From: Nathan Bossart <nathandbossart@gmail.com>
To: Rahila Syed <rahilasyed90@gmail.com>
Cc: Robert Haas <robertmhaas@gmail.com>
Cc: PostgreSQL Hackers <pgsql-hackers@lists.postgresql.org>
Subject: Re: show size of DSAs and dshash tables in pg_dsm_registry_allocations
Date: Mon, 1 Dec 2025 09:29:43 -0600
Message-ID: <aS20Z42rPHecTljS@nathan> (raw)
In-Reply-To: <CAH2L28tr5q965MbD2TroyTc-Ey+b-z3=3kpWAuGg6w7qQwwChA@mail.gmail.com>
References: <aSeEDeznAsHR1_YF@nathan>
<CAH2L28tr5q965MbD2TroyTc-Ey+b-z3=3kpWAuGg6w7qQwwChA@mail.gmail.com>
On Thu, Nov 27, 2025 at 01:51:39PM +0530, Rahila Syed wrote:
> Thank you for this improvement. I think it will be very helpful to have
> the ability to report dsa and dshash memory sizes.
Thanks for reviewing!
> 1. As a result of this change, the following comment in
> pg_get_dsm_registry_allocations is incorrect.
Fixed.
> + LWLockAcquire(&control->lock, LW_SHARED);
>
> The dsa_get_total_size function takes LW_EXCLUSIVE lock for the same
> purpose. I wonder why that is the case and whether both should be
> consistent.
That function was added by commit ee1b30f, which AFAICT used an exclusive
lock just to stay consistent with the rest of dsa.c [0]. I don't see any
discussion about this in the original DSA thread [1]. Perhaps we could go
through dsa.c and switch to LW_SHARED where appropriate, although I doubt
it makes much difference.
> +size_t
> +dsa_get_total_size_from_handle(dsa_handle handle)
>
> I believe this function will report the size as long as the dsa control
> structure is created within a dsm segment, since all dsm segments are
> tracked by the global list - dsm_segment_list, regardless of whether the
> dsa is created with dsa_create or dsa_create_in_place. In that case,
> perhaps we should update the comment above to reflect this.
Sorry, I'm not following what you think we should update the comment to
say.
> 4. Since, with this change, the size column will show memory allocation
> regardless of whether it is currently mapped in the local process, I
> think it would be helpful to add a boolean column to display the mapped
> status as a future enhancement.
Maybe, although I'm struggling to think of a scenario where that
information would be useful.
[0] https://postgr.es/m/CAD21AoDoWrbNf%2BK2Fwg2n%3DCZDHigjkndwqy_86BGgXBp9Kbq4Q%40mail.gmail.com
[1] https://postgr.es/m/CAEepm%3D1z5WLuNoJ80PaCvz6EtG9dN0j-KuHcHtU6QEfcPP5-qA%40mail.gmail.com
--
nathan
Attachments:v2-0001-rework-DSM-registry-view.patch (text/plain, 6.1K) download
From df18d34c52f0f744995c4f01b3b05205ed01b535 Mon Sep 17 00:00:00 2001
From: Nathan Bossart <nathan@postgresql.org>
Date: Wed, 26 Nov 2025 16:48:13 -0600
Subject: [PATCH v2 1/1] rework DSM registry view
---
doc/src/sgml/system-views.sgml | 4 +--
src/backend/storage/ipc/dsm_registry.c | 22 ++++--------
src/backend/utils/mmgr/dsa.c | 35 +++++++++++++++++++
src/include/utils/dsa.h | 1 +
.../expected/test_dsm_registry.out | 4 +--
.../sql/test_dsm_registry.sql | 4 +--
6 files changed, 49 insertions(+), 21 deletions(-)
diff --git a/doc/src/sgml/system-views.sgml b/doc/src/sgml/system-views.sgmlindex 7db8f73eba2..162c76b729a 100644--- a/doc/src/sgml/system-views.sgml+++ b/doc/src/sgml/system-views.sgml@@ -1150,8 +1150,8 @@ AND c1.path[c2.level] = c2.path[c2.level];
<structfield>size</structfield> <type>int8</type>
</para>
<para>
- Size of the allocation in bytes. NULL for entries of type- <literal>area</literal> and <literal>hash</literal>.+ Size of the allocation in bytes. NULL for entries that failed+ initialization.
</para></entry>
</row>
</tbody>
diff --git a/src/backend/storage/ipc/dsm_registry.c b/src/backend/storage/ipc/dsm_registry.cindex ef6533b1100..0f3e08c44d8 100644--- a/src/backend/storage/ipc/dsm_registry.c+++ b/src/backend/storage/ipc/dsm_registry.c@@ -463,26 +463,18 @@ pg_get_dsm_registry_allocations(PG_FUNCTION_ARGS)
Datum vals[3];
bool nulls[3] = {0};
- /* Do not show partially-initialized entries. */- if (entry->type == DSMR_ENTRY_TYPE_DSM &&- entry->dsm.handle == DSM_HANDLE_INVALID)- continue;- if (entry->type == DSMR_ENTRY_TYPE_DSA &&- entry->dsa.handle == DSA_HANDLE_INVALID)- continue;- if (entry->type == DSMR_ENTRY_TYPE_DSH &&- entry->dsh.dsa_handle == DSA_HANDLE_INVALID)- continue;-
vals[0] = CStringGetTextDatum(entry->name);
vals[1] = CStringGetTextDatum(DSMREntryTypeNames[entry->type]);
- /*- * Since we can't know the size of DSA/dshash entries without first- * attaching to them, return NULL for those.- */+ /* Be careful to only return the sizes of initialized entries. */
if (entry->type == DSMR_ENTRY_TYPE_DSM)
vals[2] = Int64GetDatum(entry->dsm.size);
+ else if (entry->type == DSMR_ENTRY_TYPE_DSA &&+ entry->dsa.handle != DSA_HANDLE_INVALID)+ vals[2] = Int64GetDatum(dsa_get_total_size_from_handle(entry->dsa.handle));+ else if (entry->type == DSMR_ENTRY_TYPE_DSH &&+ entry->dsh.dsa_handle !=DSA_HANDLE_INVALID)+ vals[2] = Int64GetDatum(dsa_get_total_size_from_handle(entry->dsh.dsa_handle));
else
nulls[2] = true;
diff --git a/src/backend/utils/mmgr/dsa.c b/src/backend/utils/mmgr/dsa.cindex be43e9351c3..030f9463712 100644--- a/src/backend/utils/mmgr/dsa.c+++ b/src/backend/utils/mmgr/dsa.c@@ -1050,6 +1050,41 @@ dsa_get_total_size(dsa_area *area)
return size;
}
+/*+ * Same as dsa_get_total_size(), but accepts a DSA handle. The area must have+ * been created with dsa_create (not dsa_create_in_place).+ */+size_t+dsa_get_total_size_from_handle(dsa_handle handle)+{+ size_t size;+ bool already_attached;+ dsm_segment *segment;+ dsa_area_control *control;++ already_attached = dsa_is_attached(handle);+ if (already_attached)+ segment = dsm_find_mapping(handle);+ else+ segment = dsm_attach(handle);++ if (segment == NULL)+ ereport(ERROR,+ (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),+ errmsg("could not attach to dynamic shared area")));++ control = (dsa_area_control *) dsm_segment_address(segment);++ LWLockAcquire(&control->lock, LW_SHARED);+ size = control->total_segment_size;+ LWLockRelease(&control->lock);++ if (!already_attached)+ dsm_detach(segment);++ return size;+}+
/*
* Aggressively free all spare memory in the hope of returning DSM segments to
* the operating system.
diff --git a/src/include/utils/dsa.h b/src/include/utils/dsa.hindex f2104dacbfc..42449ff22de 100644--- a/src/include/utils/dsa.h+++ b/src/include/utils/dsa.h@@ -161,6 +161,7 @@ extern dsa_pointer dsa_allocate_extended(dsa_area *area, size_t size, int flags)
extern void dsa_free(dsa_area *area, dsa_pointer dp);
extern void *dsa_get_address(dsa_area *area, dsa_pointer dp);
extern size_t dsa_get_total_size(dsa_area *area);
+extern size_t dsa_get_total_size_from_handle(dsa_handle handle);
extern void dsa_trim(dsa_area *area);
extern void dsa_dump(dsa_area *area);
diff --git a/src/test/modules/test_dsm_registry/expected/test_dsm_registry.out b/src/test/modules/test_dsm_registry/expected/test_dsm_registry.outindex ca8abbb377e..75d9eda0756 100644--- a/src/test/modules/test_dsm_registry/expected/test_dsm_registry.out+++ b/src/test/modules/test_dsm_registry/expected/test_dsm_registry.out@@ -1,4 +1,4 @@-SELECT name, type, size IS DISTINCT FROM 0 AS size+SELECT name, type, size > 0 AS size
FROM pg_dsm_registry_allocations
WHERE name like 'test_dsm_registry%' ORDER BY name;
name | type | size
@@ -32,7 +32,7 @@ SELECT get_val_in_hash('test');
(1 row)
\c
-SELECT name, type, size IS DISTINCT FROM 0 AS size+SELECT name, type, size > 0 AS size
FROM pg_dsm_registry_allocations
WHERE name like 'test_dsm_registry%' ORDER BY name;
name | type | size
diff --git a/src/test/modules/test_dsm_registry/sql/test_dsm_registry.sql b/src/test/modules/test_dsm_registry/sql/test_dsm_registry.sqlindex 965a3f1ebb6..1b9ee3ebf18 100644--- a/src/test/modules/test_dsm_registry/sql/test_dsm_registry.sql+++ b/src/test/modules/test_dsm_registry/sql/test_dsm_registry.sql@@ -1,4 +1,4 @@-SELECT name, type, size IS DISTINCT FROM 0 AS size+SELECT name, type, size > 0 AS size
FROM pg_dsm_registry_allocations
WHERE name like 'test_dsm_registry%' ORDER BY name;
CREATE EXTENSION test_dsm_registry;
@@ -8,6 +8,6 @@ SELECT set_val_in_hash('test', '1414');
SELECT get_val_in_shmem();
SELECT get_val_in_hash('test');
\c
-SELECT name, type, size IS DISTINCT FROM 0 AS size+SELECT name, type, size > 0 AS size
FROM pg_dsm_registry_allocations
WHERE name like 'test_dsm_registry%' ORDER BY name;
--
2.39.5 (Apple Git-154)
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: pgsql-hackers@postgresql.org
Cc: nathandbossart@gmail.com, rahilasyed90@gmail.com, robertmhaas@gmail.com, pgsql-hackers@lists.postgresql.org
Subject: Re: show size of DSAs and dshash tables in pg_dsm_registry_allocations
In-Reply-To: <aS20Z42rPHecTljS@nathan>
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
This inbox is served by DDX for PostgreSQL; see mirroring instructions
for how to clone and mirror all data and code used for this inbox