Ardb is a BSD licensed, redis-protocol compatible persistent nosql, it support multiple storage engines as backend like Google's LevelDB, Facebook's RocksDB, OpenLDAP's LMDB, WiredTiger, the default backend is Facebook's RocksDB.
Rocksdb is the default storage engine, to compile with rocksdb, just type make
to compile server & lib & tests.
To use LMDB or LevelDB or WiredTiger as storage engine, you should set env storage_engine
first.
storage_engine=lmdb make
storage_engine=leveldb make
storage_engine=wiredtiger make
It should compile to several executables in src
directory, such as ardb-server, ardb-test etc.
- Full redis-protocol compatibility
- 2d spatial index supported. Spatial Index
- Most redis commands supported, and a few new commands
- Different storage engine supported (LevelDB/LMDB/RocksDB)
- Replication compatible with Redis 2.6/2.8
- Ardb instance work as slave of Redis 2.6/2.8+ instance
- Ardb instance work as master of Redis 2.6/2.8+ instance
- Ardb instance work as slave of Ardb instance
- Replication detail
- Auto failover support by redis-sentinel
- Lua Scripting support
- Backup data online
- Use 'save/bgsave' to backup data
- Use 'import' to import backup data
- Backup & Restore
Since ardb is a full redis-protocol compatible server, you can use most existed redis client to connect it without any problem. Here lists all redis clients. http://www.redis.io/clients
-
Known Issues:
- For Node.js, the recommand client node_redis would try to parse
redis_version:x.y.z
frominfo
command's output, Ardb users should uncommentadditional-misc-info
in ardb.conf to makesure thatredis_version:x.y.z
exists ininfo
command's output. There is an online redis GUI admin service redsmin build on node_redis, users can test ardb's redis protocol conformance by a visual way.
- For Node.js, the recommand client node_redis would try to parse
Benchmarks were all performed on a four-core Intel(R) Xeon(R) CPU [email protected], with 64 GB of DDR3 RAM, 500 GB of SCSI disk
The benchmark tool is 'redis-benchmark' from redis,50 parallel clients, 10000000 requests, 1000000 random keys each test case.
GCC Version:4.8.3
OS Version: Red Hat Enterprise Linux AS release 4 (Nahant Update 3)
Kernel Version: 2.6.32_1-10-6-0
Redis Version: 2.8.9
Ardb Version: 0.7.2(LMDB 0.9.11, LevelDB1.16.0, RocksDB3.1)
LevelDB Options: thread_pool_size=2, block_cache_size=512m, write_buffer_size=128m, compression=snappy
RocksDB Options: thread_pool_size=2, block_cache_size=512m, write_buffer_size=128m, compression=snappy
LMDB Option: thread_pool_size=2, database_max_size=10G, readahead=no
Becnhmark data(./redis-benchmark -r 10000000 -n 10000000):
LevelDB LMDB RocksDB Redis
PING_INLINE 95075.11 91945.56 92274.76 79669.85
PING_BULK 99265.43 92988.66 95721.27 90044.66
SET 62361.64 72490.03 55567.91 73692.51
GET 69045.99 93805.12 68078.16 82459.95
INCR 47572.16 59805.03 34883.32 74940.61
LPUSH 47369.57 27713.11 40584.42 105466.32
LPOP 31787.41 14711.51 10088.27 105797.72
SADD 37583.39 41779.82 24421.81 100405.64
SPOP 8538.13 17614.32 7955.45 75253.04
LPUSH(for LRANGE) 49504.95 16998.71 40719.93 87989.45
LRANGE_100 11090.16 17639.79 10302.91 45831.61
LRANGE_300 4453.15 6578.86 4127.29 17608.11
LRANGE_500 3503.85 4552.06 2185.72 12345.07
LRANGE_600 2680.82 3082.94 1650.46 7906.01
MSET (10 keys) 9675.48 7204.56 5255.44 35967.6
- Note:
- Ardb uses 2 threads in benchmark test, while redis is single threaded application. That's the reason ardb is faster than redis in some test cases.
- LevelDB & LMDB & RocksDB all use tree like structure on disk, more data is stored, the server is slower for query operations in theory.
Cross compilation is a work in progress. It has been shown to work for the LevelDB engine and the LMDB engine when compiling on x86 for 32bit ARM. The RocksDB engine has not been shown to work. To cross compile, set the make variables XC_BUILD, XC_HOST, and XC_TGT. These make variables correspond to autoconf's notion of build, host, and target. This [link] (https://www.gnu.org/software/autoconf/manual/autoconf-2.69/html_node/Specifying-Target-Triplets.html#Specifying-Target-Triplets) explains what each means. You will also need to set appropriate values for LD, CC, and CXX to point to the appropriate compilers and linkers for the platform are used. For example, to build on x86 for 32 bit ARM, the make invocation could look like:
make AR= LD= CC= CXX=<path to ARM g++> XC_HOST="x86-unknown-linux" XC_HOST="arm-unknown-linux"
If you see an error like:
#error "Missing implementation for 32-bit atomic operations"
Then you can either use -DJE_FORCE_SYNC_COMPARE_AND_SWAP_4
as a CPP flag or use libc for malloc by setting the make variable MALLOC
to libc
.
- Unsupported Redis Commands:
- DUMP
- MIGRATE
- OBJECT
- RESTORE
- CONFIG RESETSTAT
- DEBUG
- MONITOR
- BITPOS
- PUBSUB
- New Ardb Commands:
- SUnionCount/SInterCount/SDiffCount
- HMIncrby
- CompactDB/CompactAll
- BitOPCount
- Import
- KeysCount
- GeoAdd
- GeoSearch
- Cache
###Memory Usage In basho's fork of leveldb project, they give a way to estimate memory usage of a database.
write_buffer_size*2
+ max_open_files*4194304
+ (size in NewLRUCache initialization)
If you want to limit the total memory usage, you should tweak configuration options whoose names start with leveldb.
in ardb.conf.
- Join the mailing list(Subscribe via email)