Fix infinite wait when reading a partially written WAL record
authorAlexander Korotkov <akorotkov@postgresql.org>
Sat, 19 Jul 2025 11:13:40 +0000 (14:13 +0300)
committerAlexander Korotkov <akorotkov@postgresql.org>
Sat, 19 Jul 2025 11:13:58 +0000 (14:13 +0300)
If a crash occurs while writing a WAL record that spans multiple pages, the
recovery process marks the page with the XLP_FIRST_IS_OVERWRITE_CONTRECORD
flag.  However, logical decoding currently attempts to read the full WAL
record based on its expected size before checking this flag, which can lead
to an infinite wait if the remaining data is never written (e.g., no activity
after crash).

This patch updates the logic first to read the page header and check for
the XLP_FIRST_IS_OVERWRITE_CONTRECORD flag before attempting to reconstruct
the full WAL record.  If the flag is set, decoding correctly identifies
the record as incomplete and avoids waiting for WAL data that will never
arrive.

Discussion: http://postgr.es/m/CAAKRu_ZCOzQpEumLFgG_%2Biw3FTa%2BhJ4SRpxzaQBYxxM_ZAzWcA%40mail.gmail.com
Discussion: http://postgr.es/m/CALDaNm34m36PDHzsU_GdcNXU0gLTfFY5rzh9GSQv%3Dw6B%2BQVNRQ%40mail.gmail.com
Author: Vignesh C <vignesh21@gmail.com>
Reviewed-by: Hayato Kuroda <kuroda.hayato@fujitsu.com>
Reviewed-by: Dilip Kumar <dilipbalaut@gmail.com>
Reviewed-by: Michael Paquier <michael@paquier.xyz>
Reviewed-by: Alexander Korotkov <aekorotkov@gmail.com>
Backpatch-through: 13

src/backend/access/transam/xlogreader.c

index 939dd3e0d191afb660e3af5835a613087c48d509..80a453a4ebeaf33114c2c95f75d6914e50a57d6d 100644 (file)
@@ -407,11 +407,12 @@ restart:
            /* Calculate pointer to beginning of next page */
            targetPagePtr += XLOG_BLCKSZ;
 
-           /* Wait for the next page to become available */
-           readOff = ReadPageInternal(state, targetPagePtr,
-                                      Min(total_len - gotlen + SizeOfXLogShortPHD,
-                                          XLOG_BLCKSZ));
-
+           /*
+            * Read the page header before processing the record data, so we
+            * can handle the case where the previous record ended as being a
+            * partial one.
+            */
+           readOff = ReadPageInternal(state, targetPagePtr, SizeOfXLogShortPHD);
            if (readOff < 0)
                goto err;
 
@@ -458,6 +459,13 @@ restart:
                goto err;
            }
 
+           /* Wait for the next page to become available */
+           readOff = ReadPageInternal(state, targetPagePtr,
+                                      Min(total_len - gotlen + SizeOfXLogShortPHD,
+                                          XLOG_BLCKSZ));
+           if (readOff < 0)
+               goto err;
+
            /* Append the continuation from this page to the buffer */
            pageHeaderSize = XLogPageHeaderSize(pageHeader);