Multithreaded Key-value store based on a Trie data structure. Trie is a kind of trees in which each node is a prefix for a key, the node position define the keys and the associated values are set on the last node of each key. They provide a big-O runtime complexity of O(m) on worst case, for insertion and lookup, where m is the length of the key. The main advantage is the possibility to query the tree by prefix, executing range scans in an easy way, while also maintaining the keyspace ordered.
Almost all commands supported has a "prefix" version which apply the command itself on a prefix instead of a full key.
$ cmake .
$ makeInside bin/ directory will be placed triedb and triedb_tests executables.
TrieDB uses a custom binary protocol to communicate, the structure of a packet is pretty simple, it is formed by a 1 byte header, the size of the payload and the payload itself.
The header, as shown below, is formed by 4 MSB to define the OPCODE of the
command, 1 bit as a PREFIX flag for commands which supports PREFIX operations
(e.g. range queries) and the remaining are reserved for future uses (perhaps
distribution on cluster).
Bytes from 2 to 5 are used for storing the length, behaving much like MQTT
length algorithm, for every byte only the first 7 bits are used to store
values, the last one bit is just a flag indicating wether the size is stored
also in the next byte or not (so 127 as 1 byte, 16129 on 2 bytes and like
this).
| Bit | 7 | 6 | 5 | 4 | 3 | 2 | 1 | 0 |
|--------|---------------|---------------|
| Byte 1 | opcode | p | reserved |
|--------|-------------------------------|
| Byte 2 | |
| . | Remaning Length |
| . | |
| Byte 5 | |As written before, each TrieDB command is identified by the 7-4 bits of every header which can be summarized by the following table:
OPCODE | BIN | HEX | OPCODE |
-------|-----------|----------------|
ACK | 00000000 | 0x00 | 0 |
PUT | 00010000 | 0x10 | 1 |
GET | 00100000 | 0x20 | 2 |
DEL | 00110000 | 0x30 | 3 |
TTL | 01000000 | 0x40 | 4 |
INC | 01010000 | 0x50 | 5 |
DEC | 01100000 | 0x60 | 6 |
CNT | 01110000 | 0x70 | 7 |
USE | 10000000 | 0x80 | 8 |
KEYS | 10010000 | 0x90 | 9 |
PING | 10100000 | 0xa0 | 10 |
QUIT | 10110000 | 0xb0 | 11 |
DB | 11000000 | 0xc0 | 12 |
INFO | 11010000 | 0xd0 | 13 |
FLUSH | 11100000 | 0xe0 | 14 |
JOIN | 11110000 | 0xf0 | 15 |Header byte can be manipulated at bit level to toggle bit flags: e.g
PUT with PREFIX = 1 is 00010000 | (00010000 >> 1) -> 00011000 -> 0x24
TrieDB server module define a classic TCP server, based on I/O multiplexing but sharing I/O and work loads between thread pools. The main thread have the exclusive responsibility of accepting connections and pass them to IO threads. From now on read and write operations for the connection will be handled by a dedicated thread pool, which after every read will decode the bytearray according to the protocol definition of each packet and finally pass the resulting packet to the worker thread pool, where, according to the OPCODE of the packet, the operation will be executed and the result will be returned back to the IO thread that will write back to the client the response packed into a bytestream.
MAIN 1...N 1...N
[EPOLL] [IO EPOLL] [WORK EPOLL]
ACCEPT THREAD IO THREAD POOL WORKER THREAD POOL
------------- -------------- ------------------
| | |
ACCEPT | |
| --------------> | |
| READ AND DECODE |
| | ---------------> |
| | WORK
| | <--------------- |
| WRITE |
| | |
ACCEPT | |
| --------------> | |By tuning the number of IO threads and worker threads based on the number of core of the host machine, it is possible to increase the number of served concurrent requests per seconds.
The underlying Trie data strucure accessed on the worker thread, in case of multiple worker threads it' s guarded by a spinlock, and being generally fast operations it shouldn't suffer high contentions by the threads and thus being really fast. It's a poor model of concurrency for the computation part but as of now it should be more than enough.
See the CHANGELOG file.