Unlogged index fix v15 (#262)

* Avoid errors when accessing indexes of unlogge tables after compute restart

* Support unlogged sequences

* Extract sequence start value from pg_sequence

* Initialize unlogged index undex eclusive lock

Author: knizhnik

src/backend/commands/sequence.c

@@ -98,7 +98,7 @@ static HTAB *seqhashtab = NULL; /* hash table for SeqTable items */
 static SeqTableData *last_used_seq = NULL;
 
 static void fill_seq_with_data(Relation rel, HeapTuple tuple);
-static void fill_seq_fork_with_data(Relation rel, HeapTuple tuple, ForkNumber forkNum);
+static void fill_seq_fork_with_data(Relation rel, HeapTuple tuple, ForkNumber forkNum, Buffer buf);
 static Relation lock_and_open_sequence(SeqTable seq);
 static void create_seq_hashtable(void);
 static void init_sequence(Oid relid, SeqTable *p_elm, Relation *p_rel);
@@ -351,7 +351,7 @@ ResetSequence(Oid seq_relid)
 static void
 fill_seq_with_data(Relation rel, HeapTuple tuple)
 {
-	fill_seq_fork_with_data(rel, tuple, MAIN_FORKNUM);
+	fill_seq_fork_with_data(rel, tuple, MAIN_FORKNUM, InvalidBuffer);
 
 	if (rel->rd_rel->relpersistence == RELPERSISTENCE_UNLOGGED)
 	{
@@ -360,7 +360,7 @@ fill_seq_with_data(Relation rel, HeapTuple tuple)
 		srel = smgropen(rel->rd_node, InvalidBackendId, rel->rd_rel->relpersistence);
 		smgrcreate(srel, INIT_FORKNUM, false);
 		log_smgrcreate(&rel->rd_node, INIT_FORKNUM);
-		fill_seq_fork_with_data(rel, tuple, INIT_FORKNUM);
+		fill_seq_fork_with_data(rel, tuple, INIT_FORKNUM, InvalidBuffer);
 		FlushRelationBuffers(rel);
 		smgrclose(srel);
 	}
@@ -370,28 +370,28 @@ fill_seq_with_data(Relation rel, HeapTuple tuple)
  * Initialize a sequence's relation fork with the specified tuple as content
  */
 static void
-fill_seq_fork_with_data(Relation rel, HeapTuple tuple, ForkNumber forkNum)
+fill_seq_fork_with_data(Relation rel, HeapTuple tuple, ForkNumber forkNum, Buffer buf)
 {
-	Buffer		buf;
 	Page		page;
 	sequence_magic *sm;
 	OffsetNumber offnum;
+	bool lockBuffer = false;
 
 	/* Initialize first page of relation with special magic number */
-
-	buf = ReadBufferExtended(rel, forkNum, P_NEW, RBM_NORMAL, NULL);
-	Assert(BufferGetBlockNumber(buf) == 0);
-
+	if (buf == InvalidBuffer)
+	{
+		buf = ReadBufferExtended(rel, forkNum, P_NEW, RBM_NORMAL, NULL);
+		Assert(BufferGetBlockNumber(buf) == 0);
+		LockBuffer(buf, BUFFER_LOCK_EXCLUSIVE);
+		lockBuffer = true;
+	}
 	page = BufferGetPage(buf);
-
 	PageInit(page, BufferGetPageSize(buf), sizeof(sequence_magic));
 	sm = (sequence_magic *) PageGetSpecialPointer(page);
 	sm->magic = SEQ_MAGIC;
 
 	/* Now insert sequence tuple */
 
-	LockBuffer(buf, BUFFER_LOCK_EXCLUSIVE);
-
 	/*
 	 * Since VACUUM does not process sequences, we have to force the tuple to
 	 * have xmin = FrozenTransactionId now.  Otherwise it would become
@@ -440,7 +440,8 @@ fill_seq_fork_with_data(Relation rel, HeapTuple tuple, ForkNumber forkNum)
 
 	END_CRIT_SECTION();
 
-	UnlockReleaseBuffer(buf);
+	if (lockBuffer)
+		UnlockReleaseBuffer(buf);
 }
 
 /*
@@ -1215,9 +1216,29 @@ read_seq_tuple(Relation rel, Buffer *buf, HeapTuple seqdatatuple)
 	sm = (sequence_magic *) PageGetSpecialPointer(page);
 
 	if (sm->magic != SEQ_MAGIC)
-		elog(ERROR, "bad magic number in sequence \"%s\": %08X",
-			 RelationGetRelationName(rel), sm->magic);
-
+	{
+		/* NEON: reinitialize unlogged sequence */
+		if (rel->rd_rel->relpersistence == RELPERSISTENCE_UNLOGGED)
+		{
+			Datum		value[SEQ_COL_LASTCOL] = {0};
+			bool		null[SEQ_COL_LASTCOL] = {false};
+			HeapTuple tuple;
+			Form_pg_sequence pgsform;
+
+			tuple = SearchSysCache1(SEQRELID, RelationGetRelid(rel));
+			if (!HeapTupleIsValid(tuple))
+				elog(ERROR, "cache lookup failed for sequence %u", RelationGetRelid(rel));
+			pgsform = (Form_pg_sequence) GETSTRUCT(tuple);
+			value[SEQ_COL_LASTVAL-1] = Int64GetDatumFast(pgsform->seqstart);
+			ReleaseSysCache(tuple);
+
+			tuple = heap_form_tuple(RelationGetDescr(rel), value, null);
+			fill_seq_fork_with_data(rel, tuple, MAIN_FORKNUM, *buf);
+		}
+		else
+			elog(ERROR, "bad magic number in sequence \"%s\": %08X",
+				 RelationGetRelationName(rel), sm->magic);
+	}
 	lp = PageGetItemId(page, FirstOffsetNumber);
 	Assert(ItemIdIsNormal(lp));
 

src/backend/optimizer/util/plancat.c

@@ -27,6 +27,7 @@
 #include "access/xlog.h"
 #include "catalog/catalog.h"
 #include "catalog/heap.h"
+#include "catalog/index.h"
 #include "catalog/pg_am.h"
 #include "catalog/pg_proc.h"
 #include "catalog/pg_statistic_ext.h"
@@ -47,6 +48,8 @@
 #include "rewrite/rewriteManip.h"
 #include "statistics/statistics.h"
 #include "storage/bufmgr.h"
+#include "storage/buf_internals.h"
+#include "storage/lmgr.h"
 #include "utils/builtins.h"
 #include "utils/lsyscache.h"
 #include "utils/partcache.h"
@@ -81,6 +84,39 @@ static void set_baserel_partition_key_exprs(Relation relation,
 static void set_baserel_partition_constraint(Relation relation,
 											 RelOptInfo *rel);
 
+static bool
+is_index_valid(Relation index, LOCKMODE lmode)
+{
+	if (!index->rd_index->indisvalid)
+		return false;
+
+	if (index->rd_rel->relpersistence == RELPERSISTENCE_UNLOGGED)
+	{
+		while (true)
+		{
+			Buffer metapage = ReadBuffer(index, 0);
+			bool isNew = PageIsNew(BufferGetPage(metapage));
+			ReleaseBuffer(metapage);
+			if (isNew)
+			{
+				Relation heap;
+				if (lmode != ExclusiveLock)
+				{
+					UnlockRelation(index, lmode);
+					LockRelation(index, ExclusiveLock);
+					lmode = ExclusiveLock;
+					continue;
+				}
+				DropRelFileNodesAllBuffers(&index->rd_smgr, 1);
+				heap = RelationIdGetRelation(index->rd_index->indrelid);
+				index->rd_indam->ambuild(heap, index, BuildIndexInfo(index));
+				RelationClose(heap);
+			}
+			break;
+		}
+	}
+	return true;
+}
 
 /*
  * get_relation_info -
@@ -224,7 +260,7 @@ get_relation_info(PlannerInfo *root, Oid relationObjectId, bool inhparent,
 			 * still needs to insert into "invalid" indexes, if they're marked
 			 * indisready.
 			 */
-			if (!index->indisvalid)
+			if (!is_index_valid(indexRelation, lmode))
 			{
 				index_close(indexRelation, NoLock);
 				continue;

src/backend/storage/smgr/md.c

@@ -525,6 +525,13 @@ mdopenfork(SMgrRelation reln, ForkNumber forknum, int behavior)
 
 	fd = PathNameOpenFile(path, O_RDWR | PG_BINARY);
 
+	/*
+	 * NEON: unlogged relation files are lost after compute restart - we need to implicitly recreate them
+	 * to allow data insertion
+	 */
+	if (fd < 0 && (behavior & EXTENSION_CREATE))
+		fd = PathNameOpenFile(path, O_RDWR | O_CREAT | PG_BINARY);
+
 	if (fd < 0)
 	{
 		if ((behavior & EXTENSION_RETURN_NULL) &&
@@ -689,9 +696,23 @@ mdread(SMgrRelation reln, ForkNumber forknum, BlockNumber blocknum,
 										reln->smgr_rnode.node.relNode,
 										reln->smgr_rnode.backend);
 
+	/* NEON: md smgr is used in Neon for unlogged and temp relations.
+	 * After compute node restart their data is deleted but unlogged tables are still present in system catalog.
+	 * This is a difference with Vanilla Postgres where unlogged relations are truncated only after abnormal termination.
+	 * To avoid "could not open file" we have to use EXTENSION_RETURN_NULL hear instead of EXTENSION_FAIL
+	 */
 	v = _mdfd_getseg(reln, forknum, blocknum, false,
-					 EXTENSION_FAIL | EXTENSION_CREATE_RECOVERY);
-
+					 RelFileNodeBackendIsTemp(reln->smgr_rnode)
+					 ? EXTENSION_FAIL | EXTENSION_CREATE_RECOVERY
+					 : EXTENSION_RETURN_NULL);
+	if (v == NULL)
+	{
+		char* path = relpath(reln->smgr_rnode, forknum);
+		(void)PathNameOpenFile(path, O_RDWR | O_CREAT | PG_BINARY);
+		pfree(path);
+		MemSet(buffer, 0, BLCKSZ);
+		return;
+	}
 	seekpos = (off_t) BLCKSZ * (blocknum % ((BlockNumber) RELSEG_SIZE));
 
 	Assert(seekpos < (off_t) BLCKSZ * RELSEG_SIZE);