Unlogged index fix v14 (#259)

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

* Address review complaints: add comment to mdopenfork

* Initialize unlogged index undex eclusive lock

Author: knizhnik

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"
@@ -46,6 +47,8 @@
 #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"
@@ -80,6 +83,39 @@ 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 -
@@ -221,7 +257,7 @@ 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

@@ -490,6 +490,13 @@ 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) &&
@@ -652,9 +659,23 @@ 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);