ZanRedisDB

What is ZanRedisDB

ZanRedisDB is a distributed redis cluster with strong consistency.

• Compatible with redis protocol: most redis command is supported in ZanRedisDB, so you can replace Redis with ZanRedisDB and get a more powerful cluster with unlimited data.
• Namespace and set supported: You can use the namespace and set as prefix to isolate different data to make operation more easily.
• Strong consistence and persistence: Use raft and rocksdb to make sure all the data have the consistence replication and persistent on stable storage.
• Scalability: You can easily increase the cluster capacity just by adding more machines.
• Multi-DC support: multi data centers deployment can be supported with rackaware feature.

Build

Install the compress library

yum install snappy-devel (for CentOS)
apt-get install libsnappy1 libsnappy-dev (for Debian/Ubuntu)
brew install snappy (for Mac)

Build the rocksdb with jemalloc

git clone https://github.com/absolute8511/rocksdb.git
cd rocksdb
git checkout v6.4.6-patched
PORTABLE=1 USE_SSE=1 USE_PCLMUL=1 WITH_JEMALLOC_FLAG=1 JEMALLOC=1 make static_lib

Install the dependency (for old go version only, if using go1.13+, it will be done in go modules):

CGO_CFLAGS="-I/path/to/rocksdb/include" CGO_LDFLAGS="-L/path/to/rocksdb -lrocksdb -lstdc++ -lm -lsnappy -lrt -ljemalloc" go get github.com/youzan/gorocksdb

CGO_CFLAGS="-I/path/to/rocksdb/include" CGO_LDFLAGS="-L/path/to/rocksdb -lrocksdb -lstdc++ -lm -lsnappy -ljemalloc" go get github.com/youzan/gorocksdb (for MacOS)

use the dep ensure to install other dependencies or use go modules for go1.13+

Build zankv and placedriver from the source (only support go version 1.10+, gcc 4.9+ or xcode-command-line-tools on Mac):

ROCKSDB=/path/to/rocksdb make

If you want package the binary release run the scripts

./pre-dist.sh
ROCKSDB=/path/to/rocksdb ./dist.sh

Deploy

• Deploy the rsync daemon which is needed on all server node to transfer the snapshot data for raft
• Deploy etcd cluster which is needed for the meta data for the namespaces
• Deploy the placedriver which is used for data placement: placedriver -config=/path/to/config
• Deploy the zankv for data storage server zankv -config=/path/to/config
• Init a namespace using the create the namespace API in placedriver

OS-Level Tuning

• Setting vm.swappiness=0
• Setting vm.min_free_kbytes to at least 1GB (8GB on larger memory system)
• Disable NUMA zone reclaim with vm.zone_reclaim_mode=0
• Disable THP(transparent huge pages)
• Avoid the tcp delay ack by echo 4 > /proc/sys/net/ipv4/tcp_delack_min (for old OS only)

API

placedriver has several HTTP APIs to manager the namespace

• list the namespace: GET /namespaces
• list the data nodes: GET /datanodes
• list the placedriver nodes: GET /listpd
• query the namespace meta info: GET /query/namespace_name
• create the namespace (handle only by leader) : POST /cluster/namespace/create?namespace=test_p16&partition_num=16&replicator=3
• delete the namespace (handle only by leader): POST /cluster/namespace/delete?namespace=test_p16&partition=**

storage server HTTP APIs for stats:

• namespace stats : GET /stats
• namespace raft internal stats : GET /raft/stats
• optimize the data storage : POST /kv/optimize
• get the raft leader of the namespace partition: GET /cluster/leader/namespace-partition

storage server also support the redis apis for read/write :

• KV:
• Hash Set:
• List:
• Sorted Set:
• ZSet:

Client

Golang client SDK : client-sdk , a redis proxy can be deployed based on this golang sdk if you want use the redis client in other language.

Architechture

Roadmap

• Redis data structures
  • [x] KV
  • [x] Hash
  • [x] List
  • [x] Set
  • [x] Sorted Set
  • [x] GeoHash
  • [x] Expires
  • [x] HyperLogLog
  • [x] JSON
• Distributed system
  • [x] Raft based replication
  • [x] Partitions
  • [x] Auto balance and migrate
  • [x] Support namespace
  • [x] High available
  • [x] Distributed scan on table
  • [x] Rackaware deployment used for across data center deployment
• Searchable and Indexing
  • [ ] Secondary index support on Hash fields
  • [ ] Secondary index support for json kv
  • [ ] Full text search support
• Operation
  • [x] Backup and restore for cluster
  • [x] More stats for read/write performance and errors.
• Client
  • [x] High available for redis commands (Retry on fail)
  • [ ] Extand redis commands to support index and search
  • [x] Extand redis commands for advance scan
• Others (maybe)
  • [ ] Support configure for Column storage friendly for OLAP
  • [x] Pebble as the storage engine
  • [x] Support other memory storage engine
  • [ ] Support export data to other systems

Thanks

Many thanks for these great projects which make this project possible: etcd, RocksDB, ledisdb, pika.