[sheepdog] [PATCH v2 3/3] lib: add cache version of writer and reader

[Robin Dong]

From: Robin Dong <sanbai at taobao.com>

Add new functions to repleace default writer and reader so it can cache
entire ext-node when set_vid_for_btree() need to access ext-node.

Signed-off-by: Robin Dong <sanbai at taobao.com>
---
 lib/sd_inode.c      | 109 ++++++++++++++++++++++++++++++++++++++++++++++++++++
 sheep/http/oalloc.c |   4 +-
 2 files changed, 111 insertions(+), 2 deletions(-)

diff --git a/lib/sd_inode.c b/lib/sd_inode.c
index 53035b0..a02d472 100644
--- a/lib/sd_inode.c
+++ b/lib/sd_inode.c
@@ -216,6 +216,106 @@ static void dump_btree(read_node_fn reader, struct sd_inode *inode)
 }
 
 /*
+ * This is the cache for inode and ext-node (B-tree), so we name it 'icache'.
+ * Cache of the same inode and ext-node dose not support concurrent operations
+ * so it could only be used in sd_inode_set_vid() which will be protected by
+ * distributed lock and should be released in the end of sd_inode_set_vid().
+ */
+static write_node_fn raw_writer;
+static read_node_fn raw_reader;
+
+/* no rationale */
+#define NUMBER_OF_CACHE	4
+
+struct inode_cache {
+	uint64_t oid;
+	unsigned char mem[SD_INODE_DATA_INDEX_SIZE];
+} cache_array[NUMBER_OF_CACHE];
+static int cache_idx;
+
+static void icache_init(void)
+{
+	cache_idx = 0;
+}
+
+static void icache_writeout(write_node_fn writer, int copies, int policy)
+{
+	int i;
+	for (i = 0; i < cache_idx; i++) {
+		writer(cache_array[i].oid, cache_array[i].mem,
+		       SD_INODE_DATA_INDEX_SIZE, 0, 0, copies, policy,
+		       false, false);
+	}
+}
+
+static void icache_release(write_node_fn writer, int copies, int policy)
+{
+	icache_writeout(writer, copies, policy);
+	icache_init();
+}
+
+static void icache_insert(write_node_fn writer, int copies, int policy,
+			 uint64_t oid, void *mem)
+{
+	int i;
+	for (i = 0; i < cache_idx; i++) {
+		if (oid == cache_array[i].oid) {
+			memcpy(cache_array[i].mem, mem,
+			       SD_INODE_DATA_INDEX_SIZE);
+			return;
+		}
+	}
+
+	if (cache_idx == (NUMBER_OF_CACHE - 1)) {
+		sd_info("cache for B-tree is full, so write all out");
+		icache_release(writer, copies, policy);
+	}
+
+	/* insert new cache */
+	cache_array[cache_idx].oid = oid;
+	memcpy(cache_array[cache_idx].mem, mem, SD_INODE_DATA_INDEX_SIZE);
+	cache_idx++;
+}
+
+static void *icache_find(uint64_t oid)
+{
+	int i;
+	for (i = 0; i < cache_idx; i++) {
+		if (cache_array[i].oid == oid)
+			return cache_array[i].mem;
+	}
+	return NULL;
+}
+
+static int icache_writer(uint64_t id, void *mem, unsigned int len,
+			 uint64_t offset, uint32_t flags, int copies,
+			 int copy_policy, bool create, bool direct)
+{
+	/* Only try to cache entire ext-node */
+	if (!offset && !create && !direct && len == SD_INODE_DATA_INDEX_SIZE) {
+		icache_insert(raw_writer, copies, copy_policy, id, mem);
+		return SD_RES_SUCCESS;
+	}
+	return raw_writer(id, mem, len, offset, flags, copies, copy_policy,
+			  create, direct);
+}
+
+static int icache_reader(uint64_t id, void **mem, unsigned int len,
+			 uint64_t offset)
+{
+	void *data;
+
+	if (!offset && len == SD_INODE_DATA_INDEX_SIZE) {
+		data = icache_find(id);
+		if (data) {
+			memcpy(*mem, data, len);
+			return SD_RES_SUCCESS;
+		}
+	}
+	return raw_reader(id, mem, len, offset);
+}
+
+/*
  * Search for the key in a B-tree node. If can't find it, return the position
  * for insert operation. So we can't just use xbsearch().
  */
@@ -626,6 +726,13 @@ void sd_inode_set_vid(write_node_fn writer, read_node_fn reader,
 	struct sd_extent_header *header;
 	int idx;
 
+	/* save default writer and reader */
+	raw_writer = writer;
+	raw_reader = reader;
+	/* use cache version of writer and reader */
+	writer = icache_writer;
+	reader = icache_reader;
+
 	for (idx = idx_start; idx <= idx_end; idx++) {
 		if (inode->store_policy == 0)
 			inode->data_vdi_id[idx] = vdi_id;
@@ -641,6 +748,8 @@ void sd_inode_set_vid(write_node_fn writer, read_node_fn reader,
 	}
 	if (inode->store_policy != 0)
 		dump_btree(reader, inode);
+
+	icache_release(raw_writer, inode->nr_copies, inode->copy_policy);
 }
 
 /*
diff --git a/sheep/http/oalloc.c b/sheep/http/oalloc.c
index b3d9583..3f3fb8b 100644
--- a/sheep/http/oalloc.c
+++ b/sheep/http/oalloc.c
@@ -178,8 +178,8 @@ int oalloc_new_finish(uint32_t vid, uint64_t start, uint64_t count)
 		goto out;
 	}
 
-	for (uint64_t i = 0; i < count; i++)
-		INODE_SET_VID(inode, start + i, vid);
+	sd_info("start %"PRIu64" end %"PRIu64, start, start + count - 1);
+	INODE_SET_VID_RANGE(inode, start, (start + count - 1), vid);
 
 	ret = sd_inode_write(sheep_bnode_writer, inode, 0, false, false);
 	if (ret != SD_RES_SUCCESS) {
-- 
1.7.12.4