mirror of
https://github.com/tailscale/tailscale.git
synced 2026-11-20 00:31:17 -08:00
macOS's mDNSResponder negatively caches NXDOMAIN/NODATA answers on its own schedule, ignoring the negative TTL the MagicDNS resolver asks for. When a Mac resolves a peer's name shortly before that node joins (or is renamed to that name), it keeps answering "no such host" long after Tailscale itself knows the peer. Users report this as the new node not existing. Add scripts/macos-dns-cache-check.sh, which runs "tailscale status --json", asks mDNSResponder (via dscacheutil, which uses the same lookup path as other macOS apps; dig and host bypass it) for the A/AAAA records of every peer's DNSName, and compares them to the peer's TailscaleIPs. It reports how many peers mismatch and how (no answer, missing addresses, unexpected addresses), and if any do, flushes both cache layers (dscacheutil -flushcache and killall -HUP mDNSResponder), then re-checks the mismatched peers. Under sudo it runs "tailscale status" as the invoking user first, since the App Store variant only lets the CLI find the GUI's local API when they run as the same user. It uses jq when present and falls back to JavaScript for Automation for JSON parsing, since older macOS versions don't ship jq. Updates tailscale/corp#45631 Signed-off-by: Brad Fitzpatrick <[email protected]> Change-Id: I7c3e9a1b52d84f6e0a9d3c7b8e1f4a2d6c5b9e08