Switch to varchar keys and arbitrary-length values in raftable.

kvap · kvap · commit 3307c9aae377 · 2016-03-05T00:15:57.000+03:00
diff --git a/contrib/raftable/raftable--1.0.sql b/contrib/raftable/raftable--1.0.sql
@@ -2,19 +2,19 @@
 \echo Use "CREATE EXTENSION raftable" to load this file. \quit
 
 -- get
-CREATE FUNCTION raftable(key int)
+CREATE FUNCTION raftable(key varchar(64))
 RETURNS text
 AS 'MODULE_PATHNAME','raftable_sql_get'
 LANGUAGE C;
 
 -- set
-CREATE FUNCTION raftable(key int, value text)
+CREATE FUNCTION raftable(key varchar(64), value text)
 RETURNS void
 AS 'MODULE_PATHNAME','raftable_sql_set'
 LANGUAGE C;
 
 -- list
 CREATE FUNCTION raftable()
-RETURNS table (key int, value text)
+RETURNS table (key varchar(64), value text)
 AS 'MODULE_PATHNAME','raftable_sql_list'
 LANGUAGE C;
diff --git a/contrib/raftable/raftable.c b/contrib/raftable/raftable.c
@@ -16,9 +16,36 @@
 
 #include "raftable.h"
 
-#define RAFTABLE_SHMEM_SIZE (16 * 1024)
+#define RAFTABLE_KEY_LEN (64)
+#define RAFTABLE_BLOCK_LEN (256)
+#define RAFTABLE_BLOCKS (4096)
+#define RAFTABLE_BLOCK_MEM (RAFTABLE_BLOCK_LEN * (RAFTABLE_BLOCKS - 1) + sizeof(RaftableBlockMem))
 #define RAFTABLE_HASH_SIZE (127)
-#define RAFTABLE_VALUE_LEN 64
+#define RAFTABLE_SHMEM_SIZE ((1024 * 1024) + RAFTABLE_BLOCK_MEM)
+
+typedef struct RaftableBlock
+{
+	struct RaftableBlock *next;
+	char data[RAFTABLE_BLOCK_LEN - sizeof(void*)];
+} RaftableBlock;
+
+typedef struct RaftableKey
+{
+	char data[RAFTABLE_KEY_LEN];
+} RaftableKey;
+
+typedef struct RaftableEntry
+{
+	RaftableKey key;
+	int len;
+	RaftableBlock *value;
+} RaftableEntry;
+
+typedef struct RaftableBlockMem
+{
+	RaftableBlock *free_blocks;
+	RaftableBlock blocks[1];
+} RaftableBlockMem;
 
 void _PG_init(void);
 void _PG_fini(void);
@@ -29,53 +56,183 @@ PG_FUNCTION_INFO_V1(raftable_sql_get);
 PG_FUNCTION_INFO_V1(raftable_sql_set);
 PG_FUNCTION_INFO_V1(raftable_sql_list);
 
-static HTAB *data;
-static LWLockId datalock;
+static HTAB *hashtable;
+static LWLockId hashlock;
+
+static RaftableBlockMem *blockmem;
+static LWLockId blocklock;
+
 static shmem_startup_hook_type PreviousShmemStartupHook;
 
-typedef struct RaftableEntry
+static RaftableBlock *block_alloc(void)
 {
-	int key;
-	char value[RAFTABLE_VALUE_LEN];
-} RaftableEntry;
+	RaftableBlock *result;
+
+	LWLockAcquire(blocklock, LW_EXCLUSIVE);
+
+	result = blockmem->free_blocks;
+	if (result == NULL)
+		elog(ERROR, "raftable memory limit hit");
+
+
+	blockmem->free_blocks = blockmem->free_blocks->next;
+	result->next = NULL;
+	LWLockRelease(blocklock);
+	return result;
+}
+
+static void block_free(RaftableBlock *block)
+{
+	RaftableBlock *new_free_head = block;
+	Assert(block != NULL);
+	while (block->next != NULL) {
+		block = block->next;
+	}
+	LWLockAcquire(blocklock, LW_EXCLUSIVE);
+	block->next = blockmem->free_blocks;
+	blockmem->free_blocks = new_free_head;
+	LWLockRelease(blocklock);
+}
+
+
+static text *entry_to_text(RaftableEntry *e)
+{
+	char *cursor, *buf;
+	RaftableBlock *block;
+	text *t;
+	int len;
+
+	buf = palloc(e->len + 1);
+	cursor = buf;
+
+	block = e->value;
+	len = e->len;
+	while (block != NULL)
+	{
+		int tocopy = len;
+		if (tocopy > sizeof(block->data))
+			tocopy = sizeof(block->data);
+
+		memcpy(cursor, block->data, tocopy);
+		cursor += tocopy;
+		len -= tocopy;
+
+		Assert(cursor - buf <= e->len);
+		block = block->next;
+	}
+	Assert(cursor - buf == e->len);
+	*cursor = '\0';
+	t = cstring_to_text_with_len(buf, e->len);
+	pfree(buf);
+	return t;
+}
+
+static void text_to_entry(RaftableEntry *e, text *t)
+{
+	char *buf, *cursor;
+	int len;
+	RaftableBlock *block;
+
+	buf = text_to_cstring(t);
+	cursor = buf;
+	len = strlen(buf);
+	e->len = len;
+
+	if (e->len > 0)
+	{
+		if (e->value == NULL)
+			e->value = block_alloc();
+		Assert(e->value != NULL);
+	}
+	else
+	{
+		if (e->value != NULL)
+		{
+			block_free(e->value);
+			e->value = NULL;
+		}
+	}
+
+	block = e->value;
+	while (len > 0)
+	{
+		int tocopy = len;
+		if (tocopy > sizeof(block->data))
+		{
+			tocopy = sizeof(block->data);
+			if (block->next == NULL)
+				block->next = block_alloc();
+		}
+		else
+		{
+			if (block->next != NULL)
+			{
+				block_free(block->next);
+				block->next = NULL;
+			}
+		}
+
+		memcpy(block->data, cursor, tocopy);
+		cursor += tocopy;
+		len -= tocopy;
+
+		block = block->next;
+	}
+
+	pfree(buf);
+	Assert(block == NULL);
+}
 
 Datum
 raftable_sql_get(PG_FUNCTION_ARGS)
 {
 	RaftableEntry *entry;
-	int key = PG_GETARG_INT32(0);
+	RaftableKey key;
+	text_to_cstring_buffer(PG_GETARG_TEXT_P(0), key.data, sizeof(key.data));
 
-	LWLockAcquire(datalock, LW_SHARED);
-	entry = hash_search(data, &key, HASH_FIND, NULL);
+	LWLockAcquire(hashlock, LW_SHARED);
+	entry = hash_search(hashtable, &key, HASH_FIND, NULL);
 
 	if (entry)
 	{
-		text *t = cstring_to_text(entry->value);
-		LWLockRelease(datalock);
+		text *t = entry_to_text(entry);
+		LWLockRelease(hashlock);
 		PG_RETURN_TEXT_P(t);
 	}
 	else
 	{
-		LWLockRelease(datalock);
+		LWLockRelease(hashlock);
 		PG_RETURN_NULL();
 	}
 }
 
 Datum
 raftable_sql_set(PG_FUNCTION_ARGS)
 {
-	int key = PG_GETARG_INT32(0);
+	RaftableKey key;
+	text_to_cstring_buffer(PG_GETARG_TEXT_P(0), key.data, sizeof(key.data));
 
-	LWLockAcquire(datalock, LW_EXCLUSIVE);
+	LWLockAcquire(hashlock, LW_EXCLUSIVE);
 	if (PG_ARGISNULL(1))
-		hash_search(data, &key, HASH_REMOVE, NULL);
+	{
+		RaftableEntry *entry = hash_search(hashtable, key.data, HASH_FIND, NULL);
+		if ((entry != NULL) && (entry->len > 0))
+			block_free(entry->value);
+		hash_search(hashtable, key.data, HASH_REMOVE, NULL);
+	}
 	else
 	{
-		RaftableEntry *entry = hash_search(data, &key, HASH_ENTER, NULL);
-		entry->key = key;
-		text_to_cstring_buffer(PG_GETARG_TEXT_P(1), entry->value, sizeof(entry->value));
+		bool found;
+		RaftableEntry *entry = hash_search(hashtable, key.data, HASH_ENTER, &found);
+		if (!found)
+		{
+			entry->key = key;
+			entry->value = NULL;
+			entry->len = 0;
+		}
+		text_to_entry(entry, PG_GETARG_TEXT_P(1));
 	}
-	LWLockRelease(datalock);
+	LWLockRelease(hashlock);
 
 	PG_RETURN_VOID();
 }
@@ -96,12 +253,15 @@ raftable_sql_list(PG_FUNCTION_ARGS)
 		oldcontext = MemoryContextSwitchTo(funcctx->multi_call_memory_ctx);
 
 		tfc = get_call_result_type(fcinfo, NULL, &funcctx->tuple_desc);
-		Assert(tfc == TYPEFUNC_COMPOSITE);
+		if (tfc != TYPEFUNC_COMPOSITE)
+		{
+			elog(ERROR, "raftable listing function should be composite");
+		}
 		funcctx->tuple_desc = BlessTupleDesc(funcctx->tuple_desc);
 
 		scan = (HASH_SEQ_STATUS *)palloc(sizeof(HASH_SEQ_STATUS));
-		LWLockAcquire(datalock, LW_SHARED);
-		hash_seq_init(scan, data);
+		LWLockAcquire(hashlock, LW_SHARED);
+		hash_seq_init(scan, hashtable);
 
 		MemoryContextSwitchTo(oldcontext);
 
@@ -117,52 +277,59 @@ raftable_sql_list(PG_FUNCTION_ARGS)
 		Datum  vals[2];
 		bool isnull[2];
 
-		vals[0] = Int32GetDatum(entry->key);
-		vals[1] = CStringGetTextDatum(entry->value);
+		vals[0] = CStringGetTextDatum(entry->key.data);
+		vals[1] = PointerGetDatum(entry_to_text(entry));
 		isnull[0] = isnull[1] = false;
 
 		tuple = heap_form_tuple(funcctx->tuple_desc, vals, isnull);
 		SRF_RETURN_NEXT(funcctx, HeapTupleGetDatum(tuple));
 	}
 	else
 	{
-		LWLockRelease(datalock);
+		LWLockRelease(hashlock);
 		SRF_RETURN_DONE(funcctx);
 	}
 
 }
 
-static uint32 raftable_hash_fn(const void *key, Size keysize)
-{
-	return (uint32)*(int*)key;
-}
-
-static int raftable_match_fn(const void *key1, const void *key2, Size keysize)
-{
-	return *(int*)key1 != *(int*)key2;
-}
-
 static void startup_shmem(void)
 {
-	HASHCTL info;
-
 	if (PreviousShmemStartupHook){
 		PreviousShmemStartupHook();
 	}
 
-	datalock = LWLockAssign();
+	{
+		HASHCTL info;
+		hashlock = LWLockAssign();
 
-	info.keysize = sizeof(int);
-	info.entrysize = sizeof(RaftableEntry);
-	info.hash = raftable_hash_fn;
-	info.match = raftable_match_fn;
+		info.keysize = sizeof(RaftableKey);
+		info.entrysize = sizeof(RaftableEntry);
 
-	data = ShmemInitHash(
-		"raftable",
-		RAFTABLE_HASH_SIZE, RAFTABLE_HASH_SIZE,
-		&info,
-		HASH_ELEM | HASH_FUNCTION | HASH_COMPARE
-	);
+		hashtable = ShmemInitHash(
+			"raftable_hashtable",
+			RAFTABLE_HASH_SIZE, RAFTABLE_HASH_SIZE,
+			&info, HASH_ELEM
+		);
+	}
+
+	{
+		bool found;
+		int i;
+
+		blocklock = LWLockAssign();
+
+		blockmem = ShmemInitStruct(
+			"raftable_blockmem",
+			RAFTABLE_BLOCK_MEM,
+			&found
+		);
+
+		for (i = 0; i < RAFTABLE_BLOCKS - 1; i++) {
+			blockmem->blocks[i].next = blockmem->blocks + i + 1;
+		}
+		blockmem->blocks[RAFTABLE_BLOCKS - 1].next = NULL;
+		blockmem->free_blocks = blockmem->blocks;
+	}
 }
 
 void
@@ -171,7 +338,7 @@ _PG_init(void)
 	if (!process_shared_preload_libraries_in_progress)
 		elog(ERROR, "please add 'raftable' to shared_preload_libraries list");
 	RequestAddinShmemSpace(RAFTABLE_SHMEM_SIZE);
-	RequestAddinLWLocks(1);
+	RequestAddinLWLocks(2);
 
 	PreviousShmemStartupHook = shmem_startup_hook;
 	shmem_startup_hook = startup_shmem;
