Benchmarks: proven under real load
Requests per second for REST APIs and connected clients for WebSockets, measured on cloud servers from $5 a month to 8 vCPUs. The method and the raw output of every run are public, so anyone can check them.
From one server to a fleetEvery server, measured
Load tests on 7 Linode servers, from a $5 one with 1 vCPU and 1 GB of RAM up to 8 vCPUs, shared and dedicated CPU, with MongoDB, PostgreSQL or MySQL and Redis on the same machine. You can deploy API Maker on any cloud provider.
Each tab : get all on MongoDB, public, requests per second. Choose a server to see every database.
- No cache
- With Redis cache
- One scale for every server
- MongoDB8 vCPU
12,640req/s
43,980 req/s with Redis cache
- Get by id
- 14,224 req/s
- Both tokens
- 12,106 req/s
- Custom API
- 8,893 req/s
- Custom API, sandbox
- 7,569 req/s
- Pre hook, both tokens
- 9,279 req/s
- Stream
- 11,898 req/s
- Get all in 15 s
- 192k requests
- PostgreSQL8 vCPU
12,641req/s
43,173 req/s with Redis cache
- Get by id
- 13,571 req/s
- Both tokens
- 11,913 req/s
- Custom API
- 9,176 req/s
- Custom API, sandbox
- 7,916 req/s
- Pre hook, both tokens
- 9,524 req/s
- Stream
- 13,073 req/s
- Get all in 15 s
- 188k requests
- MySQL8 vCPU
13,748req/s
44,027 req/s with Redis cache
- Get by id
- 14,510 req/s
- Both tokens
- 12,712 req/s
- Custom API
- 9,820 req/s
- Custom API, sandbox
- 8,657 req/s
- Pre hook, both tokens
- 10,204 req/s
- Stream
- 13,095 req/s
- Get all in 15 s
- 207k requests
- 8 vCPU · 16 GB$144/mo
111MB
API Maker at rest, its sandboxes beside it
- Under load
- 176 MB at most
- CPU
- Dedicated
- Requests
- 32.43M
- Errors
- 0
Watch it measured
Why Rust: API Maker Load-Tested Live on a $5 Server
API Maker is written in Rust. Put it on the smallest server Linode sells, five dollars a month, and load it live: 1,900 requests a second from MongoDB, 9,100 from the cache, memory flat at about 51 MB, and a restart in about a tenth of a second.
The performance playlist
Why it stays small
The numbers above come from how API Maker is built, not from tuning. Memory still depends on your traffic, your data and your code: measure with your own load.
- One process, every coreOne program whose threads use every core: one cache of your settings, one set of database connections.
- Connections on demandThe pool of an instance opens on first use and closes after 10 minutes without use.
- No code per tableFive tables or five thousand: the same generic handlers serve them.
- Streams that waitStream APIs read 1,000 rows at a time and wait for slow clients.
- Bounded queuesAt most 10,000 logs wait in memory; very large answers are not cached.
- Your code outsideCustom code runs in sandbox containers, apart from the API Maker process.