dynamoredis — a truly serverless Redis

ElastiCache / Redis bills by the hour while idle. dynamoredis does the same job on one DynamoDB table behind a Lambda Function URL and costs ≈ $0 when nothing happens. Same client API (redis-py); when you outgrow it, go back to a real Redis by changing one import line.

中文 · compatibility report · tasks · changelog

	ElastiCache Serverless (Redis)	dynamoredis
Idle cost / month	~$90 — billed per GB-hour with a 1 GB minimum ($0.125 × 24 × 30), ECPUs on top	≈ $0
Endpoint	host:6379, RESP protocol, VPC	Lambda Function URL, HTTPS, anywhere
Storage	memory on nodes	one DynamoDB table (pay per request, TTL, multi-AZ, durable)
Client	import redis	from dynamoredis import redis
Console	RedisInsight	built-in web console (Lambda URL)
Client compatibility	—	see the report

Why

Most Redis use is a cache, a few counters, sessions, a small queue — bursts of activity and long idle stretches, for which ElastiCache Serverless still charges every hour for at least 1 GB (≈ $90 a month before a single request). DynamoDB already is a serverless key-value store with atomic counters, sets, lists and TTL; what is missing is the Redis API. dynamoredis adds it: a Lambda translates redis commands into DynamoDB calls, the driver speaks redis-py.

How it works

Your code calls the Lambda Function URL with HTTP Basic auth (the service's user / password); the Lambda runs each command against one DynamoDB table — item = key, type, value, optional expiry. Where Redis is atomic, DynamoDB is atomic: INCR is an ADD, LPUSH is list_append, SADD is a set ADD, a type mismatch is a condition failure (WRONGTYPE). Expiry is honoured on read at once; DynamoDB's TTL sweep deletes the item later. Durable the moment a command returns, multi-AZ, no node to run out of memory.

Quick start

1. Deploy (pick one)

Terraform

cp terraform/example.json terraform/my.json      # set aws_profile, env, console_auth_username, console_auth_password
terraform -chdir=terraform init
terraform -chdir=terraform apply -var-file=my.json
terraform -chdir=terraform output                 # console_url
terraform -chdir=terraform destroy -var-file=my.json   # tear everything down (the table included)

CloudFormation (AWS CLI only)

cp cloudformation/example-params.json cloudformation/params.json
aws cloudformation package --template-file cloudformation/template.yaml \
    --s3-bucket <any-bucket-you-own> --output-template-file /tmp/dynamoredis.yaml
aws cloudformation deploy --template-file /tmp/dynamoredis.yaml --stack-name dynamoredis \
    --capabilities CAPABILITY_IAM --parameter-overrides file://cloudformation/params.json

Both create one DynamoDB table (<name>-<env>, pay per request, TTL on exp), one Lambda with a public Function URL, and one scoped IAM role.

2. Open the console

Open console_url in a browser, sign in with user / password. Keys by prefix on the left (SCAN), the value on the right by type (string, hash, list, set, sorted set) with its TTL, a command line on top — SET greeting hello, HGETALL user:1, ZADD board 10 alice. Every call shows the server time and the DynamoDB capacity units it consumed.

Paste this into the command box to get a small dataset to click around (one command per line; FLUSHDB clears it):

HSET user:tom name Tom age 31 city Boston
HSET user:mary name Mary age 28 city Seattle
HSET user:alice name Alice age 30 city Austin
SET user:tom:last_login 2026-10-26T05:00:00Z
INCR user:tom:visits
SADD user:tom:roles admin editor
ZADD user:leaderboard 120 tom 95 mary 150 alice
RPUSH user:tom:inbox "hello tom" "meeting at 3"
SET config:feature:dark_mode on
SETEX session:abc123 3600 tom
INCR counter:pageviews
LPUSH jobs:pending job-1 job-2

Then look at the DynamoDB table in the AWS console: one item per key — t is the type, v the value (hash = Map, list = List, set = String Set, sorted set = Map of member → score, counters = Number), session:abc123 carries exp.

3. Connect from code

pip install dynamoredis

from dynamoredis import redis        # later: `import redis` against a real Redis, nothing else changes

r = redis.Redis(host="xxx.lambda-url.us-east-2.on.aws", username="admin", password="...")
r.set("greeting", "hello", ex=60)
r.get("greeting")                    # b"hello"   (decode_responses=True -> "hello")
r.incr("hits")                       # atomic
r.hset("user:1", mapping={"name": "alice", "age": 30}); r.hgetall("user:1")
r.rpush("queue", "job-1", "job-2"); r.lpop("queue")
r.zadd("board", {"alice": 10, "bob": 20}); r.zrange("board", 0, -1, withscores=True)
with r.pipeline(transaction=False) as p:   # one HTTP request for the batch
    p.set("a", 1); p.get("a"); p.execute()

Compatibility

tests/compat/REPORT.md is regenerated by every test run: the same redis-py calls run on a real Redis 7 (Docker) and on dynamoredis, comparing replies (type included) and errors.

Status	Meaning
✅ verified	identical to redis-py against Redis 7; safe to rely on
❌ gap	known difference; the note names the task in TASKS.md that closes it
⚪ quirk	documented behaviour of Redis we deliberately do not imitate
⛔ unsupported	refused with NotSupportedError — never silently different

Run it yourself — the same commands CI runs on every push:

以上為 README 的開頭部分。其餘內容請閱讀原文。 檢視完整 README ↗