Location via proxy:   [ UP ]  
[Report a bug]   [Manage cookies]                
doc: PG 13 relnotes: hash_mem_multiplier can restore old behav.
authorBruce Momjian <bruce@momjian.us>
Mon, 3 Aug 2020 21:01:42 +0000 (17:01 -0400)
committerBruce Momjian <bruce@momjian.us>
Mon, 3 Aug 2020 21:01:42 +0000 (17:01 -0400)
Document that hash_mem_multiplier can get query behavior closer to the
pre-PG 13 behavior of allowing hashing to use more memory.

Reported-by: Peter Geoghegan
Discussion: https://postgr.es/m/CAH2-Wzn3kwQm_pe6g2=ki+P7+ZRqH5GvFGn6SWfv_j7UUgcLdQ@mail.gmail.com

Backpatch-through: 13 only

doc/src/sgml/release-13.sgml

index 18e6497a7f21b18d046b2e0d83ac34378ac8bb51..0b1051ba8245d4c4363ee3ab1c04def9593a02e2 100644 (file)
@@ -649,7 +649,10 @@ Author: Jeff Davis <jdavis@postgresql.org>
 
        <para>
         Previously, hash aggregation was avoided if it was expected to use
-        more than <xref linkend="guc-work-mem"/> memory.
+        more than <xref linkend="guc-work-mem"/> memory.  To reduce the
+        likelihood of using disk storage for hash aggregation and attain
+        behavior similar to previous Postgres releases, increase <xref
+        linkend="guc-hash-mem-multiplier"/>.
        </para>
       </listitem>