pg.ddx.io  pgbackrest/pgbackrest GitHub issues and pull requests  
help / color / mirror / Atom feed
[pgbackrest/pgbackrest] PR #2906: Support alternate relation segment sizes for page checksums.
17+ messages / 3 participants
[nested] [flat]

* [pgbackrest/pgbackrest] PR #2906: Support alternate relation segment sizes for page checksums.
@ 2026-11-06 11:24  "dwsteele (@dwsteele)" <noreply+dwsteele@github.com>
  0 siblings, 0 replies; 17+ messages in thread

From: dwsteele (@dwsteele) @ 2026-11-06 11:24 UTC (permalink / raw)
  To: pgbackrest/pgbackrest <noreply+pgbackrest-pgbackrest@github.com>

Page checksum validation assumed a 1GB relation segment. The block number is part of the page checksum so on a cluster built with a different segment size pages in every segment after the first failed validation. With larger segments the retry also read the wrong page past the first 1GB of a file and treated the page as changed, which prevented real corruption in that range from being reported in any segment. The restore command has errored on files with invalid page checksums by default since 2.60.0.

Get the total pages per segment from relseg_size in pg_control and error if the value is outside the range allowed by PostgreSQL. The value is only used during backup so it does not need to be stored in the repository.

Remove the assert that required the number of pages per segment to be divisible by the page size in bytes. This compared a page count to a byte count and failed for valid segment sizes, e.g. --with-segsize-blocks=6.

inline patch 1: src/command/backup/backup.c, src/command/backup/file.c.inc, src/command/backup/pageChecksum.c and 10 more (+69/-26 lines) download inline-1.patchdiff --git a/src/command/backup/backup.c b/src/command/backup/backup.c index ffaba65694..94806ae5a3 100644 --- a/src/command/backup/backup.c +++ b/src/command/backup/backup.c @@ -67,6 +67,7 @@ typedef struct BackupData unsigned int version; // PostgreSQL version unsigned int walSegmentSize; // PostgreSQL wal segment size PgPageSize pageSize; // PostgreSQL page size + unsigned int segmentPageTotal; // PostgreSQL total pages in a relation segment } BackupData; static BackupData * @@ -133,6 +134,7 @@ backupInit(const InfoBackup *const infoBackup) result->version = pgControl.version; result->walSegmentSize = pgControl.walSegmentSize; result->pageSize = pgControl.pageSize; + result->segmentPageTotal = pgControl.segmentPageTotal; // Validate pg_control info against the stanza if (result->version != infoPg.version || pgControl.systemId != infoPg.systemId) diff --git a/src/command/backup/file.c.inc b/src/command/backup/file.c.inc index ec1fe887d6..ca8db64fe7 100644 --- a/src/command/backup/file.c.inc +++ b/src/command/backup/file.c.inc @@ -151,7 +151,7 @@ static List * backupFile( const String *const repoFile, const uint64_t bundleId, const bool bundleRaw, const unsigned int blockIncrReference, const CompressType repoFileCompressType, const int repoFileCompressLevel, const CipherSpec *const cipherSpecBackup, - const String *const pgVersionForce, const PgPageSize pageSize, const List *const fileList) + const String *const pgVersionForce, const PgPageSize pageSize, const unsigned int segmentPageTotal, const List *const fileList) { FUNCTION_LOG_BEGIN(logLevelDebug); FUNCTION_LOG_PARAM(STRING, repoFile); // Repo file @@ -162,6 +162,7 @@ backupFile( FUNCTION_LOG_PARAM(INT, repoFileCompressLevel); // Compression level for repo file FUNCTION_LOG_PARAM(CIPHER_SPEC, cipherSpecBackup); // Cipher spec to encrypt the backup file FUNCTION_LOG_PARAM(ENUM, pageSize); // Page size + FUNCTION_LOG_PARAM(UINT, segmentPageTotal); // Total pages in a relation segment FUNCTION_LOG_PARAM(STRING, pgVersionForce); // Force pg version FUNCTION_LOG_PARAM(LIST, fileList); // List of files to backup FUNCTION_LOG_END(); @@ -170,6 +171,7 @@ backupFile( ASSERT(cipherSpecBackup != NULL); ASSERT(fileList != NULL && !lstEmpty(fileList)); ASSERT(pgPageSizeValid(pageSize)); + ASSERT(segmentPageTotal > 0); // Backup file results List *const result = lstNewP(sizeof(BackupFileResult)); @@ -356,8 +358,8 @@ backupFile( ioFilterGroupAdd( ioReadFilterGroup(readIo), pageChecksumNew( - segmentNumber(file->pgFile), PG_SEGMENT_SIZE_DEFAULT / pageSize, pageSize, - file->pgFilePageHeaderCheck, storagePathP(storagePg(), file->pgFile))); + segmentNumber(file->pgFile), segmentPageTotal, pageSize, file->pgFilePageHeaderCheck, + storagePathP(storagePg(), file->pgFile))); } // Compress filter diff --git a/src/command/backup/pageChecksum.c b/src/command/backup/pageChecksum.c index 653bf160fc..75cdd97627 100644 --- a/src/command/backup/pageChecksum.c +++ b/src/command/backup/pageChecksum.c @@ -238,7 +238,7 @@ pageChecksumNew( FUNCTION_LOG_END(); ASSERT(pgPageSizeValid(pageSize)); - ASSERT(segmentPageTotal > 0 && segmentPageTotal % pageSize == 0); + ASSERT(segmentPageTotal > 0); OBJ_NEW_BEGIN(PageChecksum, .childQty = MEM_CONTEXT_QTY_MAX) { diff --git a/src/command/backup/process.c.inc b/src/command/backup/process.c.inc index e58e3db4e6..0feababa36 100644 --- a/src/command/backup/process.c.inc +++ b/src/command/backup/process.c.inc @@ -340,6 +340,7 @@ typedef struct BackupJobData RegExp *standbyExp; // Identify files that may be copied from the standby const CipherSpec *const cipherSpecBackup; // Cipher spec used to encrypt files in the backup const PgPageSize pageSize; // Page size + const unsigned int segmentPageTotal; // Total pages in a relation segment const CompressType compressType; // Backup compression type const int compressLevel; // Compress level if backup is compressed const bool delta; // Is this a checksum delta backup? @@ -692,6 +693,7 @@ backupJobCallback(void *const data, const unsigned int clientIdx) pckWriteI32P(param, jobData->compressLevel); cipherSpecPack(param, jobData->cipherSpecBackup); pckWriteU32P(param, jobData->pageSize); + pckWriteU32P(param, jobData->segmentPageTotal); pckWriteStrP(param, cfgOptionStrNull(cfgOptPgVersionForce)); } @@ -806,6 +808,7 @@ backupProcess(const BackupData *const backupData, Manifest *const manifest, cons .compressLevel = cfgOptionInt(cfgOptCompressLevel), .cipherSpecBackup = manifestCipherSpec(manifest), .pageSize = backupData->pageSize, + .segmentPageTotal = backupData->segmentPageTotal, .delta = cfgOptionBool(cfgOptDelta), .bundle = cfgOptionBool(cfgOptRepoBundle), .bundleId = 1, diff --git a/src/command/backup/protocol.c b/src/command/backup/protocol.c index 88c0ed8489..235fdab94d 100644 --- a/src/command/backup/protocol.c +++ b/src/command/backup/protocol.c @@ -117,6 +117,7 @@ backupFileProtocol(PackRead *const param) const int repoFileCompressLevel = pckReadI32P(param); const CipherSpec *const cipherSpecBackup = cipherSpecNewPack(param); const PgPageSize pageSize = pckReadU32P(param); + const unsigned int segmentPageTotal = pckReadU32P(param); const String *const pgVersionForce = pckReadStrP(param); // Build the file list @@ -163,7 +164,7 @@ backupFileProtocol(PackRead *const param) // Backup file const List *const resultList = backupFile( repoFile, bundleId, bundleRaw, blockIncrReference, repoFileCompressType, repoFileCompressLevel, cipherSpecBackup, - pgVersionForce, pageSize, fileList); + pgVersionForce, pageSize, segmentPageTotal, fileList); // Return result PackWrite *const data = protocolServerResultData(result); diff --git a/src/postgres/interface.c b/src/postgres/interface.c index 1bead8d27b..070ad0ea14 100644 --- a/src/postgres/interface.c +++ b/src/postgres/interface.c @@ -52,6 +52,11 @@ concurrent reads and writes so retries are required. #define PG_CONTROL_SIZE 8192 #define PG_CONTROL_DATA_SIZE 512 +/*********************************************************************************************************************************** +Maximum pages in a relation segment +***********************************************************************************************************************************/ +#define PG_SEGMENT_PAGE_TOTAL_MAX 2147483647 + /*********************************************************************************************************************************** PostgreSQL interface definitions @@ -317,6 +322,15 @@ pgControlFromBuffer(const Buffer *const controlFile, const String *const pgVersi // Check the page size pgPageSizeCheck(result.pageSize); + // Check the relation segment size + if (result.segmentPageTotal == 0 || result.segmentPageTotal > PG_SEGMENT_PAGE_TOTAL_MAX) + { + THROW_FMT( + FormatError, + "relation segment size is %u pages but must be between 1 and " STRINGIFY(PG_SEGMENT_PAGE_TOTAL_MAX) " inclusive", + result.segmentPageTotal); + } + // Check the checksum version if (result.pageChecksumVersion > PG_DATA_CHECKSUM_VERSION) { diff --git a/src/postgres/interface.h b/src/postgres/interface.h index 9ec94b11e0..3674e5dd1c 100644 --- a/src/postgres/interface.h +++ b/src/postgres/interface.h @@ -73,13 +73,6 @@ typedef enum pgPageSize32 = 32 * 1024, } PgPageSize; -/*********************************************************************************************************************************** -Define default segment size and pages per segment - -Segment size can only be changed at compile time and is not known to be well-tested, so only the default segment size is supported. -***********************************************************************************************************************************/ -#define PG_SEGMENT_SIZE_DEFAULT ((unsigned int)(1 * 1024 * 1024 * 1024)) - /*********************************************************************************************************************************** WAL header size. It doesn't seem worth tracking the exact size of the WAL header across versions of PostgreSQL so just set it to something far larger needed but <= the minimum read size on just about any system. @@ -104,6 +97,7 @@ typedef struct PgControl uint32_t timeline; // Current timeline PgPageSize pageSize; + unsigned int segmentPageTotal; // Total pages in a relation segment unsigned int walSegmentSize; unsigned int pageChecksumVersion; // Page checksum version (0 if no checksum, 1 if checksum) diff --git a/src/postgres/interface/version.intern.h b/src/postgres/interface/version.intern.h index 84f1f9fe5b..1280d1b845 100644 --- a/src/postgres/interface/version.intern.h +++ b/src/postgres/interface/version.intern.h @@ -84,6 +84,7 @@ Read the version specific pg_control into a general data structure .checkpoint = ((const ControlFileData *)controlFile)->checkPoint, \ .timeline = ((const ControlFileData *)controlFile)->checkPointCopy.ThisTimeLineID, \ .pageSize = ((const ControlFileData *)controlFile)->blcksz, \ + .segmentPageTotal = ((const ControlFileData *)controlFile)->relseg_size, \ .walSegmentSize = ((const ControlFileData *)controlFile)->xlog_seg_size, \ .pageChecksumVersion = ((const ControlFileData *)controlFile)->data_checksum_version, \ }; \ diff --git a/test/src/harness/postgres.c b/test/src/harness/postgres.c index b7acfc7e47..48939dc7f4 100644 --- a/test/src/harness/postgres.c +++ b/test/src/harness/postgres.c @@ -88,6 +88,9 @@ hrnPgControlToBuffer(unsigned int controlVersion, const unsigned int crc, PgCont // Set defaults if values are not passed pgControl.pageSize = pgControl.pageSize == 0 ? pgPageSize8 : pgControl.pageSize; + pgControl.segmentPageTotal = + pgControl.segmentPageTotal == UINT_MAX ? + 0 : (pgControl.segmentPageTotal == 0 ? HRN_PG_SEGMENT_SIZE_DEFAULT / pgControl.pageSize : pgControl.segmentPageTotal); pgControl.walSegmentSize = pgControl.walSegmentSize == UINT_MAX ? 0 : (pgControl.walSegmentSize == 0 ? HRN_PG_WAL_SEGMENT_SIZE_DEFAULT : pgControl.walSegmentSize); diff --git a/test/src/harness/postgres.h b/test/src/harness/postgres.h index 3fcc81ee9e..09b60738ce 100644 --- a/test/src/harness/postgres.h +++ b/test/src/harness/postgres.h @@ -19,6 +19,11 @@ Default wal segment size ***********************************************************************************************************************************/ #define HRN_PG_WAL_SEGMENT_SIZE_DEFAULT ((unsigned int)(16 * 1024 * 1024)) +/*********************************************************************************************************************************** +Default relation segment size +***********************************************************************************************************************************/ +#define HRN_PG_SEGMENT_SIZE_DEFAULT ((unsigned int)(1024 * 1024 * 1024)) + /*********************************************************************************************************************************** System id constants by version diff --git a/test/src/harness/postgres/version.intern.h b/test/src/harness/postgres/version.intern.h index c71a28f5a1..4a85f226e5 100644 --- a/test/src/harness/postgres/version.intern.h +++ b/test/src/harness/postgres/version.intern.h @@ -29,6 +29,7 @@ Create a pg_control file .ThisTimeLineID = pgControl.timeline, \ }, \ .blcksz = pgControl.pageSize, \ + .relseg_size = pgControl.segmentPageTotal, \ .xlog_seg_size = pgControl.walSegmentSize, \ .data_checksum_version = pgControl.pageChecksumVersion, \ }; \ diff --git a/test/src/module/command/backupTest.c b/test/src/module/command/backupTest.c index e7d7ca54a8..c80f39c9b9 100644 --- a/test/src/module/command/backupTest.c +++ b/test/src/module/command/backupTest.c @@ -670,7 +670,7 @@ testRun(void) // ***************************************************************************************************************************** if (testBegin("PageChecksum")) { - #define PG_SEGMENT_PAGE_DEFAULT (PG_SEGMENT_SIZE_DEFAULT / pgPageSize8) + #define PG_SEGMENT_PAGE_DEFAULT (HRN_PG_SEGMENT_SIZE_DEFAULT / pgPageSize8) // ------------------------------------------------------------------------------------------------------------------------- TEST_TITLE("segment page default"); @@ -2068,8 +2068,8 @@ testRun(void) "P00 INFO: new backup label = [FULL-1]\n" "P00 INFO: full backup size = 8KB, file total = 2", TEST_64BIT() ? - (TEST_BIG_ENDIAN() ? "a73da1cfddf849880b4f6f0f4209ce71043f2ac1" : "807d4464138ef085e41844a9f488f067b3d46d46") : - "c09a7f1d2ff6131ec467e18b781970a86efd4e5b"); + (TEST_BIG_ENDIAN() ? "884c00e2f6fcfc2c73364a285175483783a7cace" : "7583cfcf2d4c882b248898162a17a19929ed6f61") : + "5cf90a168630adb21cd2352036e2be3af86d7474"); // Make pg no longer appear to be running HRN_STORAGE_REMOVE(storagePgWrite(), PG_FILE_POSTMTRPID, .errorOnMissing = true); @@ -3048,7 +3048,8 @@ testRun(void) { // Update pg_control - HRN_PG_CONTROL_PUT(storagePgWrite(), PG_VERSION_11, .pageChecksumVersion = 1, .walSegmentSize = 1024 * 1024); + HRN_PG_CONTROL_PUT( + storagePgWrite(), PG_VERSION_11, .pageChecksumVersion = 1, .segmentPageTotal = 8, .walSegmentSize = 1024 * 1024); // Update version HRN_STORAGE_PUT_Z(storagePgWrite(), PG_FILE_PGVERSION, PG_VERSION_11_Z, .timeModified = backupTimeStart); @@ -3123,17 +3124,19 @@ testRun(void) HRN_STORAGE_PUT(storagePgWrite(), PG_PATH_BASE "/1/3", relation, .timeModified = backupTimeStart); - // File with bad page checksum + // File with bad page checksum in the second relation segment + HRN_STORAGE_REMOVE(storagePgWrite(), "base/1/4", .errorOnMissing = true); + relation = bufNew(pgPageSize8 * 3); memset(bufPtr(relation), 0, bufSize(relation)); *(PageHeaderData *)(bufPtr(relation) + (pgPageSize8 * 0x00)) = (PageHeaderData){.pd_upper = 0x00}; *(PageHeaderData *)(bufPtr(relation) + (pgPageSize8 * 0x01)) = (PageHeaderData){.pd_upper = 0x08}; *(PageHeaderData *)(bufPtr(relation) + (pgPageSize8 * 0x02)) = (PageHeaderData){.pd_upper = 0xFF}; ((PageHeaderData *)(bufPtr(relation) + (pgPageSize8 * 0x02)))->pd_checksum = pgPageChecksum( - bufPtr(relation) + (pgPageSize8 * 0x02), 2, pgPageSize8); + bufPtr(relation) + (pgPageSize8 * 0x02), 10, pgPageSize8); bufUsedSet(relation, bufSize(relation)); - HRN_STORAGE_PUT(storagePgWrite(), PG_PATH_BASE "/1/4", relation, .timeModified = backupTimeStart); + HRN_STORAGE_PUT(storagePgWrite(), PG_PATH_BASE "/1/4.1", relation, .timeModified = backupTimeStart); // Add a tablespace HRN_STORAGE_PATH_CREATE(storagePgWrite(), PG_PATH_PGTBLSPC); @@ -3166,8 +3169,8 @@ testRun(void) "P00 INFO: check archive for segment 0000000105DB5DE000000000\n" "P01 DETAIL: backup file " TEST_PATH "/pg1/base/1/3 (40KB, [PCT]) checksum [SHA1]\n" "P00 WARN: invalid page checksums found in file " TEST_PATH "/pg1/base/1/3 at pages 0, 2-4\n" - "P01 DETAIL: backup file " TEST_PATH "/pg1/base/1/4 (24KB, [PCT]) checksum [SHA1]\n" - "P00 WARN: invalid page checksum found in file " TEST_PATH "/pg1/base/1/4 at page 1\n" + "P01 DETAIL: backup file " TEST_PATH "/pg1/base/1/4.1 (24KB, [PCT]) checksum [SHA1]\n" + "P00 WARN: invalid page checksum found in file " TEST_PATH "/pg1/base/1/4.1 at page 9\n" "P01 DETAIL: backup file " TEST_PATH "/pg1/base/1/1 (16KB, [PCT]) checksum [SHA1]\n" "P01 DETAIL: backup file " TEST_PATH "/pg1/base/1/2 (8.5KB, [PCT]) checksum [SHA1]\n" "P00 WARN: page misalignment in file " TEST_PATH "/pg1/base/1/2: file size 8704 is not divisible by page size" @@ -3195,7 +3198,7 @@ testRun(void) "pg_data/base/1/1.gz {s=16384, ckp=t}\n" "pg_data/base/1/2.gz {s=8704, ckp=t}\n" "pg_data/base/1/3.gz {s=40960, ckp=[0,[2,4]]}\n" - "pg_data/base/1/4.gz {s=24576, ckp=[1]}\n" + "pg_data/base/1/4.1.gz {s=24576, ckp=[9]}\n" "pg_data/global/pg_control.gz {s=8192}\n" "pg_data/pg_tblspc/\n" "pg_data/pg_wal/0000000105DB5DE000000000.gz {s=1048576, ts=+2}\n" @@ -3213,7 +3216,7 @@ testRun(void) // Remove test files but base/1/3 (with its invalid page checksums) for the following test HRN_STORAGE_REMOVE(storagePgWrite(), "base/1/2", .errorOnMissing = true); - HRN_STORAGE_REMOVE(storagePgWrite(), "base/1/4", .errorOnMissing = true); + HRN_STORAGE_REMOVE(storagePgWrite(), "base/1/4.1", .errorOnMissing = true); } // ------------------------------------------------------------------------------------------------------------------------- diff --git a/test/src/module/postgres/interfaceTest.c b/test/src/module/postgres/interfaceTest.c index cefa6f9e21..3db39fcda2 100644 --- a/test/src/module/postgres/interfaceTest.c +++ b/test/src/module/postgres/interfaceTest.c @@ -94,7 +94,7 @@ testRun(void) "HINT: calculated 0x%x but expected value is 0xfadefade\n" "HINT: is pg_control corrupt?\n" "HINT: does pg_control have a different layout than expected?", - (uint32_t)(TEST_BIG_ENDIAN() ? 0x4e206eeb : (TEST_64BIT() ? 0x4ad387b2 : 0x3ca3a1ec))); + (uint32_t)(TEST_BIG_ENDIAN() ? 0xb0421e53 : (TEST_64BIT() ? 0x7bf250ff : 0x0d8276a1))); // ------------------------------------------------------------------------------------------------------------------------- TEST_TITLE("invalid CRC on force control version"); @@ -108,12 +108,12 @@ testRun(void) "HINT: checksum values may be misleading due to forced version scan\n" "HINT: is pg_control corrupt?\n" "HINT: does pg_control have a different layout than expected?", - (uint32_t)(TEST_BIG_ENDIAN() ? 0x27dc2b85 : (TEST_64BIT() ? 0xa9264d94 : 0xb371302a))); + (uint32_t)(TEST_BIG_ENDIAN() ? 0x07653056 : (TEST_64BIT() ? 0xadb340b4 : 0xd9c9d3d5))); // ------------------------------------------------------------------------------------------------------------------------- HRN_PG_CONTROL_PUT( storageTest, PG_VERSION_11, .systemId = 0xFACEFACE, .checkpoint = 0xEEFFEEFFAABBAABB, .timeline = 47, - .walSegmentSize = 1024 * 1024); + .segmentPageTotal = INT_MAX, .walSegmentSize = 1024 * 1024); PgControl info = {0}; TEST_ASSIGN(info, pgControlFromFile(storageTest, NULL), "get control info v11"); @@ -122,6 +122,7 @@ testRun(void) TEST_RESULT_UINT(info.catalogVersion, 201809051, " check catalog version"); TEST_RESULT_UINT(info.checkpoint, 0xEEFFEEFFAABBAABB, "check checkpoint"); TEST_RESULT_UINT(info.timeline, 47, "check timeline"); + TEST_RESULT_UINT(info.segmentPageTotal, INT_MAX, "check segment page total"); // ------------------------------------------------------------------------------------------------------------------------- HRN_PG_CONTROL_PUT(storageTest, PG_VERSION_11, .walSegmentSize = UINT_MAX); // UINT_MAX forces size to 0 @@ -155,6 +156,19 @@ testRun(void) pgControlFromFile(storageTest, NULL), FormatError, "page size is 65536 but only 1024, 2048, 4096, 8192, 16384, and 32768 are supported"); + // ------------------------------------------------------------------------------------------------------------------------- + HRN_PG_CONTROL_PUT(storageTest, PG_VERSION_11, .segmentPageTotal = UINT_MAX); // UINT_MAX forces size to 0 + + TEST_ERROR( + pgControlFromFile(storageTest, NULL), FormatError, + "relation segment size is 0 pages but must be between 1 and 2147483647 inclusive"); + + HRN_PG_CONTROL_PUT(storageTest, PG_VERSION_11, .segmentPageTotal = (unsigned int)INT_MAX + 1); + + TEST_ERROR( + pgControlFromFile(storageTest, NULL), FormatError, + "relation segment size is 2147483648 pages but must be between 1 and 2147483647 inclusive"); + // ------------------------------------------------------------------------------------------------------------------------- HRN_PG_CONTROL_PUT(storageTest, PG_VERSION_11, .pageChecksumVersion = 2);
^ permalink raw [nested|flat] 17+ messages in thread

* Re: [pgbackrest/pgbackrest] PR #2906: Support alternate relation segment sizes for page checksums.
@ 2026-11-06 11:24  "dwsteele (@dwsteele)" <noreply+dwsteele@github.com>
  15 siblings, 0 replies; 17+ messages in thread

From: dwsteele (@dwsteele) @ 2026-11-06 11:24 UTC (permalink / raw)
  To: pgbackrest/pgbackrest <noreply+pgbackrest-pgbackrest@github.com>

@dwsteele set milestone to "2.61.0".

^ permalink  raw  [nested|flat] 17+ messages in thread

* Re: [pgbackrest/pgbackrest] PR #2906: Support alternate relation segment sizes for page checksums.
@ 2026-11-06 11:24  "dwsteele (@dwsteele)" <noreply+dwsteele@github.com>
  15 siblings, 0 replies; 17+ messages in thread

From: dwsteele (@dwsteele) @ 2026-11-06 11:24 UTC (permalink / raw)
  To: pgbackrest/pgbackrest <noreply+pgbackrest-pgbackrest@github.com>

@dwsteele assigned @dwsteele.

^ permalink  raw  [nested|flat] 17+ messages in thread

* Re: [pgbackrest/pgbackrest] PR #2906: Support alternate relation segment sizes for page checksums.
@ 2026-11-06 11:24  "dwsteele (@dwsteele)" <noreply+dwsteele@github.com>
  15 siblings, 0 replies; 17+ messages in thread

From: dwsteele (@dwsteele) @ 2026-11-06 11:24 UTC (permalink / raw)
  To: pgbackrest/pgbackrest <noreply+pgbackrest-pgbackrest@github.com>

@dwsteele applied label "module [core]".

^ permalink  raw  [nested|flat] 17+ messages in thread

* Re: [pgbackrest/pgbackrest] PR #2906: Support alternate relation segment sizes for page checksums.
@ 2026-11-06 11:24  "dwsteele (@dwsteele)" <noreply+dwsteele@github.com>
  15 siblings, 0 replies; 17+ messages in thread

From: dwsteele (@dwsteele) @ 2026-11-06 11:24 UTC (permalink / raw)
  To: pgbackrest/pgbackrest <noreply+pgbackrest-pgbackrest@github.com>

@dwsteele applied label "enhancement".

^ permalink  raw  [nested|flat] 17+ messages in thread

* Re: [pgbackrest/pgbackrest] PR #2906 (approved): Support alternate relation segment sizes for page checksums.
@ 2026-11-06 15:07  "hunleyd (@hunleyd)" <noreply+hunleyd@github.com>
  15 siblings, 0 replies; 17+ messages in thread

From: hunleyd (@hunleyd) @ 2026-11-06 15:07 UTC (permalink / raw)
  To: pgbackrest/pgbackrest <noreply+pgbackrest-pgbackrest@github.com>



^ permalink  raw  [nested|flat] 17+ messages in thread

* Re: [pgbackrest/pgbackrest] PR #2906 (approved): Support alternate relation segment sizes for page checksums.
@ 2026-11-06 15:41  "pgstef (@pgstef)" <noreply+pgstef@github.com>
  15 siblings, 0 replies; 17+ messages in thread

From: pgstef (@pgstef) @ 2026-11-06 15:41 UTC (permalink / raw)
  To: pgbackrest/pgbackrest <noreply+pgbackrest-pgbackrest@github.com>

Looks good to me too!

One small nitpick: could we mention in the release notes that backups taken with earlier versions on clusters using a non-default relation segment size may contain false page checksum errors in their manifests? Since 2.60.0, these could cause `restore` to fail or `expire` to refuse expiration, requiring `checksum-page-error=n` for affected backups.

For segment sizes larger than 1GB, earlier backups may also have missed actual page corruption beyond the first 1GB of each segment file.

^ permalink  raw  [nested|flat] 17+ messages in thread

* Re: [pgbackrest/pgbackrest] PR #2906: Support alternate relation segment sizes for page checksums.
@ 2026-11-09 09:58  "dwsteele (@dwsteele)" <noreply+dwsteele@github.com>
  15 siblings, 0 replies; 17+ messages in thread

From: dwsteele (@dwsteele) @ 2026-11-09 09:58 UTC (permalink / raw)
  To: pgbackrest/pgbackrest <noreply+pgbackrest-pgbackrest@github.com>

@dwsteele applied label "bug".

^ permalink  raw  [nested|flat] 17+ messages in thread

* Re: [pgbackrest/pgbackrest] PR #2906: Support alternate relation segment sizes for page checksums.
@ 2026-11-09 09:58  "dwsteele (@dwsteele)" <noreply+dwsteele@github.com>
  15 siblings, 0 replies; 17+ messages in thread

From: dwsteele (@dwsteele) @ 2026-11-09 09:58 UTC (permalink / raw)
  To: pgbackrest/pgbackrest <noreply+pgbackrest-pgbackrest@github.com>

@dwsteele removed label "enhancement".

^ permalink  raw  [nested|flat] 17+ messages in thread

* Re: [pgbackrest/pgbackrest] PR #2906: Support alternate relation segment sizes for page checksums.
@ 2026-11-09 10:03  "dwsteele (@dwsteele)" <noreply+dwsteele@github.com>
  15 siblings, 0 replies; 17+ messages in thread

From: dwsteele (@dwsteele) @ 2026-11-09 10:03 UTC (permalink / raw)
  To: pgbackrest/pgbackrest <noreply+pgbackrest-pgbackrest@github.com>

> One small nitpick: could we mention in the release notes that backups taken with earlier versions on clusters using a non-default relation segment size may contain false page checksum errors in their manifests?

@pgstef Good idea. I added this to release note but it does not get a separate "important note" in the release since I judge that it affects a small number of users.

^ permalink  raw  [nested|flat] 17+ messages in thread

* Re: [pgbackrest/pgbackrest] PR #2906: Support alternate relation segment sizes for page checksums.
@ 2026-11-09 10:07  "pgstef (@pgstef)" <noreply+pgstef@github.com>
  15 siblings, 0 replies; 17+ messages in thread

From: pgstef (@pgstef) @ 2026-11-09 10:07 UTC (permalink / raw)
  To: pgbackrest/pgbackrest <noreply+pgbackrest-pgbackrest@github.com>

> I added this to release note but it does not get a separate "important note" in the release since I judge that it affects a small number of users.

Works for me, thanks :+1: 


^ permalink  raw  [nested|flat] 17+ messages in thread

* Re: [pgbackrest/pgbackrest] PR #2906: Support alternate relation segment sizes for page checksums.
@ 2026-11-09 10:27  "dwsteele (@dwsteele)" <noreply+dwsteele@github.com>
  15 siblings, 0 replies; 17+ messages in thread

From: dwsteele (@dwsteele) @ 2026-11-09 10:27 UTC (permalink / raw)
  To: pgbackrest/pgbackrest <noreply+pgbackrest-pgbackrest@github.com>

@dwsteele merged this (commit 9d3de51a6d45).

^ permalink  raw  [nested|flat] 17+ messages in thread

* Re: [pgbackrest/pgbackrest] PR #2906: Support alternate relation segment sizes for page checksums.
@ 2026-11-09 10:27  "dwsteele (@dwsteele)" <noreply+dwsteele@github.com>
  15 siblings, 0 replies; 17+ messages in thread

From: dwsteele (@dwsteele) @ 2026-11-09 10:27 UTC (permalink / raw)
  To: pgbackrest/pgbackrest <noreply+pgbackrest-pgbackrest@github.com>

@dwsteele closed this.

^ permalink  raw  [nested|flat] 17+ messages in thread

* Re: [pgbackrest/pgbackrest] PR #2906: Support alternate relation segment sizes for page checksums.
@ 2026-11-13 15:11  "dwsteele (@dwsteele)" <noreply+dwsteele@github.com>
  15 siblings, 0 replies; 17+ messages in thread

From: dwsteele (@dwsteele) @ 2026-11-13 15:11 UTC (permalink / raw)
  To: pgbackrest/pgbackrest <noreply+pgbackrest-pgbackrest@github.com>

@hunleyd @pgstef Considering including this in 2.60.1. There is a workaround but since the restore errors are new to 2.60.0 because of the new error, it could be argued this is a new bug. Thoughts?

^ permalink  raw  [nested|flat] 17+ messages in thread

* Re: [pgbackrest/pgbackrest] PR #2906: Support alternate relation segment sizes for page checksums.
@ 2026-11-13 15:16  "pgstef (@pgstef)" <noreply+pgstef@github.com>
  15 siblings, 0 replies; 17+ messages in thread

From: pgstef (@pgstef) @ 2026-11-13 15:16 UTC (permalink / raw)
  To: pgbackrest/pgbackrest <noreply+pgbackrest-pgbackrest@github.com>

> Considering including this in 2.60.1. There is a workaround but since the restore errors are new to 2.60.0 because of the new error, it could be argued this is a new bug. Thoughts?

I'm OK with both as I don't have a strong opinion on either choice. The workaround "rule" looks good enough for me and not so many users did complain about it, so 2.61 doesn't seem very far away tbh.

^ permalink  raw  [nested|flat] 17+ messages in thread

* Re: [pgbackrest/pgbackrest] PR #2906: Support alternate relation segment sizes for page checksums.
@ 2026-11-13 15:28  "dwsteele (@dwsteele)" <noreply+dwsteele@github.com>
  15 siblings, 0 replies; 17+ messages in thread

From: dwsteele (@dwsteele) @ 2026-11-13 15:28 UTC (permalink / raw)
  To: pgbackrest/pgbackrest <noreply+pgbackrest-pgbackrest@github.com>

> The workaround "rule" looks good enough for me and not so many users did complain about it

Hmm, yeah, this is a good point that I was forgetting. Very, very few users should be affected.

^ permalink  raw  [nested|flat] 17+ messages in thread

* Re: [pgbackrest/pgbackrest] PR #2906: Support alternate relation segment sizes for page checksums.
@ 2026-11-13 16:31  "dwsteele (@dwsteele)" <noreply+dwsteele@github.com>
  15 siblings, 0 replies; 17+ messages in thread

From: dwsteele (@dwsteele) @ 2026-11-13 16:31 UTC (permalink / raw)
  To: pgbackrest/pgbackrest <noreply+pgbackrest-pgbackrest@github.com>

I prepared the patch release without this PR.

^ permalink  raw  [nested|flat] 17+ messages in thread


end of thread, other threads:[~2026-11-13 16:31 UTC | newest]

Thread overview: 17+ messages (download: mbox mbox.gz follow: Atom feed)
-- links below jump to the message on this page --
2026-11-06 11:24 [pgbackrest/pgbackrest] PR #2906: Support alternate relation segment sizes for page checksums. "dwsteele (@dwsteele)" <noreply+dwsteele@github.com>
2026-11-06 11:24 ` "dwsteele (@dwsteele)" <noreply+dwsteele@github.com>
2026-11-06 11:24 ` "dwsteele (@dwsteele)" <noreply+dwsteele@github.com>
2026-11-06 11:24 ` "dwsteele (@dwsteele)" <noreply+dwsteele@github.com>
2026-11-06 11:24 ` "dwsteele (@dwsteele)" <noreply+dwsteele@github.com>
2026-11-06 15:07 ` Re: [pgbackrest/pgbackrest] PR #2906 (approved): Support alternate relation segment sizes for page checksums. "hunleyd (@hunleyd)" <noreply+hunleyd@github.com>
2026-11-06 15:41 ` Re: [pgbackrest/pgbackrest] PR #2906 (approved): Support alternate relation segment sizes for page checksums. "pgstef (@pgstef)" <noreply+pgstef@github.com>
2026-11-09 09:58 ` "dwsteele (@dwsteele)" <noreply+dwsteele@github.com>
2026-11-09 09:58 ` "dwsteele (@dwsteele)" <noreply+dwsteele@github.com>
2026-11-09 10:03 ` "dwsteele (@dwsteele)" <noreply+dwsteele@github.com>
2026-11-09 10:07 ` "pgstef (@pgstef)" <noreply+pgstef@github.com>
2026-11-09 10:27 ` "dwsteele (@dwsteele)" <noreply+dwsteele@github.com>
2026-11-09 10:27 ` "dwsteele (@dwsteele)" <noreply+dwsteele@github.com>
2026-11-13 15:11 ` "dwsteele (@dwsteele)" <noreply+dwsteele@github.com>
2026-11-13 15:16 ` "pgstef (@pgstef)" <noreply+pgstef@github.com>
2026-11-13 15:28 ` "dwsteele (@dwsteele)" <noreply+dwsteele@github.com>
2026-11-13 16:31 ` "dwsteele (@dwsteele)" <noreply+dwsteele@github.com>

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