healthcheck doesn't respect port changes

[its-a-feature]

Version Information

Server Version: v2.44.0 CLI Version (for CLI related issue):

Environment

Docker

What is the current behaviour?

In docker, I set the HASURA_GRAPHL_SERVER_PORT environment variable so that Hasura uses a port other than 8080. This works fine so that the server is running on a different port. However, the healthz periodic healthcheck always seems to use port 8080. This means that when I switch my port, even though everything is working, I get an unhealthy status in Docker. This also means that if something else uses port 8080, then it gets spammed with these healthz checks.

What is the expected behaviour?

If I change the port of the server to something else, then the healthz healthchecks should also use that port for checking.

How to reproduce the issue?

1. Use Docker and set HASURA_GRAPHL_SERVER_PORT=8081
2. Start the container and wait a minute or two
3. Check the health of the container and it'll be unhealthy

Please provide any traces or logs that could help here.

Any possible solutions/workarounds you're aware of?

None that I know of outside of maybe wrapping the container in another container and recreating the healthcheck with the right port?

Keywords

healthz, 8080, healthcheck

[rakeshkky]

Hi @its-a-feature, can you please confirm the env var name?

In docker, I set the HASURA_GRAPHL_SERVER_PORT environment variable

It is not a valid env var name. HASURA_GRAPHQL_SERVER_PORT it is. See https://hasura.io/docs/2.0/deployment/graphql-engine-flags/reference/#server-port for more details.

[its-a-feature]

Sorry, meant to reply sooner. Yes, I'm using HASURA_GRAPHQL_SERVER_PORT (it was a typo in this post). I know that's working too because the server port is being changed from 8080 to whatever I specify, but it's just the healthcheck isn't respecting it. Not sure if there's a different environment variable I should be setting for that or not

[its-a-feature]

My initial thought is that since this is the Docker version, maybe a pre-compiled version of whatever is doing the healthcheck is hardcoded to 8080 or maybe the built-in healthcheck for the hasura container is hardcoded somewhere?

[its-a-feature]

https://hub.docker.com/layers/hasura/graphql-engine/v2.36.8-ce.cli-migrations-v2/images/sha256-741c6ed6b2cdc16faf206df9e006a1ae34d1e31dbd6028ff2aa958597690a29e?context=explore ^ right there in the Docker image it's hardcoded to 8080

[rakeshkky]

Oh! Okay. The cli-migrations image has the health check port hard-coded. Let's see if we can override the healthcheck defined in a dockerfile.

• Using docker run command - https://docs.docker.com/engine/containers/run/#healthchecks.
• In docker compose file - https://docs.docker.com/reference/compose-file/services/#healthcheck.

If this solves your issue, please feel free to close this ticket. Thank you.

[its-a-feature]

I can make the mod in a custom Docker image/Compose file, but you should note in the docs for healthcheck that it doesn't respect the port changes. Similarly, when I try to use hasura-cli within the Docker image, it also doesn't respect the port change:

root@mythic_graphql:/test# hasura-cli metadata export
ERRO connecting to graphql-engine server failed   
INFO possible reasons:                            
INFO 1) Provided root endpoint of graphql-engine server is wrong. Verify endpoint key in config.yaml or/and value of --endpoint flag 
INFO 2) Endpoint should NOT be your GraphQL API, ie endpoint is NOT https://hasura-cloud-app.io/v1/graphql it should be: https://hasura-cloud-app.io 
INFO 3) Server might be unhealthy and is not running/accepting API requests 
INFO 4) Admin secret is not correct/set           
INFO                                              
FATA[0000] making http request failed: Get "http://localhost:8080/v1/version": dial tcp [::1]:8080: connect: connection refused 

Is there some other variable or thing I can do to get the hasura-cli to respect the port change as well?

[rakeshkky]

INFO 1) Provided root endpoint of graphql-engine server is wrong. Verify endpoint key in config.yaml or/and value of --endpoint flag

If not, please try this instruction to make hasura-cli to contact the engine URL with different port.

[nikola-s7]

I just ran into this issue as well. Any updates on whether this will be addressed?