Legacy 8080 Health Checks Overriding Explicit Port 80

[yaniv]

Despite explicit configuration for port 80 health checks and disable_implicit_checks=true, Fly.io’s infrastructure continues to enforce legacy 8080 health checks, causing false “critical” status and automatic machine stoppages.

Evidence of Configuration:

1. fly.toml Configuration:

[http_service]
  disable_implicit_checks = true
  internal_port = 80

[[http_service.checks]]
  type = "http"
  port = 80
  path = "/health"
  interval = "10s"
  timeout = "2s"
  grace_period = "5s"

2. Current Health Check Status:

$ fly checks list -a x-realestate
NAME                      | STATUS   | OUTPUT
servicecheck-00-http-8080 | critical | connect: connection refused

Diagnostic Evidence:

1. Port Verification:

$ fly ssh console -a x-realestate -C "netstat -tuln"
Active Internet connections (only servers)
Proto  Local Address        Foreign Address      State
tcp    0.0.0.0:80          0.0.0.0:*            LISTEN
tcp6   :::80               :::*                 LISTEN

2. Health Check Simulation:

# Port 80 succeeds:
$ curl -I http://localhost:80/health
HTTP/1.1 200 OK

# Port 8080 fails (as expected):
$ curl -I http://localhost:8080/health
curl: (7) Failed to connect to port 8080

3. Deployment Logs Show Conflict:

2025-06-23T15:58:29Z proxy[...] waiting for machine on 0.0.0.0:8080
2025-06-23T15:58:32Z proxy[...] failed to connect to machine: gave up after 15 attempts

Steps Taken to Resolve:

1. ✓ Confirmed NGINX only listens on port 80
2. ✓ Verified disable_implicit_checks is properly set
3. ✓ Multiple deployments with --force-manifest
4. ✓ Complete machine restarts
5. ✓ Attempted health check handoff procedures

Technical Environment:

• Fly CLI v0.1.47
• NGINX 1.27.5
• App Region: ams
• Machine Size: shared-cpu-1x

Request:
Please have Fly’s platform team:

1. Manually clear legacy health check configurations for x-realestate
2. Verify backend service registration respects explicit port 80 checks
3. Confirm complete removal of implicit 8080 check behavior

Supporting Data:
Full logs and configuration available upon request. This behavior persists across multiple app restarts and deployments.

[mayailurus]

Hi… Overall, I’d suggest removing all keys and sections from fly.toml that aren’t explicitly documented in the following:

https://fly.io/docs/reference/configuration/

For example, http_service.checks doesn’t allow port or type fields, as far as I know.

Also, look for overlapping [http_service] and [[services]] sections (i.e., ones having the same internal_port or port), since those tend to confuse the Fly.io infrastructure.

If you still encounter problems, please do post the full, revised fly.toml.

Hope this helps a little!