Networking between concurrent jobs on GitHub-hosted runners in the same Workflow run

[iskunk]

I am using a Workflow run to build a large project (the Chromium browser). A single GitHub-hosted runner does not have enough CPU power to complete the build within the 6 hour job time limit.

I'd thus like to make use of multiple concurrent jobs, with distcc or similar on each runner. This will distribute the compilation workload so that the build can finish in the allotted time.

However, this requires the jobs/runners to communicate with each other on the network. The runners appear to run sshd, but when one runner attempts to connect to the SSH port of another, the connection times out. Flushing the iptables/nft firewall rules does not help. There appears to be some default network isolation in place, enforced outside of the runner itself.

I can distribute the required host key and private keys securely using the artifacts and secrets mechanisms, but at a minimum I need to get a TCP connection going between the lead runner and its helpers. How can I set this up?

(Note that using an external "bounce host" is not feasible, because (1) such a host could pose a security risk if compromised, given that intermediate build artifacts will pass through it; (2) the communication bandwidth is likely to be significant (multiple gigabytes), and (3) no such host is available to me.)