DEV Community

Lourdes Paul Agilan
Lourdes Paul Agilan

Posted on

I load-tested PocketBase and Supabase with the same tests: the numbers

I ran the same tests on PocketBase and Supabase from one small VPS. These are the numbers, and what they do and don't show.

Setup

  • Table: contacts with name, email, a notes field of about 200 characters and a created timestamp.
  • Client: a 1 vCPU, 4 GB VPS in Boston running curl and Apache Bench.
  • PocketBase 0.40.6: on the same VPS, plain HTTP.
  • Supabase: free plan (Nano compute), East US, reached over the internet from the same VPS.
  • Rows: about 300, then 10,000, then 100,000. Each test ran once.

PocketBase never touched a network. Supabase did.

Single requests (median)

Request PocketBase Supabase
Get one row by id 0.79 ms 228 ms
List 20 rows 2.2 ms 234 ms

The VPS is about 9 ms from Supabase's edge, and a request rejected at the edge takes 40 to 55 ms. Requests that reach the database took about 230 ms. A Supabase project in Sydney gave a similar median (253 ms), so distance does not explain it. I don't know the cause.

Load

PocketBase, load tool sharing the single CPU:

  • Reading one row: about 3,800 requests a second at 50 connections, 4,167 at 1,000.
  • Creating rows: about 2,300 to 2,600 requests a second.
  • 0 failed requests at 1,000 connections. Memory rose from 123 MB to 317 MB.

Supabase, free plan:

Connections Requests/s Median latency
1 2 319 ms
50 153 301 ms
200 548 311 ms
400 663 532 ms
800 806 936 ms

0 failed requests and no throttling at any step. One connection manages about 3 requests a second because each request waits about 300 ms, so throughput only rises with concurrency.

The 100,000-row list

Both slowed down when sorting and filtering a large table.

  • PocketBase: with 25 users at once, a sorted list of 20 fell to 8.5 requests a second (median 1.9 s, p95 10.8 s). An index on the sort field plus skipTotal=1 brought one list to about 1 ms and 25 users to 1,311 requests a second.
  • Supabase: without an index, 25 users got about 15 requests a second (median 1.35 s). An index on created_at raised it to 69 requests a second at 25 connections and 577 at 200.

Getting one row by id stayed fast at every size. The fix is the same on both: index the column you sort by. PocketBase also counts matching rows on every list call unless you pass skipTotal=1.

Which one

  • PocketBase if you want the cheapest setup, can run a server and your backend code sits next to the database.
  • Supabase if you want Postgres, someone else running the database, or per-user data rules through row level security.

Limits

One run per test, one small VPS, one client location. PocketBase shared a CPU with the load tool and ran over loopback. Supabase was tested on the smallest compute. I did not test file uploads, Edge Functions, PocketBase hooks, or Supabase sign-in and realtime.

The full write-up, with the login and realtime results, is on Indie Torch: PocketBase vs Supabase: Benchmarked Side by Side.

Top comments (0)