depends_on alone only waits for the container to start, not for the process inside to be ready. Add a healthcheck and depend on that condition:

1services:
2  db:
3    image: postgres:17
4    healthcheck:
5      test: ["CMD-SHELL", "pg_isready -U postgres"]
6      interval: 5s
7      timeout: 3s
8      retries: 10
9
10  app:
11    depends_on:
12      db:
13        condition: service_healthy

1services:
2  db:
3    image: postgres:17
4    healthcheck:
5      test: ["CMD-SHELL", "pg_isready -U postgres"]
6      interval: 5s
7      timeout: 3s
8      retries: 10
9
10  app:
11    depends_on:
12      db:
13        condition: service_healthy

That said — make the app retry its initial connection anyway. In production there is no Compose orchestrating startup order, and databases restart independently of your app. The healthcheck fixes local development; retry logic fixes reality.

Second the retry point. Exponential backoff with a cap, and log each attempt — otherwise a misconfigured host string looks identical to a slow-starting database.