Resolve issues #5 and #1: reduce number of collisions in the ptrack map

.gitignore

@@ -1,6 +1,4 @@
 .deps
 *.so
 *.o
-ptrack--2.0.sql
 Dockerfile
-

Makefile

@@ -2,13 +2,11 @@
 
 MODULE_big = ptrack
 OBJS = ptrack.o datapagemap.o engine.o $(WIN32RES)
-EXTENSION = ptrack
-EXTVERSION = 2.1
-DATA = ptrack.sql ptrack--2.0--2.1.sql
-DATA_built = $(EXTENSION)--$(EXTVERSION).sql
 PGFILEDESC = "ptrack - block-level incremental backup engine"
 
-EXTRA_CLEAN = $(EXTENSION)--$(EXTVERSION).sql
+EXTENSION = ptrack
+EXTVERSION = 2.2
+DATA = ptrack--2.1.sql ptrack--2.0--2.1.sql ptrack--2.1--2.2.sql
 
 TAP_TESTS = 1
 
@@ -22,13 +20,3 @@ top_builddir = ../..
 include $(top_builddir)/src/Makefile.global
 include $(top_srcdir)/contrib/contrib-global.mk
 endif
-
-$(EXTENSION)--$(EXTVERSION).sql: ptrack.sql
-	cat $^ > $@
-
-# temp-install: EXTRA_INSTALL=contrib/ptrack
-
-# check-tap: temp-install
-# 	$(prove_check)
-
-# check: check-tap

README.md

@@ -65,7 +65,8 @@ To disable `ptrack` and clean up all remaining service files set `ptrack.map_siz
 
  * ptrack_version() — returns ptrack version string.
  * ptrack_init_lsn() — returns LSN of the last ptrack map initialization.
- * ptrack_get_pagemapset('LSN') — returns a set of changed data files with bitmaps of changed blocks since specified LSN.
+ * ptrack_get_pagemapset(start_lsn pg_lsn) — returns a set of changed data files with bitmaps of changed blocks since specified `start_lsn`.
+ * ptrack_get_change_stat(start_lsn pg_lsn) — returns statistic of changes (number of files, pages and size in MB) since specified `start_lsn`.
 
 Usage example:
 
@@ -102,6 +103,10 @@ Usually, you have to only install new version of `ptrack` and do `ALTER EXTENSIO
 * Do `ALTER EXTENSION 'ptrack' UPDATE;`.
 * Restart your server.
 
+#### Upgrading from 2.1.* to 2.2.*:
+
+Since version 2.2 we use a different algorithm for tracking changed pages. Thus, data recorded in the `ptrack.map` using pre 2.2 versions of `ptrack` is incompatible with newer versions. After extension upgrade and server restart old `ptrack.map` will be discarded with `WARNING` and initialized from the scratch.
+
 ## Limitations
 
 1. You can only use `ptrack` safely with `wal_level >= 'replica'`. Otherwise, you can lose tracking of some changes if crash-recovery occurs, since [certain commands are designed not to write WAL at all if wal_level is minimal](https://www.postgresql.org/docs/12/populate.html#POPULATE-PITR), but we only durably flush `ptrack` map at checkpoint time.

engine.c

@@ -156,6 +156,8 @@ ptrackMapInit(void)
 	sprintf(ptrack_path, "%s/%s", DataDir, PTRACK_PATH);
 	sprintf(ptrack_mmap_path, "%s/%s", DataDir, PTRACK_MMAP_PATH);
 
+ptrack_map_reinit:
+
 	/* Remove old PTRACK_MMAP_PATH file, if exists */
 	if (ptrack_file_exists(ptrack_mmap_path))
 		durable_unlink(ptrack_mmap_path, LOG);
@@ -175,18 +177,15 @@ ptrackMapInit(void)
 	if (stat(ptrack_path, &stat_buf) == 0)
 	{
 		copy_file(ptrack_path, ptrack_mmap_path);
-		is_new_map = false;		/* flag to check checksum */
+		is_new_map = false;		/* flag to check map file format and checksum */
 		ptrack_fd = BasicOpenFile(ptrack_mmap_path, O_RDWR | PG_BINARY);
-		if (ptrack_fd < 0)
-			elog(ERROR, "ptrack init: failed to open map file \"%s\": %m", ptrack_mmap_path);
 	}
 	else
-	{
 		/* Create new file for PTRACK_MMAP_PATH */
 		ptrack_fd = BasicOpenFile(ptrack_mmap_path, O_RDWR | O_CREAT | PG_BINARY);
-		if (ptrack_fd < 0)
-			elog(ERROR, "ptrack init: failed to open map file \"%s\": %m", ptrack_mmap_path);
-	}
+
+	if (ptrack_fd < 0)
+		elog(ERROR, "ptrack init: failed to open map file \"%s\": %m", ptrack_mmap_path);
 
 #ifdef WIN32
 	{
@@ -227,7 +226,20 @@ ptrackMapInit(void)
 			elog(ERROR, "ptrack init: wrong map format of file \"%s\"", ptrack_path);
 
 		/* Check ptrack version inside old ptrack map */
-		/* No-op for now, but may be used for future compatibility checks */
+		if (ptrack_map->version_num != PTRACK_VERSION_NUM)
+		{
+			ereport(WARNING,
+					(errcode(ERRCODE_DATA_CORRUPTED),
+					 errmsg("ptrack init: map format version %d in the file \"%s\" is incompatible with loaded version %d",
+							ptrack_map->version_num, ptrack_path, PTRACK_VERSION_NUM),
+					 errdetail("Deleting file \"%s\" and reinitializing ptrack map.", ptrack_path)));
+
+			/* Clean up everything and try again */
+			ptrackCleanFilesAndMap();
+
+			is_new_map = true;
+			goto ptrack_map_reinit;
+		}
 
 		/* Check CRC */
 		INIT_CRC32C(crc);
@@ -378,7 +390,7 @@ ptrackCheckpoint(void)
 	/*
 	 * We are writing ptrack map values to file, but we want to simply map it
 	 * into the memory with mmap after a crash/restart. That way, we have to
-	 * write values taking into account all paddings/allignments.
+	 * write values taking into account all paddings/alignments.
 	 *
 	 * Write both magic and varsion_num at once.
 	 */
@@ -435,7 +447,7 @@ ptrackCheckpoint(void)
 												* going to overflow. */
 
 			/*
-			 * We should not have any allignment issues here, since sizeof()
+			 * We should not have any alignment issues here, since sizeof()
 			 * takes into account all paddings for us.
 			 */
 			ptrack_write_chunk(ptrack_tmp_fd, &crc, (char *) buf, writesz);
@@ -446,7 +458,7 @@ ptrackCheckpoint(void)
 		}
 	}
 
-	/* Write if anythig left */
+	/* Write if anything left */
 	if ((i + 1) % PTRACK_BUF_SIZE != 0)
 	{
 		size_t		writesz = sizeof(pg_atomic_uint64) * j;
@@ -641,48 +653,56 @@ void
 ptrack_mark_block(RelFileNodeBackend smgr_rnode,
 				  ForkNumber forknum, BlockNumber blocknum)
 {
+	PtBlockId	bid;
 	size_t		hash;
+	size_t		slot1;
+	size_t		slot2;
 	XLogRecPtr	new_lsn;
-	PtBlockId	bid;
 	/*
 	 * We use pg_atomic_uint64 here only for alignment purposes, because
-	 * pg_atomic_uint64 is forcely aligned on 8 bytes during the MSVC build.
+	 * pg_atomic_uint64 is forcedly aligned on 8 bytes during the MSVC build.
 	 */
 	pg_atomic_uint64	old_lsn;
 	pg_atomic_uint64	old_init_lsn;
 
-	if (ptrack_map_size != 0 && (ptrack_map != NULL) &&
-		smgr_rnode.backend == InvalidBackendId) /* do not track temporary
-												 * relations */
-	{
-		bid.relnode = smgr_rnode.node;
-		bid.forknum = forknum;
-		bid.blocknum = blocknum;
-		hash = BID_HASH_FUNC(bid);
-
-		if (RecoveryInProgress())
-			new_lsn = GetXLogReplayRecPtr(NULL);
-		else
-			new_lsn = GetXLogInsertRecPtr();
-
-		old_lsn.value = pg_atomic_read_u64(&ptrack_map->entries[hash]);
+	if (ptrack_map_size == 0
+		|| ptrack_map == NULL
+		|| smgr_rnode.backend != InvalidBackendId) /* do not track temporary
+													* relations */
+		return;
 
-		/* Atomically assign new init LSN value */
-		old_init_lsn.value = pg_atomic_read_u64(&ptrack_map->init_lsn);
+	bid.relnode = smgr_rnode.node;
+	bid.forknum = forknum;
+	bid.blocknum = blocknum;
 
-		if (old_init_lsn.value == InvalidXLogRecPtr)
-		{
-			elog(DEBUG1, "ptrack_mark_block: init_lsn " UINT64_FORMAT " <- " UINT64_FORMAT, old_init_lsn.value, new_lsn);
+	hash = BID_HASH_FUNC(bid);
+	slot1 = hash % PtrackContentNblocks;
+	slot2 = ((hash << 32) | (hash >> 32)) % PtrackContentNblocks;
 
-			while (old_init_lsn.value < new_lsn &&
-				   !pg_atomic_compare_exchange_u64(&ptrack_map->init_lsn, (uint64 *) &old_init_lsn.value, new_lsn));
-		}
+	if (RecoveryInProgress())
+		new_lsn = GetXLogReplayRecPtr(NULL);
+	else
+		new_lsn = GetXLogInsertRecPtr();
 
-		elog(DEBUG3, "ptrack_mark_block: map[%zu]=" UINT64_FORMAT " <- " UINT64_FORMAT, hash, old_lsn.value, new_lsn);
+	/* Atomically assign new init LSN value */
+	old_init_lsn.value = pg_atomic_read_u64(&ptrack_map->init_lsn);
+	if (old_init_lsn.value == InvalidXLogRecPtr)
+	{
+		elog(DEBUG1, "ptrack_mark_block: init_lsn " UINT64_FORMAT " <- " UINT64_FORMAT, old_init_lsn.value, new_lsn);
 
-		/* Atomically assign new LSN value */
-		while (old_lsn.value < new_lsn &&
-			   !pg_atomic_compare_exchange_u64(&ptrack_map->entries[hash], (uint64 *) &old_lsn.value, new_lsn));
-		elog(DEBUG3, "ptrack_mark_block: map[%zu]=" UINT64_FORMAT, hash, pg_atomic_read_u64(&ptrack_map->entries[hash]));
+		while (old_init_lsn.value < new_lsn &&
+			   !pg_atomic_compare_exchange_u64(&ptrack_map->init_lsn, (uint64 *) &old_init_lsn.value, new_lsn));
 	}
+
+	/* Atomically assign new LSN value to the first slot */
+	old_lsn.value = pg_atomic_read_u64(&ptrack_map->entries[slot1]);
+	elog(DEBUG3, "ptrack_mark_block: map[%zu]=" UINT64_FORMAT " <- " UINT64_FORMAT, slot1, old_lsn.value, new_lsn);
+	while (old_lsn.value < new_lsn &&
+		   !pg_atomic_compare_exchange_u64(&ptrack_map->entries[slot1], (uint64 *) &old_lsn.value, new_lsn));
+	elog(DEBUG3, "ptrack_mark_block: map[%zu]=" UINT64_FORMAT, hash, pg_atomic_read_u64(&ptrack_map->entries[slot1]));
+
+	/* And to the second */
+	old_lsn.value = pg_atomic_read_u64(&ptrack_map->entries[slot2]);
+	while (old_lsn.value < new_lsn &&
+		   !pg_atomic_compare_exchange_u64(&ptrack_map->entries[slot2], (uint64 *) &old_lsn.value, new_lsn));
 }

engine.h

@@ -50,7 +50,7 @@ typedef struct PtrackMapHdr
 {
 	/*
 	 * Three magic bytes (+ \0) to be sure, that we are reading ptrack.map
-	 * with a right PtrackMapHdr strucutre.
+	 * with a right PtrackMapHdr structure.
 	 */
 	char		magic[PTRACK_MAGIC_SIZE];
 
@@ -72,7 +72,6 @@ typedef struct PtrackMapHdr
 
 typedef PtrackMapHdr * PtrackMap;
 
-/* TODO: check MAXALIGN usage below */
 /* Number of elements in ptrack map (LSN array)  */
 #define PtrackContentNblocks \
 		((ptrack_map_size - offsetof(PtrackMapHdr, entries) - sizeof(pg_crc32c)) / sizeof(pg_atomic_uint64))
@@ -84,9 +83,10 @@ typedef PtrackMapHdr * PtrackMap;
 /* CRC32 value offset in order to directly access it in the mmap'ed memory chunk */
 #define PtrackCrcOffset (PtrackActualSize - sizeof(pg_crc32c))
 
-/* Map block address 'bid' to map slot */
+/* Block address 'bid' to hash.  To get slot position in map should be divided
+ * with '% PtrackContentNblocks' */
 #define BID_HASH_FUNC(bid) \
-		(size_t)(DatumGetUInt64(hash_any_extended((unsigned char *)&bid, sizeof(bid), 0)) % PtrackContentNblocks)
+		(size_t)(DatumGetUInt64(hash_any_extended((unsigned char *)&bid, sizeof(bid), 0)))
 
 /*
  * Per process pointer to shared ptrack_map

ptrack--2.1--2.2.sql

@@ -0,0 +1,35 @@
+/* ptrack/ptrack--2.1--2.2.sql */
+
+-- Complain if script is sourced in psql, rather than via ALTER EXTENSION
+\echo Use "ALTER EXTENSION ptrack UPDATE;" to load this file.\ quit
+
+DROP FUNCTION ptrack_get_pagemapset(start_lsn pg_lsn);
+CREATE FUNCTION ptrack_get_pagemapset(start_lsn pg_lsn)
+RETURNS TABLE (path			text,
+			   pagecount	bigint,
+			   pagemap		bytea)
+AS 'MODULE_PATHNAME'
+LANGUAGE C STRICT VOLATILE;
+
+CREATE FUNCTION ptrack_get_change_stat(start_lsn pg_lsn)
+	RETURNS TABLE (
+		files bigint,
+		pages numeric,
+		"size, MB" numeric
+	) AS
+$func$
+DECLARE
+block_size bigint;
+BEGIN
+	block_size := (SELECT setting FROM pg_settings WHERE name = 'block_size');
+
+	RETURN QUERY
+	SELECT changed_files,
+		   changed_pages,
+		   block_size * changed_pages / (1024.0 * 1024)
+	FROM
+		(SELECT count(path) AS changed_files,
+				sum(pagecount) AS changed_pages
+		 FROM ptrack_get_pagemapset(start_lsn)) s;
+END
+$func$ LANGUAGE plpgsql;

ptrack.sql → ptrack--2.1.sql

@@ -1,3 +1,5 @@
+/* ptrack/ptrack--2.1.sql */
+
 -- Complain if script is sourced in psql, rather than via CREATE EXTENSION
 \echo Use "CREATE EXTENSION ptrack" to load this file. \quit