On Mon, Nov  2, 2026 at 01:42:44PM -0800, Masahiko Sawada wrote:
> I've pushed a fix (8628df8891a) for how parallel vacuum computes the
> cost-based vacuum delay. As I mentioned in the commit message, while
> the fix corrects the delay computation, it could make VACUUM with the
> PARALLEL option and cost-based delay enabled slower after upgrading to
> v19. I think it's worth mentioning this in the release notes so that
> users can re-tune the cost-based delay parameters. I've attached the
> patch to add the item to the "Migration to Version 19" section. Please
> review it.

I have slightly reworded the patch and applied it, attached, thanks.

-- 
  Bruce Momjian  <[email protected]>        https://momjian.us
  EDB                                      https://enterprisedb.com

  Do not let urgent matters crowd out time for investment in the future.
diff --git a/doc/src/sgml/release-19.sgml b/doc/src/sgml/release-19.sgml
index a818984c946..0523775183b 100644
--- a/doc/src/sgml/release-19.sgml
+++ b/doc/src/sgml/release-19.sgml
@@ -303,6 +303,23 @@ large analytical queries to manually enable <acronym>JIT</acronym>.
 </para>
 </listitem>
 
+<!--
+Author: Masahiko Sawada <[email protected]>
+2026-11-02 [8628df889] Make parallel vacuum processes share the cost limit.
+-->
+
+<listitem>
+<para>
+Change <link linkend="parallel-vacuum">parallel vacuum</link> leaders and workers to share the cost-based vacuum delay limit (Bharath Rupireddy)
+<ulink url="&commit_baseurl;8628df889">&sect;</ulink>
+</para>
+
+<para>
+Previously each worker process slept based only on the cost it incurred itself, so a parallel vacuum could consume several times the budget set by <link linkend="guc-vacuum-cost-limit"><varname>vacuum_cost_limit</varname></link>.
+With this change, <link linkend="sql-vacuum"><command>VACUUM</command></link> with parallel workers and cost-based delay enabled might run slower than in previous releases.
+</para>
+</listitem>
+
 <!--
 Author: Amit Kapila <[email protected]>
 2025-11-18 [3edaf29fa] Rename two columns in pg_stat_subscription_stats.

Reply via email to