hbcache_v2.patch

diff -x '*.sql' -x '*.o' -x '*.txt' -rupN postgresql-9.1alpha4/src/backend/utils/time/tqual.c postgresql-9.1alpha4_hbcache/src/backend/utils/time/tqual.c
--- postgresql-9.1alpha4/src/backend/utils/time/tqual.c	2011-03-09 08:19:24.000000000 -0600
+++ postgresql-9.1alpha4_hbcache/src/backend/utils/time/tqual.c	2011-04-01 11:16:03.095195001 -0500
@@ -122,6 +122,24 @@ SetHintBits(HeapTupleHeader tuple, Buffe
 }
 
 /*
+ * SetHintBitsCached()
+ *
+ * When hint bits have been read from a process local cache that is presumed
+ * to be as fast as the in tuple hint bit, the page is not marked as dirty to
+ * give the cache an opportunity to reduce i/o caused by writing out pages for
+ * which only the hint bits have been modified.
+ *
+ * The XLogNeedsFlush check is made when setting up the cache, so it does not 
+ * have to be done again here.
+ */
+static inline void
+SetHintBitsCached(HeapTupleHeader tuple, Buffer buffer,
+                        uint16 infomask, TransactionId xid)
+{
+	tuple->t_infomask |= infomask;
+}
+
+/*
  * HeapTupleSetHintBits --- exported version of SetHintBits()
  *
  * This must be separate because of C99's brain-dead notions about how to
@@ -887,6 +905,157 @@ HeapTupleSatisfiesDirty(HeapTupleHeader
 	return false;				/* updated by other */
 }
 
+
+#define COMMIT_CACHE_BUCKET_COUNT 4
+#define COMMIT_CACHE_BUCKET_BLOCKSZ BLCKSZ
+#define COMMIT_CACHE_ROLLUPSZ 100
+#define COMMIT_CACHE_HIT_THRESHOLD 5
+
+typedef struct
+{
+	int XidBucket;
+	int HitCount;
+	bool Clear;
+} CommitCacheBucketHeader;
+
+static CommitCacheBucketHeader CommitCacheBucketList[COMMIT_CACHE_BUCKET_COUNT] 
+	= {{-1, 0, false}, {-1, 0, false}, {-1, 0, false}, {-1, 0, false}};
+static char CommitCacheData[COMMIT_CACHE_BUCKET_COUNT][COMMIT_CACHE_BUCKET_BLOCKSZ];
+static TransactionId CommitCacheXidRollupList[COMMIT_CACHE_ROLLUPSZ];
+static int CommitCacheMissCount = 0;
+
+/* qsort comparison function */
+static int
+xid_cmp(const void *p1, const void *p2)
+{
+        TransactionId	v1 = *((const TransactionId *) p1);
+        TransactionId	v2 = *((const TransactionId *) p2);
+
+        if (v1 < v2)
+                return -1;
+        if (v1 > v2)
+                return 1;
+        return 0;
+}
+
+static void RollUpCommitCache()
+{
+	TransactionId LastId = CommitCacheXidRollupList[0];
+	int XidIdx;
+	int HitCount = 0;
+	int Bucket;
+	int BucketMinHits = COMMIT_CACHE_ROLLUPSZ + 1;
+	int BucketMinIdx;
+	int XidBucket;
+
+	qsort(CommitCacheXidRollupList, COMMIT_CACHE_ROLLUPSZ, sizeof(int), xid_cmp);
+
+	for (XidIdx = 0; XidIdx < COMMIT_CACHE_ROLLUPSZ; XidIdx++)
+	{
+		if(CommitCacheXidRollupList[XidIdx] == LastId)
+			HitCount++;
+		else
+		{
+			if (HitCount >= COMMIT_CACHE_HIT_THRESHOLD)
+			{
+CheckLastXidBlock:
+				for (Bucket = 0; Bucket < COMMIT_CACHE_BUCKET_COUNT; Bucket++)
+				{
+					if (CommitCacheBucketList[Bucket].HitCount < BucketMinHits)
+					{
+						BucketMinIdx = Bucket;
+						BucketMinHits = CommitCacheBucketList[Bucket].HitCount;
+					}
+				}
+				if (HitCount > BucketMinHits)
+				{
+					XidBucket = LastId / COMMIT_CACHE_BUCKET_BLOCKSZ;
+					if(XidBucket != CommitCacheBucketList[BucketMinIdx].XidBucket)
+					{
+						CommitCacheBucketList[BucketMinIdx].XidBucket = XidBucket;
+						CommitCacheBucketList[BucketMinIdx].HitCount = HitCount;
+						CommitCacheBucketList[BucketMinIdx].Clear = true;
+					}
+				}
+
+			}
+			HitCount = 1;
+			LastId = CommitCacheXidRollupList[XidIdx];
+		}
+	}
+
+	if(HitCount >= COMMIT_CACHE_HIT_THRESHOLD)
+	{
+		XidIdx--;
+		goto CheckLastXidBlock;
+	}
+
+	for (Bucket = 0; Bucket < COMMIT_CACHE_BUCKET_COUNT; Bucket++)
+	{
+		if(CommitCacheBucketList[BucketMinIdx].Clear)
+			memset(CommitCacheData[Bucket], 0, COMMIT_CACHE_BUCKET_BLOCKSZ);
+
+		CommitCacheBucketList[BucketMinIdx].Clear = 0;
+		CommitCacheBucketList[BucketMinIdx].HitCount = 0;
+	}	
+
+	CommitCacheMissCount = 0;
+	return;
+}
+
+static int log_count=0;
+static inline bool 
+IsXidInCommitCache(TransactionId xid)
+{
+	int XidBucket = xid / COMMIT_CACHE_BUCKET_BLOCKSZ;
+	int Bucket;
+
+	for(Bucket = 0; Bucket < COMMIT_CACHE_BUCKET_COUNT; Bucket++)
+	{
+		if(CommitCacheBucketList[Bucket].XidBucket == XidBucket)
+		{
+			int ByteOffset = xid / BITS_PER_BYTE;
+			int BitOffset = xid % BITS_PER_BYTE;
+			if (CommitCacheData[Bucket][ByteOffset] & (1 << BitOffset))
+			{
+				CommitCacheBucketList[Bucket].HitCount++;
+				return true;
+			}
+
+			break;
+
+		}
+	}
+
+	CommitCacheXidRollupList[CommitCacheMissCount++] = xid;
+	
+	if (CommitCacheMissCount == COMMIT_CACHE_ROLLUPSZ)
+		RollUpCommitCache();
+
+	return false;
+}
+
+
+static inline void
+SetXidInCommitCache(TransactionId xid)
+{
+        int XidBucket = xid / COMMIT_CACHE_BUCKET_BLOCKSZ;
+        int Bucket;
+
+        for(Bucket = 0; Bucket < COMMIT_CACHE_BUCKET_COUNT; Bucket++)
+        {
+                if(XidBucket == Bucket)
+                {
+                        int ByteOffset = xid / BITS_PER_BYTE;
+                        int BitOffset = xid % BITS_PER_BYTE;
+                        CommitCacheData[Bucket][ByteOffset] |= (1 << BitOffset);
+                        break;
+                }
+        }
+
+        return;
+}
+
 /*
  * HeapTupleSatisfiesMVCC
  *		True iff heap tuple is valid for the given MVCC snapshot.
@@ -916,9 +1085,14 @@ HeapTupleSatisfiesMVCC(HeapTupleHeader t
 	{
 		if (tuple->t_infomask & HEAP_XMIN_INVALID)
 			return false;
-
+		/* Check the commit cache we store. If the transaction is 
+		 * set there, we set the hint bits but don't dirty the page.
+		 */
+		if(IsXidInCommitCache(HeapTupleHeaderGetXmin(tuple)))
+                        SetHintBitsCached(tuple, buffer, HEAP_XMIN_COMMITTED,
+                                                HeapTupleHeaderGetXmin(tuple));
 		/* Used by pre-9.0 binary upgrades */
-		if (tuple->t_infomask & HEAP_MOVED_OFF)
+		else if (tuple->t_infomask & HEAP_MOVED_OFF)
 		{
 			TransactionId xvac = HeapTupleHeaderGetXvac(tuple);
 
@@ -985,8 +1159,21 @@ HeapTupleSatisfiesMVCC(HeapTupleHeader t
 		else if (TransactionIdIsInProgress(HeapTupleHeaderGetXmin(tuple)))
 			return false;
 		else if (TransactionIdDidCommit(HeapTupleHeaderGetXmin(tuple)))
+		{
+			/* Only cache this transaction as committed if it's safe
+			 * in the xlog so as to avoid making unnecessary repeated
+			 * repeated checks to XLogNeedsFlush()
+			 */
+			if (TransactionIdIsValid(HeapTupleHeaderGetXmin(tuple)))
+			{
+		                XLogRecPtr commitLSN = TransactionIdGetCommitLSN(HeapTupleHeaderGetXmin(tuple));
+
+				if (!XLogNeedsFlush(commitLSN))
+					SetXidInCommitCache(HeapTupleHeaderGetXmin(tuple));
+			}
 			SetHintBits(tuple, buffer, HEAP_XMIN_COMMITTED,
 						HeapTupleHeaderGetXmin(tuple));
+		}
 		else
 		{
 			/* it must have aborted or crashed */