hashicorp / consul

Consul is a distributed, highly available, and data center aware solution to connect and configure applications across dynamic, distributed infrastructure.
https://www.consul.io
Other
28.44k stars 4.43k forks source link

Consul TCP Checks Should Default to Bind Addr #4053

Open alkalinecoffee opened 6 years ago

alkalinecoffee commented 6 years ago

Looking at the documentation for service configs, this part has me a bit confused:

For a service definition:

The address field can be used to specify a service-specific IP address. By default, the IP address of the agent is used, and this does not need to be provided. The port field can be used as well to make a service-oriented architecture simpler to configure; this way, the address and port of a service can be discovered.

And for a TCP health check definition:

TCP + Interval - These checks make an TCP connection attempt every Interval (e.g. every 30 seconds) to the specified IP/hostname and port. If no hostname is specified, it defaults to "localhost".

It seems to me that the service's bind address should be honored as the default for all child healthcheck definitions for that service, instead of switching over to localhost. Otherwise, in the current scenario, there is no way to set up a TCP healthcheck for a service that listens only on the bind address without knowing the node's external IP in the first place.

This suggestion may be a backwards incompatible change, but it seems to be more in line with the service-to-health-check relationship.

If this is not an option, then a workaround like GetInteraceIP should perhaps be made available in the healthcheck definition as it is through command-line arguments so users can dynamically specify the target IP to a non-loopback address. Making envvars available to be injected similarly might also work, and be more flexible.

drawks commented 6 years ago

Making the service address and port available when defining checks would be great. I'd propose this be extended generally and not just to TCP checks:

Blefish commented 2 months ago

For members with dynamic IP, Consul properly reregisters itself to the cluster, however, services created manually, that have this tcp health check in place and use ip:port to register them originally, will be no longer passing health check.

Any way around this, after whole six years? Does http check require address or can it be set to just path?