network delays when fetching providers

[jbardin]

Some users have reported reproducible network issues when fetching providers. These have manifested as network timeouts, or just extremely slow throughput.

See #15798 and #15481

#16751 may also be related, but that is happening when connecting to the registry.

#16448 is more intermittent, and manifests as a TLS handshake timeout.

In each of these cases curl still appears to work as expected, which leads me to believe that the difference is related to the Go network stack. The 2 primary pieces of concern are the the Dialer and the http2 transport.

• curl is using http/1.1 by default, where as Terraform is using http2.
• curl is using the host resolver, and Terraform uses the go resolver.
• "Happy eyeballs" is not enabled by default in Terraform.

I tested various routes by spinning up hosts in most global AWS regions, and a few DO regions. None of the hosts had any trouble fetching modules or provider releases, nor were any particularly slow. Those hosts were also dual stack when possible.

What I did find was that null routing the ipv6 packets caused the exact same delay patterns as shown in the comments of #15798. Since this is the broken network configuration that rfc6555 was written for, enabling the dual stack dialer in Terraform should fix this particular issue.

[jbardin]

Confirmed that enabling "happy eyeballs" in Terraform fixes the broken ipv6 issue as expected.
I'll get a patch upstream asap.

[jbardin]

Since the only outstanding related issues seem to have been traced to internet provider and network problem I'm going to close this one as complete.

[ghost]

I'm going to lock this issue because it has been closed for 30 days ⏳. This helps our maintainers find and focus on the active issues.

If you have found a problem that seems similar to this, please open a new issue and complete the issue template so we can capture all the details necessary to investigate further.