curvefs: add a new distributed transaction model to improve rename performance

curvefs/conf/client.conf

@@ -85,6 +85,12 @@ fuseClient.getThreadPool=4
 # it gurantee the consistent of file after rename, otherwise you should
 # disable it for performance.
 fuseClient.enableMultiMountPointRename=true
+
+# the rename transaction models are different between version 1 and version 2
+# the v2 version greatly improves the performance of rename, especially in concurrent scenarios.
+# Node: v1 and v2 are incompatible and cannot be directly upgraded from a v1 cluster to v2.
+fuseClient.txVersion=1
+
 # splice will bring higher performance in some cases
 # but there might be a kernel issue that will cause kernel panic when enabling it
 # see https://lore.kernel.org/all/CAAmZXrsGg2xsP1CK+cbuEMumtrqdvD-NKnWzhNcvn71RV3c1yw@mail.gmail.com/

curvefs/conf/metaserver.conf

@@ -258,11 +258,17 @@ storage.rocksdb.unordered_write_buffer_size=67108864
 # for store inode which exclude its s3chunkinfo list (default: 3)
 storage.rocksdb.unordered_max_write_buffer_number=3
 # rocksdb column family's write_buffer_size
-# for store dentry and inode's s3chunkinfo list (unit: bytes, default: 128MB)
+# for store dentry and inode's s3chunkinfo list (unit: bytes, default: 64MB)
 storage.rocksdb.ordered_write_buffer_size=67108864
 # rocksdb column family's max_write_buffer_number
 # for store dentry and inode's s3chunkinfo list (default: 3)
 storage.rocksdb.ordered_max_write_buffer_number=3
+# rocksdb column family's write_buffer_size
+# for store tx lock and write (unit: bytes, default: 64MB)
+storage.rocksdb.tx_cf_write_buffer_size=67108864
+# rocksdb column family's max_write_buffer_number
+# for store tx lock and write (default: 3)
+storage.rocksdb.tx_cf_max_write_buffer_number=3
 # The target number of write history bytes to hold in memory (default: 20MB)
 storage.rocksdb.max_write_buffer_size_to_maintain=20971520
 # rocksdb memtable prefix bloom size ratio (size=write_buffer_size*memtable_prefix_bloom_size_ratio)
@@ -286,6 +292,8 @@ storage.rocksdb.perf_sampling_ratio=0
 # we will sending its with rpc streaming instead of
 # padding its into inode (default: 25000, about 25000 * 41 (byte) = 1MB)
 storage.s3_meta_inside_inode.limit_size=25000
+# TTL(millisecond) for tx lock
+storage.tx_lock_ttl_ms=5000
 
 # recycle options
 # metaserver scan recycle period, default 1h

curvefs/proto/common.proto

@@ -80,7 +80,7 @@ message PartitionInfo {
     // partition manage inodeid range [start, end]
     required uint64 start = 5;
     required uint64 end = 6;
-    required uint64 txId = 7;
+    optional uint64 txId = 7;
     optional uint64 nextId = 8;
     // status can change from READWRITE to READONLY, but can not chanage from READONLY to READWRITE
     // READWRITE/READONLY can change to DELETING, but DELETING can not change to READWRITE/READONLY

curvefs/proto/mds.proto

@@ -244,6 +244,14 @@ message SetClientMdsAddrsOverrideResponse {
     required FSStatusCode statusCode = 1;
 }
 
+message TsoRequest {}
+
+message TsoResponse {
+    required FSStatusCode statusCode = 1;
+    optional uint64 ts = 2;  // transaction sequence number
+    optional uint64 timestamp = 3;
+}
+
 service MdsService {
     // fs interface
     rpc CreateFs(CreateFsRequest) returns (CreateFsResponse);
@@ -259,6 +267,8 @@ service MdsService {
     rpc GetLatestTxId(GetLatestTxIdRequest) returns (GetLatestTxIdResponse);
     rpc CommitTx(CommitTxRequest) returns (CommitTxResponse);
 
+    rpc Tso(TsoRequest) returns (TsoResponse);
+
     // client lease
     rpc RefreshSession(RefreshSessionRequest) returns (RefreshSessionResponse);
 

curvefs/proto/metaserver.proto

@@ -49,6 +49,15 @@ enum MetaStatusCode {
     RPC_STREAM_ERROR = 25;
     INODE_S3_META_TOO_LARGE = 26;
     STORAGE_CLOSED = 27;
+    // tx v2 related
+    TX_FAILED = 28;
+    TX_WRITE_CONFLICT = 29;
+    TX_KEY_LOCKED = 30;
+    TX_COMMITTED = 31;
+    TX_ROLLBACKED = 32;
+    TX_TIMEOUT = 33;
+    TX_INPROGRESS = 34;
+    TX_MISMATCH = 35;
 }
 
 // dentry interface
@@ -59,7 +68,7 @@ message GetDentryRequest {
     required uint32 fsId = 4;
     required uint64 parentInodeId = 5;
     required string name = 6;
-    required uint64 txId = 7;
+    optional uint64 txId = 7;
     optional uint64 appliedIndex = 8;
 }
 
@@ -74,7 +83,8 @@ message Dentry {
     required uint64 inodeId = 2;
     required uint64 parentInodeId = 3;
     required string name = 4;
-    required uint64 txId = 5;
+    // reused txId as ts in tx v2 for compatibility in metaserver
+    optional uint64 txId = 5;
     optional uint32 flag = 6;
     optional FsFileType type = 7;
     optional uint64 txSequence = 8;
@@ -88,6 +98,7 @@ message GetDentryResponse {
     required MetaStatusCode statusCode = 1;
     optional Dentry dentry = 2;
     optional uint64 appliedIndex = 3;
+    optional TxLock txLock = 4;
 }
 
 message ListDentryRequest {
@@ -96,7 +107,7 @@ message ListDentryRequest {
     required uint32 partitionId = 3;
     required uint32 fsId = 4;
     required uint64 dirInodeId = 5;
-    required uint64 txId = 6;
+    optional uint64 txId = 6;
     optional string last = 7;     // the name of last entry
     optional uint32 count = 8;    // the number of entry required
     optional bool onlyDir = 9;
@@ -107,6 +118,7 @@ message ListDentryResponse {
     required MetaStatusCode statusCode = 1;
     repeated Dentry dentrys = 2;
     optional uint64 appliedIndex = 3;
+    optional TxLock txLock = 4;
 }
 
 message CreateDentryRequest {
@@ -120,14 +132,15 @@ message CreateDentryRequest {
 message CreateDentryResponse {
     required MetaStatusCode statusCode = 1;
     optional uint64 appliedIndex = 2;
+    optional TxLock txLock = 3;
 }
 
 message DeleteDentryRequest {
     required uint32 poolId = 1;
     required uint32 copysetId = 2;
     required uint32 partitionId = 3;
     required uint32 fsId = 4;
-    required uint64 txId = 5;
+    optional uint64 txId = 5;
     required uint64 parentInodeId = 6;
     required string name = 7;
     optional FsFileType type = 8;
@@ -137,6 +150,7 @@ message DeleteDentryRequest {
 message DeleteDentryResponse {
     required MetaStatusCode statusCode = 1;
     optional uint64 appliedIndex = 2;
+    optional TxLock txLock = 3;
 }
 
 message PrepareRenameTxRequest {
@@ -160,6 +174,85 @@ message PrepareRenameTxResponse {
     optional uint64 appliedIndex = 2;
 }
 
+message TxLock {
+    required string primaryKey = 1;
+    required uint64 startTs = 2;
+    required uint64 timestamp = 3;
+    optional uint32 index = 4;
+    optional int32 ttl = 5;
+}
+
+enum TxWriteKind {
+    Commit = 1;
+    Rollback = 2;
+}
+
+message TS {
+    required uint64 ts = 1;
+}
+
+message TxWrite {
+    required uint64 startTs = 1;
+    required TxWriteKind kind = 2;
+}
+
+message PrewriteRenameTxRequest {
+    required uint32 poolId = 1;
+    required uint32 copysetId = 2;
+    required uint32 partitionId = 3;
+    repeated Dentry dentrys = 4;
+    required TxLock txLock = 5;
+}
+
+message PrewriteRenameTxResponse {
+    required MetaStatusCode statusCode = 1;
+    repeated Dentry dentrys = 2;
+    optional TxLock txLock = 3;
+    optional uint64 appliedIndex = 4;
+}
+
+message CheckTxStatusRequest {
+    required uint32 poolId = 1;
+    required uint32 copysetId = 2;
+    required uint32 partitionId = 3;
+    required string primaryKey = 4;
+    required uint64 startTs = 5;
+    required uint64 curTimestamp = 6;
+}
+
+message CheckTxStatusResponse {
+    required MetaStatusCode statusCode = 1;
+    optional uint64 appliedIndex = 2;
+}
+
+message ResolveTxLockRequest {
+    required uint32 poolId = 1;
+    required uint32 copysetId = 2;
+    required uint32 partitionId = 3;
+    required Dentry dentry = 4;
+    required uint64 startTs = 5;
+    required uint64 commitTs = 6;
+}
+
+message ResolveTxLockResponse {
+    required MetaStatusCode statusCode = 1;
+    optional uint64 appliedIndex = 2;
+}
+
+message CommitTxRequest {
+    required uint32 poolId = 1;
+    required uint32 copysetId = 2;
+    required uint32 partitionId = 3;
+    repeated Dentry dentrys = 4;
+    required uint64 startTs = 5;
+    required uint64 commitTs = 6;
+}
+
+message CommitTxResponse {
+    required MetaStatusCode statusCode = 1;
+    optional uint64 appliedIndex = 2;
+}
+
 // inode interface
 message GetInodeRequest {
     required uint32 poolId = 1;
@@ -539,6 +632,11 @@ service MetaServerService {
     rpc CreateDentry(CreateDentryRequest) returns (CreateDentryResponse);
     rpc DeleteDentry(DeleteDentryRequest) returns (DeleteDentryResponse);
     rpc PrepareRenameTx(PrepareRenameTxRequest) returns (PrepareRenameTxResponse);
+    // tx v2
+    rpc PrewriteRenameTx(PrewriteRenameTxRequest) returns (PrewriteRenameTxResponse);
+    rpc CheckTxStatus(CheckTxStatusRequest) returns (CheckTxStatusResponse);
+    rpc ResolveTxLock(ResolveTxLockRequest) returns (ResolveTxLockResponse);
+    rpc CommitTx(CommitTxRequest) returns (CommitTxResponse);
 
     // inode interface
     rpc GetInode(GetInodeRequest) returns (GetInodeResponse);

curvefs/proto/topology.proto

@@ -61,7 +61,6 @@ message ClusterInfoData {
     required string clusterId = 1;
     // <fsId, partitionIndex of this fs>
     map<uint32, uint32> partitionIndexs = 2;
-
 }
 
 message PoolData {

curvefs/src/client/BUILD

@@ -99,6 +99,7 @@ cc_library(
         "//curvefs/src/common:metric_utils",
         "//curvefs/src/common:dynamic_vlog",
         "//curvefs/src/common:threading",
+        "//curvefs/src/metaserver:metaserver_storage_conv",
         "@com_google_absl//absl/memory",
         "@com_google_absl//absl/strings",
         "@com_google_absl//absl/synchronization",