-		qr/bind <unnamed>: SELECT '123456789ABC$/m, $log_offset),
+		qr/bind <unnamed>: SELECT '123456789ABC\.\.\.$/m, $log_offset),
 	"bind duration entry truncated");
 ok( $node->log_contains(
-		qr/execute <unnamed>: SELECT '123456789ABC$/m, $log_offset),
+		qr/execute <unnamed>: SELECT '123456789ABC\.\.\.$/m, $log_offset),
 	"execute duration entry truncated");
 
+note "Truncate prepared statement query in DETAIL";
+$log_offset = -s $node->logfile;
+$node->psql(
+	'postgres', "
+	SET log_statement_max_length TO 12;
+	PREPARE stmt AS SELECT * FROM pg_hba_file_rules WHERE address = \$1;
+	EXECUTE stmt('127.0.0.1');");
+ok( $node->log_contains(
+		qr/prepare: PREPARE stmt\.\.\.$/m, $log_offset),
+	"Truncate prepared statement query in DETAIL");
+
+note "Truncate prepared statement query in DETAIL (0 length)";
+$log_offset = -s $node->logfile;
+$node->psql(
+	'postgres', "
+	SET log_statement_max_length TO 0;
+	PREPARE stmt AS SELECT * FROM pg_hba_file_rules WHERE address = \$1;
+	EXECUTE stmt('127.0.0.1');");
+ok( $node->log_contains(
+		qr/prepare: \.\.\.$/m, $log_offset),
+	"0 logs the prepared statement body with only an ellipsis");
 $node->stop;
 done_testing();
-- 
2.55.0



  [text/x-patch] v2-0002-Simplify-callers-of-truncate_query_log.patch (3.2K, ../[email protected]/3-v2-0002-Simplify-callers-of-truncate_query_log.patch)
  download | inline diff:
From 4c1fb8acd321dea4830b96d1518b1ffd4474b164 Mon Sep 17 00:00:00 2001
From: Jim Jones <[email protected]>
Date: Wed, 8 Jul 2026 18:31:27 +0200
Subject: [PATCH v2 2/2] Simplify callers of truncate_query_log()

truncate_query_log() already returns NULL when query logging is disabled
or when no truncation is needed. Callers therefore do not need to check
log_statement_max_length before invoking it.

Remove the redundant checks and rely on truncate_query_log() to decide
whether a truncated copy should be returned. This is purely a code
cleanup with no behavioral change.

Author: Jim Jones <[email protected]>
---
 src/backend/tcop/postgres.c | 30 ++++++------------------------
 1 file changed, 6 insertions(+), 24 deletions(-)

diff --git a/src/backend/tcop/postgres.c b/src/backend/tcop/postgres.c
index 019e4ab42fe..b966cb242b7 100644
--- a/src/backend/tcop/postgres.c
+++ b/src/backend/tcop/postgres.c
@@ -1085,10 +1085,7 @@ exec_simple_query(const char *query_string)
 	/* Log immediately if dictated by log_statement */
 	if (check_log_statement(parsetree_list))
 	{
-		char	   *truncated_stmt = NULL;
-
-		if (log_statement_max_length >= 0)
-			truncated_stmt = truncate_query_log(query_string);
+		char	   *truncated_stmt = truncate_query_log(query_string);
 
 		ereport(LOG,
 				(errmsg("statement: %s",
@@ -1392,10 +1389,7 @@ exec_simple_query(const char *query_string)
 			break;
 		case 2:
 			{
-				char	   *truncated_stmt = NULL;
-
-				if (log_statement_max_length >= 0)
-					truncated_stmt = truncate_query_log(query_string);
+				char	   *truncated_stmt = truncate_query_log(query_string);
 
 				ereport(LOG,
 						(errmsg("duration: %s ms  statement: %s",
@@ -1638,10 +1632,7 @@ exec_parse_message(const char *query_string,	/* string to execute */
 			break;
 		case 2:
 			{
-				char	   *truncated_stmt = NULL;
-
-				if (log_statement_max_length >= 0)
-					truncated_stmt = truncate_query_log(query_string);
+				char	   *truncated_stmt = truncate_query_log(query_string);
 
 				ereport(LOG,
 						(errmsg("duration: %s ms  parse %s: %s",
@@ -2125,10 +2116,7 @@ exec_bind_message(StringInfo input_message)
 			break;
 		case 2:
 			{
-				char	   *truncated_stmt = NULL;
-
-				if (log_statement_max_length >= 0)
-					truncated_stmt = truncate_query_log(psrc->query_string);
+				char	   *truncated_stmt = truncate_query_log(psrc->query_string);
 
 				ereport(LOG,
 						(errmsg("duration: %s ms  bind %s%s%s: %s",
@@ -2282,10 +2270,7 @@ exec_execute_message(const char *portal_name, long max_rows)
 	/* Log immediately if dictated by log_statement */
 	if (check_log_statement(portal->stmts))
 	{
-		char	   *truncated_source = NULL;
-
-		if (log_statement_max_length >= 0)
-			truncated_source = truncate_query_log(sourceText);
+		char	   *truncated_source = truncate_query_log(sourceText);
 
 		ereport(LOG,
 				(errmsg("%s %s%s%s: %s",
@@ -2414,10 +2399,7 @@ exec_execute_message(const char *portal_name, long max_rows)
 			break;
 		case 2:
 			{
-				char	   *truncated_source = NULL;
-
-				if (log_statement_max_length >= 0)
-					truncated_source = truncate_query_log(sourceText);
+				char	   *truncated_source = truncate_query_log(sourceText);
 
 				ereport(LOG,
 						(errmsg("duration: %s ms  %s %s%s%s: %s",
-- 
2.55.0



view thread (15+ messages)

reply

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Reply to all the recipients using the --to and --cc options:
  reply via email

  To: [email protected]
  Cc: [email protected], [email protected], [email protected], [email protected], [email protected], [email protected], [email protected], [email protected], [email protected], [email protected]
  Subject: Re: Truncate logs by max_log_size
  In-Reply-To: <[email protected]>

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

This inbox is served by agora; see mirroring instructions
for how to clone and mirror all data and code used for this inbox