table database with replication
The implementation is basically a port from jraft to NPL
The core algorithm is implemented based on the TLA+ spec, whose safety is proven, and liveness is highly depended on the pseudo random number sequence, which could be fine if different servers in the same cluster are generating random numbers with different seeds.
- Core Algorithm(Leader election,Log replication), safety is proven
- Configuration Change Support, add or remove servers one by one without limitation
- Client Request Support
- Urgent commit, see below
- log compaction
Urgent Commit, is a new feature introduced by this implementation, which enables the leader asks all other servers to commit one or more logs if commit index is advanced. With Urgent Commit, the system's performance is highly improved and the heartbeat interval could be increased to seconds,depends on how long your application can abide when a leader goes down, usually, one or two seconds is fine.
it's always safer to implement such kind of algorithm based on Math description other than natural languge description. there should be an auto conversion from TLA+ to programming languages, even they are talking things in different ways, but they are identical
Now the implementation is Single Thread model.
The Core Raft algorithm logic is in RaftServer, whose implementation is straight forward.
NOTE: logic will be modified in Phase 2, to adapt to the WAL Raft log.
The client interface keeps unchanged, we provide a script/TableDB/RaftSqliteStore.lua
, which is a StorageProvider
. And we also changed the TableDatabase internal to adapt to the new Raft API,
RaftSqliteStore will send the Log Entry above to the raft cluster in each interface.
To use TableDB Raft, u need to add a tabledb.config.xml
file in the Tabledb rootFolder
.
<tabledb>
<providers>
<provider type="TableDB.RaftSqliteStore" name="raft" file="(g1)npl_mod/TableDB/RaftSqliteStore.lua">./,localhost,9004,4,rtdb
</provider>
</providers>
<tables>
<table provider="raft" name="User1"/>
<table provider="raft" name="User2"/>
<table provider="raft" name="default" />
</tables>
</tabledb>
Above is a example of the Raft
provider config.
provider
have 4 config,type
is used incommonlib.gettable(provider.attr.type)
.file
is used inNPL.load(provider.attr.file)
.name
is the name of the provider, which is used in thetable
config.values
is the init args used in providertype
's init method.
table
have 2 config,provider
is the name of the provider, correspond to thename
in provider,name
is the table name. If the name isdefault
, the corresponding provider will be set to be default provider of the database. Ifdefault
is not set, default provider will besqlite
.
Like the original interfaces, callbacks is also implemented in a async way. But we make the connect(which is called in RaftSqliteStore:init
) to be sync to get the Cluster's leader, this will boost the performance and alleviate the retry in subsequent request.
We use sqlite's Online Backup API to make snapshot.
cd setup && setup.bat
will start 3 Raft nodes. The 3 nodes will automatically elect a leader. we can stop one node(whether it is leader or not) and the cluster can still function correctly. Raft can tolerate N/2 + 1
nodes failue, where N is the total nodes in cluster.
Create setup/client/temp/test_raft_database
directory, place the tabledb.config.xml
above to the directory, and run setup.bat client appendEntries
. This will start 1 client node whose tabledb rootFolder
is temp/test_raft_database
and default provider is raft
.The client node will send commands to the cluster and automaticlly retry in a backoff way if the command not succeed. All 3 Raft nodes will recv the same commands. Whether succeed or not the client's callback will be called.
addsrv 5
will start a node whose id is 5.(Note: you need to delete the server5
folder). To add the node to the cluster, execute setup.bat client addServer 5
. The cluster will automatically sync the logs previously commited to this new added server. Now if you start a client to send commands to the cluster, the new server will also recv these commands.
Note: every time you start a new client(whether Send Commands, Add server or Remove server), you should stop the previous client.
setup.bat client removeServer 5
will remove the node 5 in the cluster. But you can not request to remove the leader.
The Raft cluster is fault tolerate and highly available. You can stop the cluster and client with stopNPL.bat
.
NPL.load("(gl)script/ide/UnitTest/luaunit.lua");
NPL.load("(gl)npl_mod/Raft/test/TestClusterConfiguration.lua");
NPL.load("(gl)npl_mod/Raft/test/TestSnapshotSyncRequest.lua");
NPL.load("(gl)npl_mod/Raft/test/TestFileBasedSequentialLogStore.lua");
NPL.load("(gl)npl_mod/Raft/test/TestServerStateManager.lua");
NPL.load("(gl)npl_mod/TableDB/test/TestRaftLogEntryValue.lua");
LuaUnit:run('TestRaftLogEntryValue');
LuaUnit:run('TestFileBasedSequentialLogStore');
LuaUnit:run('TestClusterConfiguration');
LuaUnit:run('TestSnapshotSyncRequest');
LuaUnit:run('TestServerStateManager');
ParaGlobal.Exit(0)
Welcome for more feedbacks.:)