Go's Mac builders run at Mac hosting provider, MacStadium. We have a VMware cluster with 10 physical Mac Minis on which we run two VMs each.
In addition to 20 Mac VMs, we also run one 1 Linux VM that's our bastion host & runs various services.
The bastion host is macstadiumd.golang.org and can be accessed via:
$ ssh -i ~/keys/id_ed25519_golang1 email@example.com
id_ed255519_golang is available from http://go/go-builders-ssh)
It also runs:
a DHCP server for the 20 Mac VMs to get IP addresses from (systemd unit
isc-dhcp-server.service, so watch with
journalctl -f -u isc-dhcp-server.service)
the makemac daemon daemon (systemd unit
makemac.service, so watch with
journalctl -f -u makemac). This monitors the build coordinator (farmer.golang.org) as well as the VMware cluster (via the
govc CLI tool) and makes and destroys Mac VMs as needed. It also serves plaintext HTTP status at http://macstadiumd.golang.org:8713 which used to be accessible in a browser, but recent HSTS configuration on
*.golang.org means you need to use curl now. But that‘s a good way to see what it’s doing remotely, without authentication.
WireGuard listening on port 51820, because OpenVPN + Mac‘s built-in VPN client is painful and buggy. See config in
/etc/wireguard/wg0.conf. This doesn’t yet come up on boot. It's a recent addition.
The method of last resort to access the cluster, which works even if the bastion host VM is down, is to VPN to our Cisco something gateway.
TODO: put this info somewhere. It's currently circulated in email between Brad, Russ, and Dmitri in an email with some subject “macOS VM info”.
Once you've connected to either OpenVPN or WireGuard, you can hit the VMware web UI at:
ssh -D to the bastion host to make a SOCKS tunnel and configure your browser to send proxy through that SOCKS tunnel.)
Common techniques to debug:
Can you get to the bastion host?
journalctl -f -u makemac say? Is it error looping?
Look at https://10.88.203.9/ui/ and see if VMware is unhappy about things. Did hosts die? Did storage disappear?
Need to hard reboot machines? Eventually we‘ll fix https://github.com/golang/go/issues/32033 but for now you can use https://portal.macstadium.com/subscriptions and power cycle machines that VMware reports dead (or that you see missing from https://farmer.golang.org’s reverse pool info).
Worst case, file a ticket: https://portal.macstadium.com/tickets