v4-0004-review.patch

text/x-patch

Filename: v4-0004-review.patch
Type: text/x-patch
Part: 3
Message: Re: Compress ReorderBuffer spill files using LZ4

Patch

Format: format-patch
Series: patch v4-0004
Subject: review
File+
src/backend/replication/logical/reorderbuffer.c 8 0
From 2e120c01baa97e82a62fd8c6936612f23dd4ac3c Mon Sep 17 00:00:00 2001
From: Tomas Vondra <tomas@vondra.me>
Date: Tue, 17 Sep 2024 16:47:55 +0200
Subject: [PATCH v4 04/11] review

---
 src/backend/replication/logical/reorderbuffer.c | 8 ++++++++
 1 file changed, 8 insertions(+)

diff --git a/src/backend/replication/logical/reorderbuffer.c b/src/backend/replication/logical/reorderbuffer.c
index 19b72e90af8..168956d944a 100644
--- a/src/backend/replication/logical/reorderbuffer.c
+++ b/src/backend/replication/logical/reorderbuffer.c
@@ -3805,6 +3805,14 @@ ReorderBufferSerializeTXN(ReorderBuffer *rb, ReorderBufferTXN *txn)
 
 /*
  * Serialize individual change to disk.
+ *
+ * Returns the amount of data written to disk (with compression, this is the
+ * size of the compressed representation).
+ *
+ * XXX But is this actually the right thing to do? Even without compression we
+ * don't really count the bytes written to the disk (we don't account for the
+ * DiskChange header), but rather the memory representation. So why should we
+ * do that with compression? Seems a bit strange.
  */
 static Size
 ReorderBufferSerializeChange(ReorderBuffer *rb, ReorderBufferTXN *txn,
-- 
2.46.0