public inbox for [email protected]
help / color / mirror / Atom feedFrom: Ayush Tiwari <[email protected]>
To: Ashutosh Bapat <[email protected]>
To: Peter Eisentraut <[email protected]>
Cc: SATYANARAYANA NARLAPURAM <[email protected]>
Cc: PostgreSQL Hackers <[email protected]>
Subject: Re: [Bug] Add the missing RTE_GRAPH_TABLE case to transformLockingClause()
Date: Wed, 10 Jun 2026 01:03:45 +0530
Message-ID: <CAJTYsWX8J2cUTCnVRwks0gJw4wiMCv_qVSxshFC9f8oh7Ch6JA@mail.gmail.com> (raw)
In-Reply-To: <CAExHW5sQ5CfQeO4QXQaArRpD5FrUK_ndDAqHiQFg04DaC-+NaQ@mail.gmail.com>
References: <CAHg+QDcE9wp6nOEC3SCRQ90nrCO=QF+OZq1MG8Qc6hnusmogqw@mail.gmail.com>
<CAExHW5sQ5CfQeO4QXQaArRpD5FrUK_ndDAqHiQFg04DaC-+NaQ@mail.gmail.com>
Hi,
On Wed, 13 May 2026 at 10:57, Ashutosh Bapat <[email protected]>
wrote:
> > Repro:
> >
> > CREATE TABLE v(id int PRIMARY KEY, vname text);
> > CREATE PROPERTY GRAPH g VERTEX TABLES (v);
> > SELECT * FROM GRAPH_TABLE(g MATCH (a) COLUMNS (a.vname)) gt
> > FOR UPDATE OF gt;
> > -- ERROR: unrecognized RTE type: 8
> >
> > Attached a patch that returns ERRCODE_FEATURE_NOT_SUPPORTED "FOR ...
> cannot be
> > applied to a GRAPH_TABLE" with a position pointer, matching the
> convention used by
> > the function/tablefunc etc.
>
> In a fully matured SQL/PGQ support, a row mark on GRAPH_TABLE should
> be pushed down to the resulting subqueries in rewriteGraphTable(). I
> see this similar to how the row mark is pushed down into views during
> the rewrite phase. I don't think the code changes will be massive, it
> will be a matter of invoking markQueryForLocking() on the query that
> replaces the graph pattern in rewriteGraphTable(). I am not sure
> whether we want to do that now when stabilizing the feature. Peter,
> what do you think?
>
I wanted to implement this by trying to push FOR UPDATE/SHARE
down into the relational tables underlying the GRAPH_TABLE (the
"fully matured SQL/PGQ" direction mentioned). But had
some doubts around the implementation.
I'd like input on a few design questions from you/Peter/hackers:
Q1. Is the right place to do the pushdown rewriteGraphTable(),
mirroring what ApplyRetrieveRule() does for views (i.e. call
markQueryForLocking() on the freshly built subquery)? That
would require exposing markQueryForLocking() outside
rewriteHandler.c.
Q2. How should FOR UPDATE behave when the MATCH expands to more
than one path and rewriteGraphTable() produces a UNION?
markQueryForLocking() refuses to descend into set-operations,
same as it does for views. Options:
(a) reject FOR UPDATE on multi-path MATCH with a clear
"cannot lock rows from a set-operation result" error,
matching the view behaviour;
(b) silently fall back to the current
"raise relation lock only" behaviour and document it.
I think it should be (a).
Q3. SQL/PGQ does not really define UPDATE/DELETE semantics for
graph paths. Is the intent here that FOR UPDATE/SHARE on
GRAPH_TABLE is purely a concurrency hint rather than a
precursor to UPDATE WHERE CURRENT OF? If so, that should
probably be documented next to the GRAPH_TABLE syntax in
the SQL reference.
> If we decide to not to support it, the code changes look on the right
> path. We need to update the following comment which appears earlier in
> the function to mention GRAPH_TABLE, /* ignore JOIN, SPECIAL,
> FUNCTION, VALUES, CTE RTEs */. I think it will be better to specify
> that this restriction is temporary in the comments at both the places;
> but that's arguable.
>
> > Patch includes tests for all four locking strengths.
> > Since the code path looks simple we can just keep one of them as well
> and trim other
> > tests. Thoughts?
>
> I don't think we need all the four tests, but we need a test for
> locking clause without relation since that takes a different code
> path.
>
Attached is v2-0001, rebased on master and addressing
the comments you added in case we plan on parking the feature
work for once v20 opens.
Regards,
Ayush
Attachments:
[application/octet-stream] v2-0001-Add-the-missing-RTE_GRAPH_TABLE-case-to-transform.patch (4.5K, ../CAJTYsWX8J2cUTCnVRwks0gJw4wiMCv_qVSxshFC9f8oh7Ch6JA@mail.gmail.com/3-v2-0001-Add-the-missing-RTE_GRAPH_TABLE-case-to-transform.patch)
download | inline diff:
From 8c75b0d94e507a60616d304e85153461cf96c2c5 Mon Sep 17 00:00:00 2001
From: Ayush Tiwari <[email protected]>
Date: Tue, 9 Jun 2026 20:14:13 +0530
Subject: [PATCH v2] Add the missing RTE_GRAPH_TABLE case to
transformLockingClause()
A FOR UPDATE/SHARE clause that names a GRAPH_TABLE alias hit the
internal "unrecognized RTE type" error rather than a user-friendly
"FOR ... cannot be applied to a GRAPH_TABLE" message:
CREATE TABLE v(id int PRIMARY KEY, vname text);
CREATE PROPERTY GRAPH g VERTEX TABLES (v);
SELECT * FROM GRAPH_TABLE(g MATCH (a) COLUMNS (a.vname)) gt
FOR UPDATE OF gt;
-- ERROR: unrecognized RTE type: 8
Add an explicit case for RTE_GRAPH_TABLE that raises
ERRCODE_FEATURE_NOT_SUPPORTED, matching the convention already used
for RTE_FUNCTION, RTE_TABLEFUNC, RTE_VALUES, RTE_CTE and
RTE_NAMEDTUPLESTORE. Tests cover both the named-rel and the
all-rels code paths.
Author: SATYANARAYANA NARLAPURAM <[email protected]>
Reported-by: SATYANARAYANA NARLAPURAM <[email protected]>
Reviewed-by: Zhenwei Shang <[email protected]>
Reviewed-by: Ashutosh Bapat <[email protected]>
Discussion: https://postgr.es/m/CAHg%2BQDcE9wp6nOEC3SCRQ90nrCO%3DQF%2BOZq1MG8Qc6hnusmogqw%40mail.gmail.com
---
src/backend/parser/analyze.c | 16 +++++++++++++++-
src/test/regress/expected/graph_table.out | 16 ++++++++++++++++
src/test/regress/sql/graph_table.sql | 7 +++++++
3 files changed, 38 insertions(+), 1 deletion(-)
diff --git a/src/backend/parser/analyze.c b/src/backend/parser/analyze.c
index 93fa66ae57c..74294581ef9 100644
--- a/src/backend/parser/analyze.c
+++ b/src/backend/parser/analyze.c
@@ -3866,7 +3866,7 @@ transformLockingClause(ParseState *pstate, Query *qry, LockingClause *lc,
allrels, true);
break;
default:
- /* ignore JOIN, SPECIAL, FUNCTION, VALUES, CTE RTEs */
+ /* ignore JOIN, SPECIAL, FUNCTION, VALUES, CTE, GRAPH_TABLE */
break;
}
}
@@ -4001,6 +4001,20 @@ transformLockingClause(ParseState *pstate, Query *qry, LockingClause *lc,
LCS_asString(lc->strength)),
parser_errposition(pstate, thisrel->location)));
break;
+ case RTE_GRAPH_TABLE:
+ /*
+ * XXX: push the row mark down to the
+ * relations the GRAPH_TABLE expands to
+ * during rewriting.
+ */
+ ereport(ERROR,
+ (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
+ /*------
+ translator: %s is a SQL row locking clause such as FOR UPDATE */
+ errmsg("%s cannot be applied to a GRAPH_TABLE",
+ LCS_asString(lc->strength)),
+ parser_errposition(pstate, thisrel->location)));
+ break;
/* Shouldn't be possible to see RTE_RESULT here */
diff --git a/src/test/regress/expected/graph_table.out b/src/test/regress/expected/graph_table.out
index 70d986e8ab0..d95752cf92c 100644
--- a/src/test/regress/expected/graph_table.out
+++ b/src/test/regress/expected/graph_table.out
@@ -1077,4 +1077,20 @@ SELECT src.vname, count(*) FROM v1 AS src
v13 | 1
(3 rows)
+-- FOR UPDATE/SHARE on GRAPH_TABLE: not supported when the alias is named
+-- in the locking clause; silently ignored when no relations are named.
+SELECT * FROM GRAPH_TABLE (g1 MATCH (src IS vl1) COLUMNS (src.vname)) gt
+ FOR UPDATE OF gt; -- error
+ERROR: FOR UPDATE cannot be applied to a GRAPH_TABLE
+LINE 2: FOR UPDATE OF gt;
+ ^
+SELECT * FROM GRAPH_TABLE (g1 MATCH (src IS vl1) COLUMNS (src.vname)) gt
+ FOR UPDATE; -- ok, lock silently skipped for GRAPH_TABLE
+ vname
+-------
+ v11
+ v12
+ v13
+(3 rows)
+
-- leave the objects behind for pg_upgrade/pg_dump tests
diff --git a/src/test/regress/sql/graph_table.sql b/src/test/regress/sql/graph_table.sql
index 0b44f70d7e7..3976db4e231 100644
--- a/src/test/regress/sql/graph_table.sql
+++ b/src/test/regress/sql/graph_table.sql
@@ -612,4 +612,11 @@ SELECT src.vname, count(*) FROM v1 AS src
HAVING count(*) >= (SELECT count(*) FROM GRAPH_TABLE (g1 MATCH (a IS vl1 | vl2) COLUMNS (a.vname AS n)) WHERE n = src.vname)
ORDER BY vname;
+-- FOR UPDATE/SHARE on GRAPH_TABLE: not supported when the alias is named
+-- in the locking clause; silently ignored when no relations are named.
+SELECT * FROM GRAPH_TABLE (g1 MATCH (src IS vl1) COLUMNS (src.vname)) gt
+ FOR UPDATE OF gt; -- error
+SELECT * FROM GRAPH_TABLE (g1 MATCH (src IS vl1) COLUMNS (src.vname)) gt
+ FOR UPDATE; -- ok, lock silently skipped for GRAPH_TABLE
+
-- leave the objects behind for pg_upgrade/pg_dump tests
--
2.34.1
view thread (5+ messages) latest in thread
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]
Subject: Re: [Bug] Add the missing RTE_GRAPH_TABLE case to transformLockingClause()
In-Reply-To: <CAJTYsWX8J2cUTCnVRwks0gJw4wiMCv_qVSxshFC9f8oh7Ch6JA@mail.gmail.com>
* 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