Wait loop when using JWKS

[SVerkuil]

Describe the bug
When using JWKS hasura-backend-plus will wait with starting until hasura (graphql-engine) is ready. But graphql-engine will never become ready because it will wait for the JWKS endpoint of hasura-backend-plus to become ready first.

To Reproduce
Steps to reproduce the behavior:

1. Create environment using provided docker compose
2. Change graphql-engine to use JWKS HASURA_GRAPHQL_JWT_SECRET: '{"type": "RS256", "jwk_url": "http://hasura-backend-plus:3000/auth/jwks"}'
3. Observe logs of hasura-backend-plus (crash loop) and graphql-engine (crash on JWKS)

Expected behavior
The hasura-backend-plus application should expose JWKS endpoint on app startup even if Hasura is not ready

Additional context
hasura-backend-plus 2.6.1 works fine, JWKS api is broken in 2.7.1

[pk992]

This can also happen on 2.6.1, if one container was stopped or recreated, it's impossible to sync them again.

[paulbalomiri]

second that.
I'm trying to automate deployment using ansible. The way to work around this is

• startup hasura (and hbp) excluding the config setting HASURA_GRAPHQL_JWT_SECRET
• after hbp is listening ( e.g. do a docker-compose logs -f hbp) restart the hasura service with docker-compose up -d this time with HASURA_GRAPHQL_JWT_SECRET properly set

my docker images are:

    image: hasura/graphql-engine:v2.0.3.cli-migrations-v3
    image: nhost/hasura-backend-plus:latest