Tweak fillfactor code as per my recent proposal. Fix nbtsort.c so that

it can handle small fillfactors for ordinary-sized index entries without
failing on large ones; fix nbtinsert.c to distinguish leaf and nonleaf
pages; change the minimum fillfactor to 10% for all index types.

7 files changed, 34 insertions, 21 deletions

diff --git a/doc/src/sgml/ref/create_index.sgml b/doc/src/sgml/ref/create_index.sgml
index d73642c436..e8626331dc 100644
--- a/doc/src/sgml/ref/create_index.sgml
+++ b/doc/src/sgml/ref/create_index.sgml
@@ -221,17 +221,17 @@ CREATE [ UNIQUE ] INDEX <replaceable class="parameter">name</replaceable> ON <re
     <listitem>
      <para>
       The fillfactor for an index is a percentage that determines how full
-      the index method will try to pack index pages.  For B-trees, pages
+      the index method will try to pack index pages.  For B-trees, leaf pages
       are filled to this percentage during initial index build, and also
       when extending the index at the right (largest key values).  If pages
       subsequently become completely full, they will be split, leading to
       gradual degradation in the index's efficiency.  B-trees use a default
-      fillfactor of 90, but any value from 70 to 100 can be selected.
+      fillfactor of 90, but any value from 10 to 100 can be selected.
       If the table is static then fillfactor 100 is best to minimize the
       index's physical size, but for heavily updated tables a smaller
       fillfactor is better to minimize the need for page splits.  The
       other index methods use fillfactor in different but roughly analogous
-      ways; the default fillfactor and allowed range varies.
+      ways; the default fillfactor varies between methods.
      </para>
     </listitem>
    </varlistentry>
diff --git a/src/backend/access/gin/ginutil.c b/src/backend/access/gin/ginutil.c
index a0a6592aeb..4382714a82 100644
--- a/src/backend/access/gin/ginutil.c
+++ b/src/backend/access/gin/ginutil.c
@@ -213,7 +213,7 @@ ginoptions(PG_FUNCTION_ARGS)
 	 * It's not clear that fillfactor is useful for GIN, but for the moment
 	 * we'll accept it anyway.  (It won't do anything...)
 	 */
-#define GIN_MIN_FILLFACTOR			50
+#define GIN_MIN_FILLFACTOR			10
 #define GIN_DEFAULT_FILLFACTOR		100
 
 	result = default_reloptions(reloptions, validate,
diff --git a/src/backend/access/nbtree/nbtinsert.c b/src/backend/access/nbtree/nbtinsert.c
index 4ea57fb5c9..886413e422 100644
--- a/src/backend/access/nbtree/nbtinsert.c
+++ b/src/backend/access/nbtree/nbtinsert.c
@@ -991,7 +991,8 @@ _bt_split(Relation rel, Buffer buf, OffsetNumber firstright,
  * inserting successively increasing keys (consider sequences, timestamps,
  * etc) we will end up with a tree whose pages are about fillfactor% full,
  * instead of the 50% full result that we'd get without this special case.
- * This is the same as nbtsort.c produces for a newly-created tree.
+ * This is the same as nbtsort.c produces for a newly-created tree.  Note
+ * that leaf and nonleaf pages use different fillfactors.
  *
  * We are passed the intended insert position of the new tuple, expressed as
  * the offsetnumber of the tuple it must go in front of.  (This could be
@@ -1025,10 +1026,14 @@ _bt_findsplitloc(Relation rel,
 	/* Passed-in newitemsz is MAXALIGNED but does not include line pointer */
 	newitemsz += sizeof(ItemIdData);
 	state.newitemsz = newitemsz;
-	state.fillfactor = RelationGetFillFactor(rel, BTREE_DEFAULT_FILLFACTOR);
 	state.is_leaf = P_ISLEAF(opaque);
 	state.is_rightmost = P_RIGHTMOST(opaque);
 	state.have_split = false;
+	if (state.is_leaf)
+		state.fillfactor = RelationGetFillFactor(rel,
+												 BTREE_DEFAULT_FILLFACTOR);
+	else
+		state.fillfactor = BTREE_NONLEAF_FILLFACTOR;
 
 	/* Total free space available on a btree page, after fixed overhead */
 	leftspace = rightspace =
diff --git a/src/backend/access/nbtree/nbtsort.c b/src/backend/access/nbtree/nbtsort.c
index 130f916ef8..85c5764b24 100644
--- a/src/backend/access/nbtree/nbtsort.c
+++ b/src/backend/access/nbtree/nbtsort.c
@@ -27,10 +27,10 @@
  * insertion would cause a split (and not only of the leaf page; the need
  * for a split would cascade right up the tree).  The steady-state load
  * factor for btrees is usually estimated at 70%.  We choose to pack leaf
- * pages to the user-controllable fill factor while upper pages are always
- * packed to 70%.  This gives us reasonable density (there aren't many upper
- * pages if the keys are reasonable-size) without incurring a lot of cascading
- * splits during early insertions.
+ * pages to the user-controllable fill factor (default 90%) while upper pages
+ * are always packed to 70%.  This gives us reasonable density (there aren't
+ * many upper pages if the keys are reasonable-size) without risking a lot of
+ * cascading splits during early insertions.
  *
  * Formerly the index pages being built were kept in shared buffers, but
  * that is of no value (since other backends have no interest in them yet)
@@ -349,7 +349,7 @@ _bt_pagestate(BTWriteState *wstate, uint32 level)
 	state->btps_level = level;
 	/* set "full" threshold based on level.  See notes at head of file. */
 	if (level > 0)
-		state->btps_full = (BLCKSZ * (100 - BTREE_MIN_FILLFACTOR) / 100);
+		state->btps_full = (BLCKSZ * (100 - BTREE_NONLEAF_FILLFACTOR) / 100);
 	else
 		state->btps_full = RelationGetTargetPageFreeSpace(wstate->index,
 												BTREE_DEFAULT_FILLFACTOR);
@@ -499,11 +499,16 @@ _bt_buildadd(BTWriteState *wstate, BTPageState *state, IndexTuple itup)
 				"Consider a function index of an MD5 hash of the value, "
 				"or use full text indexing.")));
 
-	if (pgspc < itupsz || pgspc < state->btps_full)
+	/*
+	 * Check to see if page is "full".  It's definitely full if the item
+	 * won't fit.  Otherwise, compare to the target freespace derived from
+	 * the fillfactor.  However, we must put at least two items on each
+	 * page, so disregard fillfactor if we don't have that many.
+	 */
+	if (pgspc < itupsz || (pgspc < state->btps_full && last_off > P_FIRSTKEY))
 	{
 		/*
-		 * Item won't fit on this page, or we feel the page is full enough
-		 * already.  Finish off the page and write it out.
+		 * Finish off the page and write it out.
 		 */
 		Page		opage = npage;
 		BlockNumber oblkno = nblkno;
@@ -522,8 +527,7 @@ _bt_buildadd(BTWriteState *wstate, BTPageState *state, IndexTuple itup)
 		 * rearrange the old page so that the 'last item' becomes its high key
 		 * rather than a true data item.  There had better be at least two
 		 * items on the page already, else the page would be empty of useful
-		 * data.  (Hence, we must allow pages to be packed at least 2/3rds
-		 * full; the 70% figure used above is close to minimum.)
+		 * data.
 		 */
 		Assert(last_off > P_FIRSTKEY);
 		ii = PageGetItemId(opage, last_off);
diff --git a/src/include/access/gist_private.h b/src/include/access/gist_private.h
index e3352e0442..619b36dd07 100644
--- a/src/include/access/gist_private.h
+++ b/src/include/access/gist_private.h
@@ -272,7 +272,7 @@ extern Datum gistgetmulti(PG_FUNCTION_ARGS);
 #define GiSTPageSize   \
     ( BLCKSZ - SizeOfPageHeaderData - MAXALIGN(sizeof(GISTPageOpaqueData)) ) 
 
-#define GIST_MIN_FILLFACTOR			50
+#define GIST_MIN_FILLFACTOR			10
 #define GIST_DEFAULT_FILLFACTOR		90
 
 extern Datum gistoptions(PG_FUNCTION_ARGS);
diff --git a/src/include/access/hash.h b/src/include/access/hash.h
index 514e5086bf..75be830788 100644
--- a/src/include/access/hash.h
+++ b/src/include/access/hash.h
@@ -167,7 +167,7 @@ typedef HashMetaPageData *HashMetaPage;
 	 MAXALIGN(sizeof(HashPageOpaqueData)) - \
 	 sizeof(ItemIdData))
 
-#define HASH_MIN_FILLFACTOR			50
+#define HASH_MIN_FILLFACTOR			10
 #define HASH_DEFAULT_FILLFACTOR		75
 
 /*
diff --git a/src/include/access/nbtree.h b/src/include/access/nbtree.h
index a382aff142..a6fb03a7df 100644
--- a/src/include/access/nbtree.h
+++ b/src/include/access/nbtree.h
@@ -107,11 +107,15 @@ typedef struct BTMetaPageData
 	  MAXALIGN(sizeof(BTPageOpaqueData))) / 3 - sizeof(ItemIdData))
 
 /*
- * Because of above, min fillfactor can't be less than 2/3rds; see notes in
- * nbtsort.c before you change these!
+ * The leaf-page fillfactor defaults to 90% but is user-adjustable.
+ * For pages above the leaf level, we use a fixed 70% fillfactor.
+ * The fillfactor is applied during index build and when splitting
+ * a rightmost page; when splitting non-rightmost pages we try to
+ * divide the data equally.
  */
-#define BTREE_MIN_FILLFACTOR		70
+#define BTREE_MIN_FILLFACTOR		10
 #define BTREE_DEFAULT_FILLFACTOR	90
+#define BTREE_NONLEAF_FILLFACTOR	70
 
 /*
  *	Test whether two btree entries are "the same".