DragonflyDB vs Redis

Kawin Viriyaprasopsook included in DevOps

2023-08-22 528 words 3 minutes

/en/posts/db/dragonfly/featured-image.webp

Contents

Normally, when we think of in-memory databases, Redis and Memcached often come to mind. Today, I’d like to introduce another interesting option: Dragonfly, which claims to be a multi-threaded Redis replacement. So, let’s test how Dragonfly compares to Redis 7.

Redis test compose

First, let’s create a compose file for testing Redis.

services:
  redis:
    image: redis
    ulimits:
      memlock: -1
    volumes:
      - redis-data:/data
    hostname: redis
    command:
      ["redis-server", "--loglevel warning", "--unixsocket", "/data/redis.sock"]

  redis_tcp:
    depends_on:
      - redis
    image: redislabs/memtier_benchmark
    volumes:
      - redis-data:/redis-data
      - ./tmp:/data
    command: ["-h", "redis", "--out-file=/data/redis-tcp.log"]

  redis_sock:
    depends_on:
      - redis
    image: redislabs/memtier_benchmark
    volumes:
      - redis-data:/redis-data
      - ./tmp:/data
    command: ["-S", "/redis-data/redis.sock", "--out-file=/data/redis-sock.log"]

volumes:
  redis-data:

Dragonfly test compose

Next, the compose file for Dragonfly.

Note

Dragonfly typically requires a minimum RAM limit of 256MB per logical CPU.

services:
  dragonfly:
    image: docker.dragonflydb.io/dragonflydb/dragonfly
    ulimits:
      memlock: -1
    volumes:
      - dragonfly-data:/data
    hostname: dragonfly
    command:
      ["dragonfly", "--logtostderr", "--unixsocket", "/data/dragonfly.sock"]

  dragonfly_tcp:
    depends_on:
      - dragonfly
    image: redislabs/memtier_benchmark
    volumes:
      - dragonfly-data:/dragonfly-data
      - ./tmp:/data
    command: ["-h", "dragonfly", "--out-file=/data/dragonfly-tcp.log"]

  dragonfly_sock:
    depends_on:
      - dragonfly
    image: redislabs/memtier_benchmark
    volumes:
      - dragonfly-data:/dragonfly-data
      - ./tmp:/data
    command: ["-S", "/dragonfly-data/dragonfly.sock", "--out-file=/data/dragonfly-sock.log"]

volumes:
  dragonfly-data:

TCP

Let’s start with the common scenario where everyone uses TCP/IP between containers.

Redis

docker compose -f docker-compose.redis.yml up redis_tcp

Type	Ops/sec	Hits/sec	Misses/sec	Avg. Latency	p50 Latency	p99 Latency	p99.9 Latency	KB/sec
Sets	6446.59	—	—	2.82053	2.68700	5.40700	10.36700	496.50
Gets	64395.05	0.00	64395.05	2.82098	2.68700	5.40700	10.68700	2508.47
Waits	0.00	—	—	—	—	—	—	—
Totals	70841.64	0.00	64395.05	2.82093	2.68700	5.40700	10.62300	3004.97

Dragonfly

docker compose -f docker-compose.dragonfly.yml up dragonfly_tcp

Type	Ops/sec	Hits/sec	Misses/sec	Avg. Latency	p50 Latency	p99 Latency	p99.9 Latency	KB/sec
Sets	12683.03	—	—	1.53450	0.66300	10.87900	16.12700	976.82
Gets	126690.88	0.00	126690.88	1.51768	0.65500	10.94300	16.31900	4935.16
Waits	0.00	—	—	—	—	—	—	—
Totals	139373.91	0.00	126690.88	1.51921	0.65500	10.94300	16.31900	5911.97

UNIX socket

Now let’s try testing with UNIX sockets between containers.

Redis

docker compose -f docker-compose.redis.yml up redis_sock

Type	Ops/sec	Hits/sec	Misses/sec	Avg. Latency	p50 Latency	p99 Latency	p99.9 Latency	KB/sec
Sets	17540.93	—	—	1.04733	0.99900	1.82300	4.99100	1350.96
Gets	175216.52	0.00	175216.52	1.04597	0.99900	1.82300	3.87100	6825.44
Waits	0.00	—	—	—	—	—	—	—
Totals	192757.45	0.00	175216.52	1.04610	0.99900	1.82300	4.03100	8176.40

Dragonfly

docker compose -f docker-compose.dragonfly.yml up dragonfly_sock

Type	Ops/sec	Hits/sec	Misses/sec	Avg. Latency	p50 Latency	p99 Latency	p99.9 Latency	KB/sec
Sets	23136.83	—	—	0.80438	0.27900	7.74300	14.39900	1781.94
Gets	231114.05	0.00	231114.05	0.78618	0.27900	7.39100	12.67100	9002.89
Waits	0.00	—	—	—	—	—	—	—
Totals	254250.88	0.00	231114.05	0.78783	0.27900	7.42300	12.79900	10784.83

Conclusion

With the power of multi-threading, Dragonfly performs very well (especially through UNIX sockets, where the difference from TCP/IP is noticeable). However, it’s worth noting that Dragonfly doesn’t currently support 100% of the Redis API. If you’re using advanced or special features, you’ll need to check if it can be used as a direct replacement. You can find more information in the command-reference. From my testing, if you normally use Redis with only a single instance, and don’t use Graph or Geo location features, primarily using Set, Get, and occasionally PubSub or Stream, you can replace it with Dragonfly. The code and connection methods remain the same.