On Tue, Jul 6, 2021 at 3:00 PM Dilip Kumar <dilipbal...@gmail.com> wrote: > > On Fri, Jun 18, 2021 at 12:50 AM Robert Haas <robertmh...@gmail.com> wrote: > > > > On Thu, Jun 17, 2021 at 2:48 PM Andres Freund <and...@anarazel.de> wrote: > > > Or do you mean that looking at the filesystem at all is bypassing shared > > > buffers? > > > > This is what I mean. I think we will end up in a better spot if we can > > avoid doing that without creating too much ugliness elsewhere. > > > > The patch was not getting applied on head so I have rebased it, along > with that now I have used bufmgr layer for writing writing/logging > destination pages as well instead of directly using sgmr layer.
I have done further cleanup of the patch and also divided it into 3 patches. 0001 - Currently, write_relmap_file and load_relmap_file are tightly coupled with shared_map and local_map. As part of the higher level patch set we need remap read/write interfaces that are not dependent upon shared_map and local_map, and we should be able to pass map memory as an external parameter instead. 0002- Support new interfaces in relmapper, 1) Support copying the relmap file from one database path to the other database path. 2) Like RelationMapOidToFilenode, provide another interface which do the same but instead of getting it for the database we are connected to it will get it for the input database path. These interfaces are required for the next patch for supporting the wal logged created database. 0003- The main patch for WAL logging the created database operation. -- Regards, Dilip Kumar EnterpriseDB: http://www.enterprisedb.com
From 2f6a5341ef7f97fc258fb7d108129a824678fd75 Mon Sep 17 00:00:00 2001 From: Dilip Kumar <dilipkumar@localhost.localdomain> Date: Wed, 1 Sep 2021 14:06:29 +0530 Subject: [PATCH v2 1/3] Refactor relmap load and relmap write functions Currently, write_relmap_file and load_relmap_file are tightly coupled with shared_map and local_map. As part of the higher level patch set we need remap read/write interfaces that are not dependent upon shared_map and local_map, and we should be able to pass map memory as an external parameter instead. --- src/backend/utils/cache/relmapper.c | 163 ++++++++++++++++++++++-------------- 1 file changed, 102 insertions(+), 61 deletions(-) diff --git a/src/backend/utils/cache/relmapper.c b/src/backend/utils/cache/relmapper.c index a6e38ad..ae62910 100644 --- a/src/backend/utils/cache/relmapper.c +++ b/src/backend/utils/cache/relmapper.c @@ -136,6 +136,12 @@ static void apply_map_update(RelMapFile *map, Oid relationId, Oid fileNode, bool add_okay); static void merge_map_updates(RelMapFile *map, const RelMapFile *updates, bool add_okay); +static void read_relmap_file(char *mapfilename, RelMapFile *map, + bool lock_held); +static void write_relmap_file_internal(char *mapfilename, RelMapFile *newmap, + bool write_wal, bool send_sinval, + bool preserve_files, Oid dbid, Oid tsid, + const char *dbpath); static void load_relmap_file(bool shared, bool lock_held); static void write_relmap_file(bool shared, RelMapFile *newmap, bool write_wal, bool send_sinval, bool preserve_files, @@ -692,31 +698,17 @@ RestoreRelationMap(char *startAddress) * Because the map file is essential for access to core system catalogs, * failure to read it is a fatal error. * - * Note that the local case requires DatabasePath to be set up. + * lock_held, pass true if caller already have the relation mapping or higher + * level lock. */ static void -load_relmap_file(bool shared, bool lock_held) +read_relmap_file(char *mapfilename, RelMapFile *map, bool lock_held) { - RelMapFile *map; - char mapfilename[MAXPGPATH]; pg_crc32c crc; int fd; int r; - if (shared) - { - snprintf(mapfilename, sizeof(mapfilename), "global/%s", - RELMAPPER_FILENAME); - map = &shared_map; - } - else - { - snprintf(mapfilename, sizeof(mapfilename), "%s/%s", - DatabasePath, RELMAPPER_FILENAME); - map = &local_map; - } - - /* Read data ... */ + /* Open the relmap file for reading. */ fd = OpenTransientFile(mapfilename, O_RDONLY | PG_BINARY); if (fd < 0) ereport(FATAL, @@ -779,62 +771,53 @@ load_relmap_file(bool shared, bool lock_held) } /* - * Write out a new shared or local map file with the given contents. - * - * The magic number and CRC are automatically updated in *newmap. On - * success, we copy the data to the appropriate permanent static variable. - * - * If write_wal is true then an appropriate WAL message is emitted. - * (It will be false for bootstrap and WAL replay cases.) - * - * If send_sinval is true then a SI invalidation message is sent. - * (This should be true except in bootstrap case.) + * load_relmap_file -- load data from the shared or local map file * - * If preserve_files is true then the storage manager is warned not to - * delete the files listed in the map. + * Because the map file is essential for access to core system catalogs, + * failure to read it is a fatal error. * - * Because this may be called during WAL replay when MyDatabaseId, - * DatabasePath, etc aren't valid, we require the caller to pass in suitable - * values. The caller is also responsible for being sure no concurrent - * map update could be happening. + * Note that the local case requires DatabasePath to be set up. */ static void -write_relmap_file(bool shared, RelMapFile *newmap, - bool write_wal, bool send_sinval, bool preserve_files, - Oid dbid, Oid tsid, const char *dbpath) +load_relmap_file(bool shared, bool lock_held) { - int fd; - RelMapFile *realmap; + RelMapFile *map; char mapfilename[MAXPGPATH]; - /* - * Fill in the overhead fields and update CRC. - */ - newmap->magic = RELMAPPER_FILEMAGIC; - if (newmap->num_mappings < 0 || newmap->num_mappings > MAX_MAPPINGS) - elog(ERROR, "attempt to write bogus relation mapping"); - - INIT_CRC32C(newmap->crc); - COMP_CRC32C(newmap->crc, (char *) newmap, offsetof(RelMapFile, crc)); - FIN_CRC32C(newmap->crc); - - /* - * Open the target file. We prefer to do this before entering the - * critical section, so that an open() failure need not force PANIC. - */ if (shared) { snprintf(mapfilename, sizeof(mapfilename), "global/%s", RELMAPPER_FILENAME); - realmap = &shared_map; + map = &shared_map; } else { snprintf(mapfilename, sizeof(mapfilename), "%s/%s", - dbpath, RELMAPPER_FILENAME); - realmap = &local_map; + DatabasePath, RELMAPPER_FILENAME); + map = &local_map; } + /* Read data ... */ + read_relmap_file(mapfilename, map, lock_held); +} + +/* + * Helper function for write_relmap_file, Read comments atop write_relmap_file + * for more details. The CRC should be computed by the caller and stored in + * the newmap. + */ +static void +write_relmap_file_internal(char *mapfilename, RelMapFile *newmap, + bool write_wal, bool send_sinval, + bool preserve_files, Oid dbid, Oid tsid, + const char *dbpath) +{ + int fd; + + /* + * Open the target file. We prefer to do this before entering the + * critical section, so that an open() failure need not force PANIC. + */ fd = OpenTransientFile(mapfilename, O_WRONLY | O_CREAT | PG_BINARY); if (fd < 0) ereport(ERROR, @@ -934,6 +917,68 @@ write_relmap_file(bool shared, RelMapFile *newmap, } } + /* Critical section done */ + if (write_wal) + END_CRIT_SECTION(); +} + +/* + * Write out a new shared or local map file with the given contents. + * + * The magic number and CRC are automatically updated in *newmap. On + * success, we copy the data to the appropriate permanent static variable. + * + * If write_wal is true then an appropriate WAL message is emitted. + * (It will be false for bootstrap and WAL replay cases.) + * + * If send_sinval is true then a SI invalidation message is sent. + * (This should be true except in bootstrap case.) + * + * If preserve_files is true then the storage manager is warned not to + * delete the files listed in the map. + * + * Because this may be called during WAL replay when MyDatabaseId, + * DatabasePath, etc aren't valid, we require the caller to pass in suitable + * values. The caller is also responsible for being sure no concurrent + * map update could be happening. + */ +static void +write_relmap_file(bool shared, RelMapFile *newmap, + bool write_wal, bool send_sinval, bool preserve_files, + Oid dbid, Oid tsid, const char *dbpath) +{ + RelMapFile *realmap; + char mapfilename[MAXPGPATH]; + + /* + * Fill in the overhead fields and update CRC. + */ + newmap->magic = RELMAPPER_FILEMAGIC; + if (newmap->num_mappings < 0 || newmap->num_mappings > MAX_MAPPINGS) + elog(ERROR, "attempt to write bogus relation mapping"); + + INIT_CRC32C(newmap->crc); + COMP_CRC32C(newmap->crc, (char *) newmap, offsetof(RelMapFile, crc)); + FIN_CRC32C(newmap->crc); + + if (shared) + { + snprintf(mapfilename, sizeof(mapfilename), "global/%s", + RELMAPPER_FILENAME); + realmap = &shared_map; + } + else + { + snprintf(mapfilename, sizeof(mapfilename), "%s/%s", + dbpath, RELMAPPER_FILENAME); + realmap = &local_map; + } + + /* Write the map to the relmap file. */ + write_relmap_file_internal(mapfilename, newmap, write_wal, + send_sinval, preserve_files, dbid, tsid, + dbpath); + /* * Success, update permanent copy. During bootstrap, we might be working * on the permanent copy itself, in which case skip the memcpy() to avoid @@ -943,10 +988,6 @@ write_relmap_file(bool shared, RelMapFile *newmap, memcpy(realmap, newmap, sizeof(RelMapFile)); else Assert(!send_sinval); /* must be bootstrapping */ - - /* Critical section done */ - if (write_wal) - END_CRIT_SECTION(); } /* -- 1.8.3.1
From 50761c0caf60b17be99c4ba063369d74e75c77fa Mon Sep 17 00:00:00 2001 From: Dilip Kumar <dilipkumar@localhost.localdomain> Date: Wed, 1 Sep 2021 14:16:35 +0530 Subject: [PATCH v2 2/3] Extend relmap interfaces Support new interfaces in relmapper, 1) Support copying the relmap file from one database path to the other database path. 2) Like RelationMapOidToFilenode, provide another interface which do the same but instead of getting it for the database we are connected to it will get it for the input database path. These interfaces are required for next patch for supporting the wal logged created database. --- src/backend/utils/cache/relmapper.c | 131 +++++++++++++++++++++++++++++++----- src/include/utils/relmapper.h | 6 +- 2 files changed, 119 insertions(+), 18 deletions(-) diff --git a/src/backend/utils/cache/relmapper.c b/src/backend/utils/cache/relmapper.c index ae62910..182054e 100644 --- a/src/backend/utils/cache/relmapper.c +++ b/src/backend/utils/cache/relmapper.c @@ -141,7 +141,7 @@ static void read_relmap_file(char *mapfilename, RelMapFile *map, static void write_relmap_file_internal(char *mapfilename, RelMapFile *newmap, bool write_wal, bool send_sinval, bool preserve_files, Oid dbid, Oid tsid, - const char *dbpath); + const char *dbpath, bool create); static void load_relmap_file(bool shared, bool lock_held); static void write_relmap_file(bool shared, RelMapFile *newmap, bool write_wal, bool send_sinval, bool preserve_files, @@ -256,6 +256,40 @@ RelationMapFilenodeToOid(Oid filenode, bool shared) } /* + * RelationMapOidToFilenodeForDatabase + * + * Find relfilenode for the given relation id in the dbpath. Returns + * InvalidOid if the relationId is not found in the relmap. + * + * This function is only called during CREATE DATABASE command, so we can pass + * lock_held as true while reading the relmap file since we are already holding + * the exclusive lock on the database. + */ +Oid +RelationMapOidToFilenodeForDatabase(char *dbpath, Oid relationId) +{ + RelMapFile map; + int i; + char mapfilename[MAXPGPATH]; + + /* Relmap file path for the given dbpath. */ + snprintf(mapfilename, sizeof(mapfilename), "%s/%s", + dbpath, RELMAPPER_FILENAME); + + /* Read the relmap file from the source database. */ + read_relmap_file(mapfilename, &map, true); + + /* Iterate over the relmap entries to find the input relation oid. */ + for (i = 0; i < map.num_mappings; i++) + { + if (relationId == map.mappings[i].mapoid) + return map.mappings[i].mapfilenode; + } + + return InvalidOid; +} + +/* * RelationMapUpdateMap * * Install a new relfilenode mapping for the specified relation. @@ -693,10 +727,47 @@ RestoreRelationMap(char *startAddress) } /* - * load_relmap_file -- load data from the shared or local map file + * CopyRelationMap * - * Because the map file is essential for access to core system catalogs, - * failure to read it is a fatal error. + * Copy relmapfile from source db path to the destination db path and WAL log + * the operation. This function is only called during the create database so + * the caller must hold the exclusive lock on the source database. Destination + * database is not yet created so we don't have any issue. + */ +void +CopyRelationMap(Oid dbid, Oid tsid, char *srcdbpath, char *dstdbpath) +{ + RelMapFile map; + char mapfilename[MAXPGPATH]; + + /* Relmap file path of the source database. */ + snprintf(mapfilename, sizeof(mapfilename), "%s/%s", + srcdbpath, RELMAPPER_FILENAME); + + /* + * Read the relmap file from the source database. We are not connected to + * the database so we can not take the relmap lock, but we are already + * holding exclusive lock on the database so pass lock_held as true. + */ + read_relmap_file(mapfilename, &map, true); + + /* Relmap file path of the destination database. */ + snprintf(mapfilename, sizeof(mapfilename), "%s/%s", + dstdbpath, RELMAPPER_FILENAME); + + /* + * Write map contents into the destination database's relmap file. + * write_relmap_file_internal, expects that the CRC should have been + * computed and stored in the input map. But, since we have read this map + * from the source database and directly writing to the destination file + * without updating it so we don't need to recompute it. + */ + write_relmap_file_internal(mapfilename, &map, true, false, true, dbid, + tsid, dstdbpath, true); +} + +/* + * read_relmap_file - read the relmap file data. * * lock_held, pass true if caller already have the relation mapping or higher * level lock. @@ -802,15 +873,18 @@ load_relmap_file(bool shared, bool lock_held) } /* - * Helper function for write_relmap_file, Read comments atop write_relmap_file - * for more details. The CRC should be computed by the caller and stored in - * the newmap. + * Helper function for write_relmap_file and CopyRelationMap, Read comments + * atop write_relmap_file for more details. The CRC should be computed by the + * caller and stored in the newmap. + * + * Pass the create = true, if we are copying the relmap file during CREATE + * DATABASE command. */ static void write_relmap_file_internal(char *mapfilename, RelMapFile *newmap, bool write_wal, bool send_sinval, bool preserve_files, Oid dbid, Oid tsid, - const char *dbpath) + const char *dbpath, bool create) { int fd; @@ -836,6 +910,7 @@ write_relmap_file_internal(char *mapfilename, RelMapFile *newmap, xlrec.dbid = dbid; xlrec.tsid = tsid; xlrec.nbytes = sizeof(RelMapFile); + xlrec.create = create; XLogBeginInsert(); XLogRegisterData((char *) (&xlrec), MinSizeOfRelmapUpdate); @@ -977,7 +1052,7 @@ write_relmap_file(bool shared, RelMapFile *newmap, /* Write the map to the relmap file. */ write_relmap_file_internal(mapfilename, newmap, write_wal, send_sinval, preserve_files, dbid, tsid, - dbpath); + dbpath, false); /* * Success, update permanent copy. During bootstrap, we might be working @@ -1069,15 +1144,37 @@ relmap_redo(XLogReaderState *record) * Write out the new map and send sinval, but of course don't write a * new WAL entry. There's no surrounding transaction to tell to * preserve files, either. - * - * There shouldn't be anyone else updating relmaps during WAL replay, - * but grab the lock to interlock against load_relmap_file(). */ - LWLockAcquire(RelationMappingLock, LW_EXCLUSIVE); - write_relmap_file((xlrec->dbid == InvalidOid), &newmap, - false, true, false, - xlrec->dbid, xlrec->tsid, dbpath); - LWLockRelease(RelationMappingLock); + if (!xlrec->create) + { + /* + * There shouldn't be anyone else updating relmaps during WAL + * replay, but grab the lock to interlock against + * load_relmap_file(). + */ + LWLockAcquire(RelationMappingLock, LW_EXCLUSIVE); + write_relmap_file((xlrec->dbid == InvalidOid), &newmap, + false, true, false, + xlrec->dbid, xlrec->tsid, dbpath); + LWLockRelease(RelationMappingLock); + } + else + { + char mapfilename[MAXPGPATH]; + + /* Construct the mapfilename. */ + snprintf(mapfilename, sizeof(mapfilename), "%s/%s", + dbpath, RELMAPPER_FILENAME); + + /* + * We don't need to take relmap lock because this wal is logged + * while creating a new database, so there could be no one else + * reading/writing the relmap file. + */ + write_relmap_file_internal(mapfilename, &newmap, false, false, + false, xlrec->dbid, xlrec->tsid, dbpath, + true); + } pfree(dbpath); } diff --git a/src/include/utils/relmapper.h b/src/include/utils/relmapper.h index c0d14da..4165f09 100644 --- a/src/include/utils/relmapper.h +++ b/src/include/utils/relmapper.h @@ -29,6 +29,7 @@ typedef struct xl_relmap_update Oid dbid; /* database ID, or 0 for shared map */ Oid tsid; /* database's tablespace, or pg_global */ int32 nbytes; /* size of relmap data */ + bool create; /* true if creating new relmap */ char data[FLEXIBLE_ARRAY_MEMBER]; } xl_relmap_update; @@ -39,6 +40,8 @@ extern Oid RelationMapOidToFilenode(Oid relationId, bool shared); extern Oid RelationMapFilenodeToOid(Oid relationId, bool shared); +extern Oid RelationMapOidToFilenodeForDatabase(char *dbpath, Oid relationId); + extern void RelationMapUpdateMap(Oid relationId, Oid fileNode, bool shared, bool immediate); @@ -62,7 +65,8 @@ extern void RelationMapInitializePhase3(void); extern Size EstimateRelationMapSpace(void); extern void SerializeRelationMap(Size maxSize, char *startAddress); extern void RestoreRelationMap(char *startAddress); - +extern void CopyRelationMap(Oid dbid, Oid tsid, char *srcdbpath, + char *dstdbpath); extern void relmap_redo(XLogReaderState *record); extern void relmap_desc(StringInfo buf, XLogReaderState *record); extern const char *relmap_identify(uint8 info); -- 1.8.3.1
From 7a734dc46fcdbb38f90f9d44e3932af5f123e154 Mon Sep 17 00:00:00 2001 From: Dilip Kumar <dilipkumar@localhost.localdomain> Date: Tue, 31 Aug 2021 15:15:00 +0530 Subject: [PATCH v2 3/3] WAL logged CREATE DATABASE Currently, CREATE DATABASE forces a checkpoint, then copies all the files, then forces another checkpoint. The comments in the createdb() function explain the reasons for this. The attached patch fixes this problem by making create database completely WAL logged so that we can avoid the checkpoints. This can also be useful for supporting the TDE. For example, if we need different encryption for the source and the target database then we can not re-encrypt the page data if we copy the whole directory. But with this patch, we are copying page by page so we have an opportunity to re-encrypt the page before copying that to the target database. --- contrib/bloom/blinsert.c | 2 +- src/backend/access/heap/heapam_handler.c | 2 +- src/backend/access/nbtree/nbtree.c | 2 +- src/backend/access/rmgrdesc/dbasedesc.c | 9 +- src/backend/access/transam/xlogutils.c | 12 +- src/backend/commands/dbcommands.c | 662 ++++++++++++++++++++----------- src/backend/commands/tablecmds.c | 59 +-- src/backend/storage/buffer/bufmgr.c | 13 +- src/bin/pg_rewind/parsexlog.c | 3 +- src/include/commands/dbcommands_xlog.h | 9 +- src/include/commands/tablecmds.h | 5 + src/include/storage/bufmgr.h | 3 +- 12 files changed, 495 insertions(+), 286 deletions(-) diff --git a/contrib/bloom/blinsert.c b/contrib/bloom/blinsert.c index 23661d1..d7054dc 100644 --- a/contrib/bloom/blinsert.c +++ b/contrib/bloom/blinsert.c @@ -173,7 +173,7 @@ blbuildempty(Relation index) * Write the page and log it. It might seem that an immediate sync would * be sufficient to guarantee that the file exists on disk, but recovery * itself might remove it while replaying, for example, an - * XLOG_DBASE_CREATE or XLOG_TBLSPC_CREATE record. Therefore, we need + * XLOG_DBASEDIR_CREATE or XLOG_TBLSPC_CREATE record. Therefore, we need * this even when wal_level=minimal. */ PageSetChecksumInplace(metapage, BLOOM_METAPAGE_BLKNO); diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c index 9befe01..dd4e038 100644 --- a/src/backend/access/heap/heapam_handler.c +++ b/src/backend/access/heap/heapam_handler.c @@ -601,7 +601,7 @@ heapam_relation_set_new_filenode(Relation rel, * even if the page has been logged, because the write did not go through * shared_buffers and therefore a concurrent checkpoint may have moved the * redo pointer past our xlog record. Recovery may as well remove it - * while replaying, for example, XLOG_DBASE_CREATE or XLOG_TBLSPC_CREATE + * while replaying, for example, XLOG_DBASEDIR_CREATE or XLOG_TBLSPC_CREATE * record. Therefore, logging is necessary even if wal_level=minimal. */ if (persistence == RELPERSISTENCE_UNLOGGED) diff --git a/src/backend/access/nbtree/nbtree.c b/src/backend/access/nbtree/nbtree.c index 30df244..5839dc7 100644 --- a/src/backend/access/nbtree/nbtree.c +++ b/src/backend/access/nbtree/nbtree.c @@ -159,7 +159,7 @@ btbuildempty(Relation index) * Write the page and log it. It might seem that an immediate sync would * be sufficient to guarantee that the file exists on disk, but recovery * itself might remove it while replaying, for example, an - * XLOG_DBASE_CREATE or XLOG_TBLSPC_CREATE record. Therefore, we need + * XLOG_DBASEDIR_CREATE or XLOG_TBLSPC_CREATE record. Therefore, we need * this even when wal_level=minimal. */ PageSetChecksumInplace(metapage, BTREE_METAPAGE); diff --git a/src/backend/access/rmgrdesc/dbasedesc.c b/src/backend/access/rmgrdesc/dbasedesc.c index 2660984..0ce38fa 100644 --- a/src/backend/access/rmgrdesc/dbasedesc.c +++ b/src/backend/access/rmgrdesc/dbasedesc.c @@ -24,12 +24,11 @@ dbase_desc(StringInfo buf, XLogReaderState *record) char *rec = XLogRecGetData(record); uint8 info = XLogRecGetInfo(record) & ~XLR_INFO_MASK; - if (info == XLOG_DBASE_CREATE) + if (info == XLOG_DBASEDIR_CREATE) { xl_dbase_create_rec *xlrec = (xl_dbase_create_rec *) rec; - appendStringInfo(buf, "copy dir %u/%u to %u/%u", - xlrec->src_tablespace_id, xlrec->src_db_id, + appendStringInfo(buf, "create dir %u/%u", xlrec->tablespace_id, xlrec->db_id); } else if (info == XLOG_DBASE_DROP) @@ -51,8 +50,8 @@ dbase_identify(uint8 info) switch (info & ~XLR_INFO_MASK) { - case XLOG_DBASE_CREATE: - id = "CREATE"; + case XLOG_DBASEDIR_CREATE: + id = "CREATE DIR"; break; case XLOG_DBASE_DROP: id = "DROP"; diff --git a/src/backend/access/transam/xlogutils.c b/src/backend/access/transam/xlogutils.c index 88a1bfd..a7a8b79 100644 --- a/src/backend/access/transam/xlogutils.c +++ b/src/backend/access/transam/xlogutils.c @@ -483,8 +483,8 @@ XLogReadBufferExtended(RelFileNode rnode, ForkNumber forknum, if (blkno < lastblock) { /* page exists in file */ - buffer = ReadBufferWithoutRelcache(rnode, forknum, blkno, - mode, NULL); + buffer = ReadBufferWithoutRelcache(rnode, forknum, blkno, mode, NULL, + RELPERSISTENCE_PERMANENT); } else { @@ -508,8 +508,8 @@ XLogReadBufferExtended(RelFileNode rnode, ForkNumber forknum, LockBuffer(buffer, BUFFER_LOCK_UNLOCK); ReleaseBuffer(buffer); } - buffer = ReadBufferWithoutRelcache(rnode, forknum, - P_NEW, mode, NULL); + buffer = ReadBufferWithoutRelcache(rnode, forknum, P_NEW, mode, + NULL, RELPERSISTENCE_PERMANENT); } while (BufferGetBlockNumber(buffer) < blkno); /* Handle the corner case that P_NEW returns non-consecutive pages */ @@ -518,8 +518,8 @@ XLogReadBufferExtended(RelFileNode rnode, ForkNumber forknum, if (mode == RBM_ZERO_AND_LOCK || mode == RBM_ZERO_AND_CLEANUP_LOCK) LockBuffer(buffer, BUFFER_LOCK_UNLOCK); ReleaseBuffer(buffer); - buffer = ReadBufferWithoutRelcache(rnode, forknum, blkno, - mode, NULL); + buffer = ReadBufferWithoutRelcache(rnode, forknum, blkno, mode, + NULL, RELPERSISTENCE_PERMANENT); } } diff --git a/src/backend/commands/dbcommands.c b/src/backend/commands/dbcommands.c index 029fab4..055c3d3 100644 --- a/src/backend/commands/dbcommands.c +++ b/src/backend/commands/dbcommands.c @@ -36,15 +36,20 @@ #include "catalog/indexing.h" #include "catalog/objectaccess.h" #include "catalog/pg_authid.h" +#include "catalog/pg_auth_members.h" #include "catalog/pg_database.h" #include "catalog/pg_db_role_setting.h" +#include "catalog/pg_proc.h" #include "catalog/pg_subscription.h" #include "catalog/pg_tablespace.h" +#include "catalog/storage.h" +#include "catalog/storage_xlog.h" #include "commands/comment.h" #include "commands/dbcommands.h" #include "commands/dbcommands_xlog.h" #include "commands/defrem.h" #include "commands/seclabel.h" +#include "commands/tablecmds.h" #include "commands/tablespace.h" #include "mb/pg_wchar.h" #include "miscadmin.h" @@ -62,6 +67,7 @@ #include "utils/builtins.h" #include "utils/fmgroids.h" #include "utils/pg_locale.h" +#include "utils/relmapper.h" #include "utils/snapmgr.h" #include "utils/syscache.h" @@ -77,6 +83,19 @@ typedef struct Oid dest_tsoid; /* tablespace we are trying to move to */ } movedb_failure_params; +/* + * While creating a database, we first scan the pg_class of the source database + * and identify all the relations to be copied to the target database. This + * is used for storing one relation entry and we will create a list of these + * entries for each valid relation of the source database. + */ +typedef struct CreateDBRelInfo +{ + RelFileNode rnode; /* physical relation identifier */ + char relpersistence; /* relation's persistence level */ +} CreateDBRelInfo; + + /* non-export function prototypes */ static void createdb_failure_callback(int code, Datum arg); static void movedb(const char *dbname, const char *tblspcname); @@ -91,6 +110,387 @@ static bool have_createdb_privilege(void); static void remove_dbtablespaces(Oid db_id); static bool check_db_file_conflict(Oid db_id); static int errdetail_busy_db(int notherbackends, int npreparedxacts); +static void CreateDirAndVersionFile(char *dbpath, Oid dbid, Oid tsid, + bool isRedo); +static List *GetDatabaseRelationList(Oid srctbid, Oid srcdbid, + Oid relfilenode); +void RelationCopyStorageUsingBuffer(SMgrRelation src, SMgrRelation dst, + ForkNumber forkNum, char relpersistence); +static void CopyDatabase(Oid src_dboid, Oid dboid, Oid src_tsid, Oid dst_tsid); + +/* + * CreateDirAndVersionFile - Create database directory and write out the + * PG_VERSION file in the database path. + * + * If isRedo is true, it's okay for the database directory to exist already. + * + * We can directly write PG_MAJORVERSION in the version file instead of copying + * from the source database file because these two must be the same. + */ +static void +CreateDirAndVersionFile(char *dbpath, Oid dbid, Oid tsid, bool isRedo) +{ + int fd; + int nbytes; + char versionfile[MAXPGPATH]; + + /* Create the empty db directory. */ + if (MakePGDirectory(dbpath) < 0) + { + /* Failure other than already exists or not in WAL replay? */ + if (errno != EEXIST || !isRedo) + ereport(ERROR, + (errcode_for_file_access(), + errmsg("could not create directory \"%s\": %m", dbpath))); + } + + /* Create PG_VERSION file in the database path. */ + snprintf(versionfile, sizeof(versionfile), "%s/%s", dbpath, "PG_VERSION"); + fd = OpenTransientFile(versionfile, O_RDWR | O_CREAT | O_EXCL | PG_BINARY); + + /* + * If file already exist and we are in WAL replay then just retry to open + * in write mode. + */ + if (fd < 0 && errno == EEXIST && isRedo) + fd = OpenTransientFile(versionfile, O_RDWR | PG_BINARY); + if (fd < 0) + ereport(ERROR, + (errcode_for_file_access(), + errmsg("could not create file \"%s\": %m", versionfile))); + + nbytes = strlen(PG_MAJORVERSION); + + /* If we are not in WAL replay then write the WAL. */ + if (!isRedo) + { + xl_dbase_create_rec xlrec; + XLogRecPtr lsn; + + /* now errors are fatal ... */ + START_CRIT_SECTION(); + + xlrec.db_id = dbid; + xlrec.tablespace_id = tsid; + xlrec.nbytes = nbytes; + + XLogBeginInsert(); + XLogRegisterData((char *) (&xlrec), MinSizeOfDbaseCreateRec); + XLogRegisterData((char *) PG_MAJORVERSION, nbytes); + + lsn = XLogInsert(RM_DBASE_ID, XLOG_DBASEDIR_CREATE); + + /* As always, WAL must hit the disk before the data update does. */ + XLogFlush(lsn); + } + + /* Write PG_MAJORVERSION in the PG_VERSION file. */ + pgstat_report_wait_start(WAIT_EVENT_COPY_FILE_WRITE); + errno = 0; + if ((int) write(fd, (char *) PG_MAJORVERSION, nbytes) != nbytes) + { + /* If write didn't set errno, assume problem is no disk space. */ + if (errno == 0) + errno = ENOSPC; + ereport(ERROR, + (errcode_for_file_access(), + errmsg("could not write to file \"%s\": %m", versionfile))); + } + pgstat_report_wait_end(); + + /* Close the version file. */ + CloseTransientFile(fd); + + /* Critical section done. */ + if (!isRedo) + END_CRIT_SECTION(); +} + +/* + * GetDatabaseRelationList - Get list of all valid relnode for the given dbid. + * + * Iterate over each block of the input relfilnode of the pg_class. We will + * scan each block and identify the all the visible tuple, from there we will + * get all valid relation in the source database. We will remember that + * information in a rnodelist and return it to the caller. + */ +static List * +GetDatabaseRelationList(Oid tbid, Oid dbid, Oid relfilenode) +{ + SMgrRelation rd_smgr; + RelFileNode rnode; + BlockNumber nblocks; + BlockNumber blkno; + OffsetNumber offnum; + OffsetNumber maxoff; + Buffer buf; + Page page; + List *rnodelist = NIL; + HeapTupleData tuple; + Form_pg_class classForm; + BufferAccessStrategy bstrategy; + + /* Prepare a relnode for pg_class relation. */ + rnode.spcNode = tbid; + rnode.dbNode = dbid; + rnode.relNode = relfilenode; + + /* + * We are not connected to the source database so open the pg_class + * relation at the smgr level and get the block count. + */ + rd_smgr = smgropen(rnode, InvalidBackendId); + nblocks = smgrnblocks(rd_smgr, MAIN_FORKNUM); + + /* + * We're going to read the whole pg_class so better to use bulk-read buffer + * access strategy. + */ + bstrategy = GetAccessStrategy(BAS_BULKREAD); + + /* Iterate over each block on the pg_class relation. */ + for (blkno = 0; blkno < nblocks; blkno++) + { + /* + * We are not connected to the source database so we can not directly + * use the top-level bufmgr interfaces. So directly use the lower + * level bufmgr interface which operates on the rnode. + */ + buf = ReadBufferWithoutRelcache(rnode, MAIN_FORKNUM, blkno, + RBM_NORMAL, bstrategy, + RELPERSISTENCE_PERMANENT); + + LockBuffer(buf, BUFFER_LOCK_SHARE); + page = BufferGetPage(buf); + if (PageIsNew(page) || PageIsEmpty(page)) + continue; + + maxoff = PageGetMaxOffsetNumber(page); + + /* Iterate over each tuple on the page. */ + for (offnum = FirstOffsetNumber; + offnum <= maxoff; + offnum = OffsetNumberNext(offnum)) + { + ItemId itemid; + + itemid = PageGetItemId(page, offnum); + + /* Nothing to do if slot is empty or already dead. */ + if (!ItemIdIsUsed(itemid) || ItemIdIsDead(itemid) || + ItemIdIsRedirected(itemid)) + continue; + + Assert(ItemIdIsNormal(itemid)); + ItemPointerSet(&(tuple.t_self), blkno, offnum); + + tuple.t_data = (HeapTupleHeader) PageGetItem(page, itemid); + tuple.t_len = ItemIdGetLength(itemid); + tuple.t_tableOid = RelationRelationId; + + /* + * If the tuple is visible then add its relfilenode info to the + * list. + */ + if (HeapTupleSatisfiesVisibility(&tuple, GetActiveSnapshot(), buf)) + { + Oid relfilenode = InvalidOid; + CreateDBRelInfo *relinfo; + + classForm = (Form_pg_class) GETSTRUCT(&tuple); + + /* We don't need to copy the shared objects to the target. */ + if (classForm->reltablespace == GLOBALTABLESPACE_OID) + continue; + + /* + * If the object doesn't have the storage then nothing to be + * done for that object so just ignore it. + */ + if (!RELKIND_HAS_STORAGE(classForm->relkind)) + continue; + + /* Built-in oids are mapped directly */ + if (classForm->oid < FirstGenbkiObjectId) + relfilenode = classForm->oid; + else if (OidIsValid(classForm->relfilenode)) + relfilenode = classForm->relfilenode; + else + continue; + + /* We must have a valid relfilenode oid. */ + Assert(OidIsValid(relfilenode)); + + /* Prepare a rel info element and add it to the list. */ + relinfo = (CreateDBRelInfo *) palloc(sizeof(CreateDBRelInfo)); + if (OidIsValid(classForm->reltablespace)) + relinfo->rnode.spcNode = classForm->reltablespace; + else + relinfo->rnode.spcNode = tbid; + + relinfo->rnode.dbNode = dbid; + relinfo->rnode.relNode = relfilenode; + relinfo->relpersistence = classForm->relpersistence; + + if (rnodelist == NULL) + rnodelist = list_make1(relinfo); + else + rnodelist = lappend(rnodelist, relinfo); + } + } + + /* Release the buffer lock. */ + UnlockReleaseBuffer(buf); + } + + return rnodelist; +} + +/* + * Copy a fork's data, block by block using buffers. Same as + * RelationCopyStorage but instead of using smgr this will copy using bufmgr. + */ +void +RelationCopyStorageUsingBuffer(SMgrRelation src, SMgrRelation dst, + ForkNumber forkNum, char relpersistence) +{ + Buffer srcBuf; + Buffer dstBuf; + Page srcPage; + Page dstPage; + bool use_wal; + bool copying_initfork; + BlockNumber nblocks; + BlockNumber blkno; + BufferAccessStrategy bstrategy = GetAccessStrategy(BAS_BULKREAD); + + /* Refer comments in RelationCopyStorage. */ + copying_initfork = relpersistence == RELPERSISTENCE_UNLOGGED && + forkNum == INIT_FORKNUM; + use_wal = XLogIsNeeded() && + (relpersistence == RELPERSISTENCE_PERMANENT || copying_initfork); + + /* Get the number of blocks in the source relation. */ + nblocks = smgrnblocks(src, forkNum); + + /* Iterate over each block of the source relation file. */ + for (blkno = 0; blkno < nblocks; blkno++) + { + /* If we got a cancel signal during the copy of the data, quit */ + CHECK_FOR_INTERRUPTS(); + + /* Read block from source relation. */ + srcBuf = ReadBufferWithoutRelcache(src->smgr_rnode.node, forkNum, + blkno, RBM_NORMAL, bstrategy, + relpersistence); + srcPage = BufferGetPage(srcBuf); + if (PageIsNew(srcPage) || PageIsEmpty(srcPage)) + { + ReleaseBuffer(srcBuf); + continue; + } + + /* Use P_NEW to extend the relation */ + dstBuf = ReadBufferWithoutRelcache(dst->smgr_rnode.node, MAIN_FORKNUM, + P_NEW, RBM_NORMAL, NULL, + relpersistence); + LockBuffer(dstBuf, BUFFER_LOCK_EXCLUSIVE); + + START_CRIT_SECTION(); + + /* Initialize the page and write the data. */ + dstPage = BufferGetPage(dstBuf); + PageInit(dstPage, BufferGetPageSize(dstBuf), 0); + memcpy(dstPage, srcPage, BLCKSZ); + MarkBufferDirty(dstBuf); + + /* WAL-log the copied page */ + if (use_wal) + log_newpage_buffer(dstBuf, true); + + END_CRIT_SECTION(); + + PageSetChecksumInplace(dstPage, blkno); + + UnlockReleaseBuffer(dstBuf); + ReleaseBuffer(srcBuf); + } +} + +/* + * Copy data block by block from the source database to the destination + * database. + */ +static void +CopyDatabase(Oid src_dboid, Oid dst_dboid, Oid src_tsid, Oid dst_tsid) +{ + char *srcpath; + char *dstpath; + Oid relfilenode; + List *rnodelist = NULL; + ListCell *cell; + RelFileNode srcrnode; + RelFileNode dstrnode; + CreateDBRelInfo *relinfo; + + /* Get the source database path. */ + srcpath = GetDatabasePath(src_dboid, src_tsid); + + /* Get the destination database path. */ + dstpath = GetDatabasePath(dst_dboid, dst_tsid); + + /* Create database directory and write PG_VERSION file. */ + CreateDirAndVersionFile(dstpath, dst_dboid, dst_tsid, false); + + /* Copy relmap file from source database to the destination database. */ + CopyRelationMap(dst_dboid, dst_tsid, srcpath, dstpath); + + /* Get pg_class relfilenode. */ + relfilenode = RelationMapOidToFilenodeForDatabase(srcpath, + RelationRelationId); + + /* Get list of all valid relnode from the source database. */ + rnodelist = GetDatabaseRelationList(src_tsid, src_dboid, + relfilenode); + Assert(rnodelist != NIL); + + /* + * Iterate over each relfilenode and copy the relation data block by block + * from source database to the destination database. + */ + foreach(cell, rnodelist) + { + SMgrRelation src_smgr; + SMgrRelation dst_smgr; + + relinfo = lfirst(cell); + srcrnode = relinfo->rnode; + + /* + * If the relation is from the default tablespace then we need to + * create it in the destinations db's default tablespace. Otherwise, + * we need to create in the same tablespace as it is created in the + * source database. + */ + if (srcrnode.spcNode != src_tsid) + dstrnode.spcNode = srcrnode.spcNode; + else + dstrnode.spcNode = dst_tsid; + + dstrnode.dbNode = dst_dboid; + dstrnode.relNode = srcrnode.relNode; + + /* Open the source and the destination relation at smgr level. */ + src_smgr = smgropen(srcrnode, InvalidBackendId); + dst_smgr = smgropen(dstrnode, InvalidBackendId); + + /* Copy relation storage from source to the destination. */ + RelationCopyAllFork(src_smgr, dst_smgr, relinfo->relpersistence, + RelationCopyStorageUsingBuffer); + } + + list_free_deep(rnodelist); +} /* @@ -99,8 +499,6 @@ static int errdetail_busy_db(int notherbackends, int npreparedxacts); Oid createdb(ParseState *pstate, const CreatedbStmt *stmt) { - TableScanDesc scan; - Relation rel; Oid src_dboid; Oid src_owner; int src_encoding = -1; @@ -562,140 +960,19 @@ createdb(ParseState *pstate, const CreatedbStmt *stmt) /* Post creation hook for new database */ InvokeObjectPostCreateHook(DatabaseRelationId, dboid, 0); - /* - * Force a checkpoint before starting the copy. This will force all dirty - * buffers, including those of unlogged tables, out to disk, to ensure - * source database is up-to-date on disk for the copy. - * FlushDatabaseBuffers() would suffice for that, but we also want to - * process any pending unlink requests. Otherwise, if a checkpoint - * happened while we're copying files, a file might be deleted just when - * we're about to copy it, causing the lstat() call in copydir() to fail - * with ENOENT. - */ - RequestCheckpoint(CHECKPOINT_IMMEDIATE | CHECKPOINT_FORCE | CHECKPOINT_WAIT - | CHECKPOINT_FLUSH_ALL); - - /* - * Once we start copying subdirectories, we need to be able to clean 'em - * up if we fail. Use an ENSURE block to make sure this happens. (This - * is not a 100% solution, because of the possibility of failure during - * transaction commit after we leave this routine, but it should handle - * most scenarios.) - */ fparms.src_dboid = src_dboid; fparms.dest_dboid = dboid; PG_ENSURE_ERROR_CLEANUP(createdb_failure_callback, PointerGetDatum(&fparms)); - { - /* - * Iterate through all tablespaces of the template database, and copy - * each one to the new database. - */ - rel = table_open(TableSpaceRelationId, AccessShareLock); - scan = table_beginscan_catalog(rel, 0, NULL); - while ((tuple = heap_getnext(scan, ForwardScanDirection)) != NULL) - { - Form_pg_tablespace spaceform = (Form_pg_tablespace) GETSTRUCT(tuple); - Oid srctablespace = spaceform->oid; - Oid dsttablespace; - char *srcpath; - char *dstpath; - struct stat st; - - /* No need to copy global tablespace */ - if (srctablespace == GLOBALTABLESPACE_OID) - continue; - - srcpath = GetDatabasePath(src_dboid, srctablespace); - - if (stat(srcpath, &st) < 0 || !S_ISDIR(st.st_mode) || - directory_is_empty(srcpath)) - { - /* Assume we can ignore it */ - pfree(srcpath); - continue; - } - - if (srctablespace == src_deftablespace) - dsttablespace = dst_deftablespace; - else - dsttablespace = srctablespace; - - dstpath = GetDatabasePath(dboid, dsttablespace); - - /* - * Copy this subdirectory to the new location - * - * We don't need to copy subdirectories - */ - copydir(srcpath, dstpath, false); - - /* Record the filesystem change in XLOG */ - { - xl_dbase_create_rec xlrec; - - xlrec.db_id = dboid; - xlrec.tablespace_id = dsttablespace; - xlrec.src_db_id = src_dboid; - xlrec.src_tablespace_id = srctablespace; - - XLogBeginInsert(); - XLogRegisterData((char *) &xlrec, sizeof(xl_dbase_create_rec)); - - (void) XLogInsert(RM_DBASE_ID, - XLOG_DBASE_CREATE | XLR_SPECIAL_REL_UPDATE); - } - } - table_endscan(scan); - table_close(rel, AccessShareLock); - - /* - * We force a checkpoint before committing. This effectively means - * that committed XLOG_DBASE_CREATE operations will never need to be - * replayed (at least not in ordinary crash recovery; we still have to - * make the XLOG entry for the benefit of PITR operations). This - * avoids two nasty scenarios: - * - * #1: When PITR is off, we don't XLOG the contents of newly created - * indexes; therefore the drop-and-recreate-whole-directory behavior - * of DBASE_CREATE replay would lose such indexes. - * - * #2: Since we have to recopy the source database during DBASE_CREATE - * replay, we run the risk of copying changes in it that were - * committed after the original CREATE DATABASE command but before the - * system crash that led to the replay. This is at least unexpected - * and at worst could lead to inconsistencies, eg duplicate table - * names. - * - * (Both of these were real bugs in releases 8.0 through 8.0.3.) - * - * In PITR replay, the first of these isn't an issue, and the second - * is only a risk if the CREATE DATABASE and subsequent template - * database change both occur while a base backup is being taken. - * There doesn't seem to be much we can do about that except document - * it as a limitation. - * - * Perhaps if we ever implement CREATE DATABASE in a less cheesy way, - * we can avoid this. - */ - RequestCheckpoint(CHECKPOINT_IMMEDIATE | CHECKPOINT_FORCE | CHECKPOINT_WAIT); - - /* - * Close pg_database, but keep lock till commit. - */ - table_close(pg_database_rel, NoLock); - - /* - * Force synchronous commit, thus minimizing the window between - * creation of the database files and committal of the transaction. If - * we crash before committing, we'll have a DB that's taking up disk - * space but is not in pg_database, which is not good. - */ - ForceSyncCommit(); - } + CopyDatabase(src_dboid, dboid, src_deftablespace, dst_deftablespace); PG_END_ENSURE_ERROR_CLEANUP(createdb_failure_callback, PointerGetDatum(&fparms)); + /* + * Close pg_database, but keep lock till commit. + */ + table_close(pg_database_rel, NoLock); + return dboid; } @@ -1190,43 +1467,12 @@ movedb(const char *dbname, const char *tblspcname) errdetail_busy_db(notherbackends, npreparedxacts))); /* - * Get old and new database paths + * Get new database path */ src_dbpath = GetDatabasePath(db_id, src_tblspcoid); dst_dbpath = GetDatabasePath(db_id, dst_tblspcoid); /* - * Force a checkpoint before proceeding. This will force all dirty - * buffers, including those of unlogged tables, out to disk, to ensure - * source database is up-to-date on disk for the copy. - * FlushDatabaseBuffers() would suffice for that, but we also want to - * process any pending unlink requests. Otherwise, the check for existing - * files in the target directory might fail unnecessarily, not to mention - * that the copy might fail due to source files getting deleted under it. - * On Windows, this also ensures that background procs don't hold any open - * files, which would cause rmdir() to fail. - */ - RequestCheckpoint(CHECKPOINT_IMMEDIATE | CHECKPOINT_FORCE | CHECKPOINT_WAIT - | CHECKPOINT_FLUSH_ALL); - - /* - * Now drop all buffers holding data of the target database; they should - * no longer be dirty so DropDatabaseBuffers is safe. - * - * It might seem that we could just let these buffers age out of shared - * buffers naturally, since they should not get referenced anymore. The - * problem with that is that if the user later moves the database back to - * its original tablespace, any still-surviving buffers would appear to - * contain valid data again --- but they'd be missing any changes made in - * the database while it was in the new tablespace. In any case, freeing - * buffers that should never be used again seems worth the cycles. - * - * Note: it'd be sufficient to get rid of buffers matching db_id and - * src_tblspcoid, but bufmgr.c presently provides no API for that. - */ - DropDatabaseBuffers(db_id); - - /* * Check for existence of files in the target directory, i.e., objects of * this database that are already in the target tablespace. We can't * allow the move in such a case, because we would need to change those @@ -1271,28 +1517,7 @@ movedb(const char *dbname, const char *tblspcname) PG_ENSURE_ERROR_CLEANUP(movedb_failure_callback, PointerGetDatum(&fparms)); { - /* - * Copy files from the old tablespace to the new one - */ - copydir(src_dbpath, dst_dbpath, false); - - /* - * Record the filesystem change in XLOG - */ - { - xl_dbase_create_rec xlrec; - - xlrec.db_id = db_id; - xlrec.tablespace_id = dst_tblspcoid; - xlrec.src_db_id = db_id; - xlrec.src_tablespace_id = src_tblspcoid; - - XLogBeginInsert(); - XLogRegisterData((char *) &xlrec, sizeof(xl_dbase_create_rec)); - - (void) XLogInsert(RM_DBASE_ID, - XLOG_DBASE_CREATE | XLR_SPECIAL_REL_UPDATE); - } + CopyDatabase(db_id, db_id, src_tblspcoid, dst_tblspcoid); /* * Update the database's pg_database tuple @@ -1326,22 +1551,6 @@ movedb(const char *dbname, const char *tblspcname) systable_endscan(sysscan); /* - * Force another checkpoint here. As in CREATE DATABASE, this is to - * ensure that we don't have to replay a committed XLOG_DBASE_CREATE - * operation, which would cause us to lose any unlogged operations - * done in the new DB tablespace before the next checkpoint. - */ - RequestCheckpoint(CHECKPOINT_IMMEDIATE | CHECKPOINT_FORCE | CHECKPOINT_WAIT); - - /* - * Force synchronous commit, thus minimizing the window between - * copying the database files and committal of the transaction. If we - * crash before committing, we'll leave an orphaned set of files on - * disk, which is not fatal but not good either. - */ - ForceSyncCommit(); - - /* * Close pg_database, but keep lock till commit. */ table_close(pgdbrel, NoLock); @@ -1350,6 +1559,23 @@ movedb(const char *dbname, const char *tblspcname) PointerGetDatum(&fparms)); /* + * Now drop all buffers holding data of the target database; they should + * no longer be dirty so DropDatabaseBuffers is safe. + * + * It might seem that we could just let these buffers age out of shared + * buffers naturally, since they should not get referenced anymore. The + * problem with that is that if the user later moves the database back to + * its original tablespace, any still-surviving buffers would appear to + * contain valid data again --- but they'd be missing any changes made in + * the database while it was in the new tablespace. In any case, freeing + * buffers that should never be used again seems worth the cycles. + * + * Note: it'd be sufficient to get rid of buffers matching db_id and + * src_tblspcoid, but bufmgr.c presently provides no API for that. + */ + DropDatabaseBuffers(db_id); + + /* * Commit the transaction so that the pg_database update is committed. If * we crash while removing files, the database won't be corrupt, we'll * just leave some orphaned files in the old directory. @@ -2138,42 +2364,14 @@ dbase_redo(XLogReaderState *record) /* Backup blocks are not used in dbase records */ Assert(!XLogRecHasAnyBlockRefs(record)); - if (info == XLOG_DBASE_CREATE) + if (info == XLOG_DBASEDIR_CREATE) { xl_dbase_create_rec *xlrec = (xl_dbase_create_rec *) XLogRecGetData(record); - char *src_path; - char *dst_path; - struct stat st; - - src_path = GetDatabasePath(xlrec->src_db_id, xlrec->src_tablespace_id); - dst_path = GetDatabasePath(xlrec->db_id, xlrec->tablespace_id); + char *dbpath; - /* - * Our theory for replaying a CREATE is to forcibly drop the target - * subdirectory if present, then re-copy the source data. This may be - * more work than needed, but it is simple to implement. - */ - if (stat(dst_path, &st) == 0 && S_ISDIR(st.st_mode)) - { - if (!rmtree(dst_path, true)) - /* If this failed, copydir() below is going to error. */ - ereport(WARNING, - (errmsg("some useless files may be left behind in old database directory \"%s\"", - dst_path))); - } - - /* - * Force dirty buffers out to disk, to ensure source database is - * up-to-date for the copy. - */ - FlushDatabaseBuffers(xlrec->src_db_id); - - /* - * Copy this subdirectory to the new location - * - * We don't need to copy subdirectories - */ - copydir(src_path, dst_path, false); + dbpath = GetDatabasePath(xlrec->db_id, xlrec->tablespace_id); + CreateDirAndVersionFile(dbpath, xlrec->db_id, xlrec->tablespace_id, + true); } else if (info == XLOG_DBASE_DROP) { diff --git a/src/backend/commands/tablecmds.c b/src/backend/commands/tablecmds.c index dbee6ae..c3e5aee 100644 --- a/src/backend/commands/tablecmds.c +++ b/src/backend/commands/tablecmds.c @@ -14189,21 +14189,13 @@ AlterTableMoveAll(AlterTableMoveAllStmt *stmt) return new_tablespaceoid; } -static void -index_copy_data(Relation rel, RelFileNode newrnode) +/* + * Copy source smgr all fork's data to the destination smgr. + */ +void +RelationCopyAllFork(SMgrRelation src_smgr, SMgrRelation dst_smgr, + char relpersistence, copy_relation_storage copy_storage) { - SMgrRelation dstrel; - - dstrel = smgropen(newrnode, rel->rd_backend); - - /* - * Since we copy the file directly without looking at the shared buffers, - * we'd better first flush out any pages of the source relation that are - * in shared buffers. We assume no new changes will be made while we are - * holding exclusive lock on the rel. - */ - FlushRelationBuffers(rel); - /* * Create and copy all forks of the relation, and schedule unlinking of * old physical files. @@ -14211,32 +14203,51 @@ index_copy_data(Relation rel, RelFileNode newrnode) * NOTE: any conflict in relfilenode value will be caught in * RelationCreateStorage(). */ - RelationCreateStorage(newrnode, rel->rd_rel->relpersistence); + RelationCreateStorage(dst_smgr->smgr_rnode.node, relpersistence); /* copy main fork */ - RelationCopyStorage(RelationGetSmgr(rel), dstrel, MAIN_FORKNUM, - rel->rd_rel->relpersistence); + copy_storage(src_smgr, dst_smgr, MAIN_FORKNUM, relpersistence); /* copy those extra forks that exist */ for (ForkNumber forkNum = MAIN_FORKNUM + 1; forkNum <= MAX_FORKNUM; forkNum++) { - if (smgrexists(RelationGetSmgr(rel), forkNum)) + if (smgrexists(src_smgr, forkNum)) { - smgrcreate(dstrel, forkNum, false); + smgrcreate(dst_smgr, forkNum, false); /* * WAL log creation if the relation is persistent, or this is the * init fork of an unlogged relation. */ - if (RelationIsPermanent(rel) || - (rel->rd_rel->relpersistence == RELPERSISTENCE_UNLOGGED && + if (relpersistence == RELPERSISTENCE_PERMANENT || + (relpersistence == RELPERSISTENCE_UNLOGGED && forkNum == INIT_FORKNUM)) - log_smgrcreate(&newrnode, forkNum); - RelationCopyStorage(RelationGetSmgr(rel), dstrel, forkNum, - rel->rd_rel->relpersistence); + log_smgrcreate(&dst_smgr->smgr_rnode.node, forkNum); + + /* Copy a fork's data, block by block. */ + copy_storage(src_smgr, dst_smgr, forkNum, relpersistence); } } +} + +static void +index_copy_data(Relation rel, RelFileNode newrnode) +{ + SMgrRelation dstrel; + + dstrel = smgropen(newrnode, rel->rd_backend); + + /* + * Since we copy the file directly without looking at the shared buffers, + * we'd better first flush out any pages of the source relation that are + * in shared buffers. We assume no new changes will be made while we are + * holding exclusive lock on the rel. + */ + FlushRelationBuffers(rel); + + RelationCopyAllFork(RelationGetSmgr(rel), dstrel, + rel->rd_rel->relpersistence, RelationCopyStorage); /* drop old relation, and close new one */ RelationDropStorage(rel); diff --git a/src/backend/storage/buffer/bufmgr.c b/src/backend/storage/buffer/bufmgr.c index bc1753a..8ec0448 100644 --- a/src/backend/storage/buffer/bufmgr.c +++ b/src/backend/storage/buffer/bufmgr.c @@ -770,24 +770,17 @@ ReadBufferExtended(Relation reln, ForkNumber forkNum, BlockNumber blockNum, /* * ReadBufferWithoutRelcache -- like ReadBufferExtended, but doesn't require * a relcache entry for the relation. - * - * NB: At present, this function may only be used on permanent relations, which - * is OK, because we only use it during XLOG replay. If in the future we - * want to use it on temporary or unlogged relations, we could pass additional - * parameters. */ Buffer ReadBufferWithoutRelcache(RelFileNode rnode, ForkNumber forkNum, BlockNumber blockNum, ReadBufferMode mode, - BufferAccessStrategy strategy) + BufferAccessStrategy strategy, char relpersistence) { bool hit; SMgrRelation smgr = smgropen(rnode, InvalidBackendId); - Assert(InRecovery); - - return ReadBuffer_common(smgr, RELPERSISTENCE_PERMANENT, forkNum, blockNum, + return ReadBuffer_common(smgr, relpersistence, forkNum, blockNum, mode, strategy, &hit); } @@ -797,7 +790,7 @@ ReadBufferWithoutRelcache(RelFileNode rnode, ForkNumber forkNum, * * *hit is set to true if the request was satisfied from shared buffer cache. */ -static Buffer +Buffer ReadBuffer_common(SMgrRelation smgr, char relpersistence, ForkNumber forkNum, BlockNumber blockNum, ReadBufferMode mode, BufferAccessStrategy strategy, bool *hit) diff --git a/src/bin/pg_rewind/parsexlog.c b/src/bin/pg_rewind/parsexlog.c index 59ebac7..0390b53 100644 --- a/src/bin/pg_rewind/parsexlog.c +++ b/src/bin/pg_rewind/parsexlog.c @@ -23,6 +23,7 @@ #include "fe_utils/archive.h" #include "filemap.h" #include "pg_rewind.h" +#include "utils/relmapper.h" /* * RmgrNames is an array of resource manager names, to make error messages @@ -370,7 +371,7 @@ extractPageInfo(XLogReaderState *record) /* Is this a special record type that I recognize? */ - if (rmid == RM_DBASE_ID && rminfo == XLOG_DBASE_CREATE) + if (rmid == RM_DBASE_ID && rminfo == XLOG_DBASEDIR_CREATE) { /* * New databases can be safely ignored. It won't be present in the diff --git a/src/include/commands/dbcommands_xlog.h b/src/include/commands/dbcommands_xlog.h index f5ed762..caf3676 100644 --- a/src/include/commands/dbcommands_xlog.h +++ b/src/include/commands/dbcommands_xlog.h @@ -18,18 +18,19 @@ #include "lib/stringinfo.h" /* record types */ -#define XLOG_DBASE_CREATE 0x00 +#define XLOG_DBASEDIR_CREATE 0x00 #define XLOG_DBASE_DROP 0x10 typedef struct xl_dbase_create_rec { - /* Records copying of a single subdirectory incl. contents */ Oid db_id; Oid tablespace_id; - Oid src_db_id; - Oid src_tablespace_id; + int32 nbytes; /* size of version data */ + char version[FLEXIBLE_ARRAY_MEMBER]; } xl_dbase_create_rec; +#define MinSizeOfDbaseCreateRec offsetof(xl_dbase_create_rec, version) + typedef struct xl_dbase_drop_rec { Oid db_id; diff --git a/src/include/commands/tablecmds.h b/src/include/commands/tablecmds.h index 336549c..e0e0aa5 100644 --- a/src/include/commands/tablecmds.h +++ b/src/include/commands/tablecmds.h @@ -19,10 +19,13 @@ #include "catalog/objectaddress.h" #include "nodes/parsenodes.h" #include "storage/lock.h" +#include "storage/smgr.h" #include "utils/relcache.h" struct AlterTableUtilityContext; /* avoid including tcop/utility.h here */ +typedef void (*copy_relation_storage) (SMgrRelation src, SMgrRelation dst, + ForkNumber forkNum, char relpersistence); extern ObjectAddress DefineRelation(CreateStmt *stmt, char relkind, Oid ownerId, ObjectAddress *typaddress, const char *queryString); @@ -42,6 +45,8 @@ extern void AlterTableInternal(Oid relid, List *cmds, bool recurse); extern Oid AlterTableMoveAll(AlterTableMoveAllStmt *stmt); +extern void RelationCopyAllFork(SMgrRelation src_smgr, SMgrRelation dst_smgr, + char relpersistence, copy_relation_storage copy_storage); extern ObjectAddress AlterTableNamespace(AlterObjectSchemaStmt *stmt, Oid *oldschema); diff --git a/src/include/storage/bufmgr.h b/src/include/storage/bufmgr.h index cfce23e..dcf42be 100644 --- a/src/include/storage/bufmgr.h +++ b/src/include/storage/bufmgr.h @@ -184,7 +184,8 @@ extern Buffer ReadBufferExtended(Relation reln, ForkNumber forkNum, BufferAccessStrategy strategy); extern Buffer ReadBufferWithoutRelcache(RelFileNode rnode, ForkNumber forkNum, BlockNumber blockNum, - ReadBufferMode mode, BufferAccessStrategy strategy); + ReadBufferMode mode, BufferAccessStrategy strategy, + char relpersistence); extern void ReleaseBuffer(Buffer buffer); extern void UnlockReleaseBuffer(Buffer buffer); extern void MarkBufferDirty(Buffer buffer); -- 1.8.3.1