control: map response long-poll timed out and Tailscale never reconnects

Fix "control: map response long-poll timed out" with no reconnect

TL;DR: When Tailscale logs this, the control-plane connection is dead and the daemon sometimes never notices. Restart tailscaled to recover, then add a tiny watchdog that restarts it automatically next time.

The error

control: map response long-poll timed out!

The node stops passing traffic and never recovers on its own. Common in containers, where nothing restarts the daemon for you.

Fix it

1. Recover right now

sudo systemctl restart tailscaled
tailscale status

Expected: status shows connected and traffic flows again.

In a container without systemd, restart the container or re-run the tailscaled process.

2. Add a watchdog so it self-heals

A simple approach: a cron job every few minutes that looks for the message and restarts the service.

#!/bin/bash
if journalctl -u tailscaled --since "10 minutes ago" | grep -q "map response long-poll timed out"; then
  systemctl restart tailscaled
fi

Expected: the next stall clears itself within minutes instead of waiting for you.

3. For containers: make the failure visible

If your orchestrator can restart on failure, have the watchdog exit nonzero instead of restarting in place, so the container gets recycled:

#!/bin/bash
if journalctl -u tailscaled --since "10 minutes ago" | grep -q "map response long-poll timed out"; then
  exit 1
fi

Expected: orchestrator restarts the container, Tailscale reconnects fresh.

When this applies

• Logs show control: map response long-poll timed out!
• The node stops working and stays broken until manual restart
• Especially containers and always-on appliances

When it does not apply

• Occasional long-poll timeouts that recover on their own (normal on flaky networks)
• DERP relay errors (different subsystem)
• Login or auth failures

Tool compatibility

All Tailscale versions; reported on Linux containers and Debian hosts. The journalctl watchdog needs systemd; adapt the log source for other setups.

Variant phrasings

Node goes dark after WAN/IP change, only restart helps

Same family. One reporter found nothing but a restart recovered the DERP path after an ISP IP change. The watchdog covers this too.

Why it happens

The long-poll is how the client hears about network updates. If the underlying TCP connection stalls silently (middleboxes, NAT timeouts), the client can sit forever waiting on a dead socket instead of reconnecting.

Edge cases

• Do not watchdog too aggressively: checking every minute and restarting on any single timeout can flap a node that would have recovered. Five to ten minute windows are saner.
• Stalled TCP variant: one reporter instead watched for unsent bytes piling up on the control-plane connection and killed just that connection. If restarts feel heavy, that is the lighter scalpel.
• This is a known upstream gap: the issue asks for self-detection; until that lands, the watchdog is the supported-by-experience answer.