Benchmark Supabase (fly.io)

[kiwicopple]

Chore

• Running Database on Supabase
• Attached PostgREST directly to the Supabase database
• Running benchmarks directly to the PostgREST (no kong)

Results

Instance	vCPU	RAM	Price	Read	Write	Comments
AWS t3a.micro	2	1	$6.80	303/s	307/s
Fly micro-1x	shared	128MB	$2.67	171/s	235/s	Reached request duration threshold
Fly micro-2x	shared	512MB	$8	206/s	277/s

Fly config

app = "postgrest"

[build]
  image = "postgrest/postgrest"

[[services]]
  internal_port = 3000
  protocol = "tcp"

  [services.concurrency]
    hard_limit = 25
    soft_limit = 20

  [[services.ports]]
    handlers = ["http"]
    port = "80"

  [[services.ports]]
    handlers = ["tls", "http"]
    port = "443"

  [[services.tcp_checks]]
    interval = 10000
    timeout = 2000

Read benchmarks

Write benchmarks

[kiwicopple]

Just to make sure I am running with the same amount of data in the benchmark database, and on the same hardware, here is my run for Supabase:

Read

Write

[kiwicopple]

TLDR

Setup:

• AWS: t3a.micro (Kong -> Postgres) -> database (EC2 t3a.micro)
• Fly: (Postgres) -> database (EC2 t3a.micro)

Read

• AWS: 303/s
• Fly: 206/s

Write

• AWS 307/s
• Fly: 277/s

Notes

• Running from my local machine, so latency included
• AWS is going through cloudflare (DNS only), and then Kong, then Postgrest.
• Strange that read is slower than write (?)
• All systems deployed in Singapore

[awalias]

why is your AWS run so much slower than my results

[kiwicopple]

I was wondering the same actually. I just ran this code against our Benchmarks project. Is that how you did it?

[kiwicopple]

Benchmarks on Fly micro 1x: not a huge difference in throughput, but it did result in some duration thresholds to drop out

[kiwicopple]

I'll add a comparison table to the top comment