Running a tailnet?

Tailscale ACLs or subnet routes misbehaving? We sort it out.

Tailscale is a wonderful thing until a tailnet lock policy quietly blocks a new node, or a subnet router stops advertising routes after a reboot, or an ACL file has a typo that nobody will own up to. We read the config, trace the actual routing path, and leave the network in a state where a new hire does not have to ask three people before they can reach the staging database.

  • Since 2015Fixing technology
  • FreeEstimates
  • US teamReal engineers

tailscaled.log

still broken

What it is

Tailscale at a glance

Tailscale is a zero-config mesh VPN built on WireGuard. Each device gets a stable 100.x address inside your tailnet, finds its peers through a coordination server, and then talks directly whenever NAT allows, falling back to DERP relays when it does not.

The product bundles identity-aware access, MagicDNS, exit nodes, subnet routers, Taildrop and Tailscale SSH. Pricing starts free for small teams and moves to per-user tiers with features like tailnet lock, SCIM provisioning and audit log streaming. Teams reach for it instead of a legacy IPsec VPN or a hand-rolled WireGuard mesh because the operational cost is low and the identity model is grounded in SSO rather than static keys.

Type
Mesh VPN on WireGuard with identity-based ACLs
First released
2019 public release
License
Proprietary control plane, open-source client (BSD-3-Clause)
Maintained by
Tailscale Inc.
Typically used for
Remote access, service-to-service networking, bastion replacement, hybrid cloud, developer connectivity

The honest version

Tailscale pros and cons

No technology is all upside. This is where Tailscale earns its place and where it tends to cause trouble.

Where Tailscale is strong

  • Setup is genuinely minutes, not daysA fresh machine joins the tailnet with a single command and an SSO login, with no certificate dance or firewall change required on either end.
  • Identity is the access modelACLs describe who can reach what in terms of users, groups and tags rather than IP ranges, which survives a team changing size or shape.
  • NAT traversal just worksDirect connections are negotiated transparently; when NAT is hostile the traffic falls back to DERP without the user noticing.
  • MagicDNS is usefulShort hostnames resolve across the tailnet without a DNS server to run, so people stop memorising ten-digit CGNAT addresses.
  • Audit and lock features for grown-upsTailnet lock, SSH session recording and streaming audit logs mean you can run it in regulated environments without bolting on a second product.

Where Tailscale bites

  • The control plane is a dependencyIf the Tailscale coordination server is unreachable, new nodes cannot join and keys cannot refresh, even though existing connections keep working. Headscale exists for the fully self-hosted case, with its own trade-offs.
  • ACLs are JSON that is easy to get wrongThe HuJSON file is powerful and unforgiving. A single typo in a tag or group can quietly cut off a team until someone runs the preview mode.
  • Pricing climbs with user count, not usageTeams that want SSO, tailnet lock or audit streaming are on the paid tier, and the per-user cost adds up once contractors and service accounts enter the picture.
  • Subnet routers are a single point of failureA tailnet reaching into a cloud VPC through one subnet router goes dark when that node reboots or fails, so high availability needs more than the default configuration.
  • Compliance posture is situationalTailscale has strong controls, but some regulated environments still require a self-hosted control plane or a review of DERP relay locations, and that conversation is not free.
  • Exit nodes are tempting and misusedRouting all traffic through a single exit node is easy to enable and easy to overload, and it hides a lot of performance problems until someone notices the latency.

Sound familiar?

Common Tailscale problems we fix

The Tailscale problems that bring people to us, and what you have once each one is fixed.

subnet route not advertised

A subnet router will not accept traffic

An advertised route that was never approved in the admin console, IP forwarding disabled on the Linux host, or an ACL that does not grant the requesting user access to that tag.

Routes advertised, approved and reachable

ACL compile error line 42

The ACL file fails to apply

A tag that was renamed in one place and not another, a group that no longer exists in the IdP sync, or a syntax error in the HuJSON that nobody spotted in review.

ACL compiles, tests pass

tailnet lock requires signing key

A new node cannot join

Tailnet lock is enabled and the node signing keys were rotated without updating the published set, so every new machine is quarantined until an admin signs for it.

Lock policy clean, nodes joining

MagicDNS NXDOMAIN

MagicDNS stops resolving internal hosts

A split DNS configuration pointing internal zones at a resolver that is itself only reachable over the tailnet, so the client cannot even find the resolver.

DNS working from a cold boot

exit node ip_forward disabled

The exit node forwards nothing

A fresh Linux exit node where net.ipv4.ip_forward and net.ipv6.conf.all.forwarding were never set, so advertised exit routes reach the node and die there.

Exit node routing cleanly

DERP relay in use

Everything is slower than it should be

Nodes falling back to DERP because direct connections cannot be negotiated, often because of symmetric NAT on both ends or a firewall blocking UDP.

Direct connections restored

key expiry in 14 days

Keys expire and services go offline

Headless servers with user keys rather than tagged keys silently expiring, taking the subnet router or the production bastion offline overnight.

Keys non-expiring where they should be

What we do

Tailscale help, from setup to day-two

Call us for one broken ACL, or hand us the whole tailnet.

Rescue
A subnet router down, ACLs blocking real work, or MagicDNS not resolving. We get traffic flowing and then explain what changed.
ACL design
A HuJSON policy that reflects your actual groups and tags, with a test file that fails loudly when something is renamed.
Subnet and exit node topology
High-availability subnet routers, sensibly placed exit nodes, and a map of what sits where.
SSO and provisioning
SCIM provisioning, SSO wired to your IdP, tailnet lock turned on sensibly, and offboarding that actually revokes access.
Migration from legacy VPN
Moving teams off OpenVPN, Cisco AnyConnect, legacy IPsec or a hand-built WireGuard mesh, without a two-week outage.
Managed Tech
ACL reviews, key rotation reminders, upgrades of the tailscaled client, and an engineer available when a subnet router falls over at an awkward hour.

How it works

From first message to fixed

  1. Look Free estimate

    Tell us what's going wrong. We look at it, then tell you plainly what's wrong and what it will cost.

  2. Fix

    We fix the cause, not the symptom. If something is down, getting it back up comes first.

  3. Harden

    Backups, monitoring, updates and access, so the same problem can't quietly come back.

  4. Hand back or stay on

    You get a plain-language write-up. Or we stay on for hosting, upkeep and new work.

Often found together

Running Tailscale alongside something else?

We work on the whole stack, so the problem can be wherever it actually is.

Questions

Tailscale questions

Something else? Call+1 737 727 7327

Is Tailscale a real replacement for a traditional VPN?

For developer access and service-to-service connectivity, yes, and it is usually a strict improvement. For some compliance regimes that require network-level isolation or a self-hosted control plane, the answer is nuanced. We look at the regulatory posture, the identity model and the traffic patterns before recommending a wholesale replacement.

Why can one machine not see another on our tailnet?

Three things cover most cases. The ACL file does not grant the source user or tag access to the destination tag. The destination is behind a subnet router that is offline or whose route was never approved. Or a firewall on the destination is blocking the Tailscale interface even though the tailnet itself is fine. Running tailscale ping and tailscale status quickly tells us which one.

Should we self-host with Headscale instead?

Sometimes. Headscale is a useful open-source control plane for teams that genuinely need no third-party coordination server, usually for compliance reasons. It costs real operational time to run, lags behind Tailscale on features, and does not include DERP by default. For most teams, the managed service is still the better trade.

How do we stop keys from expiring on our servers?

Headless servers should use tagged auth keys with the non-expiring flag set, bound to a specific tag the ACL understands. User keys are designed to expire and should not be used on production nodes. Enforcing this in the admin console removes the whole class of 2 a.m. incidents caused by a quiet expiry.

Can Tailscale reach a VPC we cannot install an agent on?

Yes, through a subnet router. A small instance inside the VPC runs the Tailscale client and advertises the private subnet as a route. For high availability you run two of them, and for cloud networks with overlapping ranges you use the 4via6 route trick. We would plan those pieces up front rather than discovering them after launch.

How much does Tailscale help cost?

Estimates are free. Tell us how many users, which clouds the tailnet reaches into, and what is currently going wrong, and we return a quote before any policy changes land.

Free estimate

Tailscale giving you trouble?

Tell us what you're running and what's going wrong. A person reads it and comes back with an estimate.

  • No charge to find out what it will take
  • If it's urgent, say so

Replies come from a person. By sending, you authorize us to email you back. No marketing.