Mike O'Driscoll
858b53b289
derp.Client.ForwardPacket flushed on every call, so each packet a derper forwards to a mesh peer cost its own TLS record and write syscall, serialized behind the client's write mutex with the syscall inside the critical section. ForwardPacket now appends the frame to one of two 16 KiB buffers and returns. A flusher goroutine, started if none is running, swaps the buffer for the spare under the mutex and writes it with the mutex released, so callers keep filling the spare while a write is in progress and frames from concurrent callers share a write. A caller that finds the spare full waits for the write to complete, preserving backpressure. The buffers are allocated on first forward, so only mesh clients carry them. Forward buffers are written directly to the connection rather than through the client's bufio.Writer. A new mutex serializes those writes with the client's other senders, since not every Conn tolerates concurrent writes and bufio can split one frame over several. A failed write is logged with the number of frames dropped, closes the connection, wakes waiting callers with the error, and is returned by every later call, so derphttp reconnects as before. Adds BenchmarkForwardPacket, which also reports percentiles of ForwardPacket's own latency, unit tests for batching, backpressure, oversized frames, write failure, and write serialization, and an end-to-end test that concurrently forwarded packets all reach the destination. Updates tailscale/corp#50628 Signed-off-by: Mike O'Driscoll <mikeo@tailscale.com>
DERP
This directory (and subdirectories) contain the DERP code. The server itself is
in ../cmd/derper.
DERP is a packet relay system (client and servers) where peers are addressed using WireGuard public keys instead of IP addresses.
It relays two types of packets:
-
"Disco" discovery messages (see
../disco) as the a side channel during NAT traversal. -
Encrypted WireGuard packets as the fallback of last resort when UDP is blocked or NAT traversal fails.
DERP Map
Each client receives a "DERP Map" from the coordination server describing the DERP servers the client should try to use.
The client picks its home "DERP home" based on latency. This is done to keep costs low by avoid using cloud load balancers (pricey) or anycast, which would necessarily require server-side routing between DERP regions.
Clients pick their DERP home and report it to the coordination server which shares it to all the peers in the tailnet. When a peer wants to send a packet and it doesn't already have a WireGuard session open, it sends disco messages (some direct, and some over DERP), trying to do the NAT traversal. The client will make connections to multiple DERP regions as needed. Only the DERP home region connection needs to be alive forever.
DERP Regions
Tailscale runs 1 or more DERP nodes (instances of cmd/derper) in various
geographic regions to make sure users have low latency to their DERP home.
Regions generally have multiple nodes per region "meshed" (routing to each
other) together for redundancy: it allows for cloud failures or upgrades without
kicking users out to a higher latency region. Instead, clients will reconnect to
the next node in the region. Each node in the region is required to be meshed
with every other node in the region and forward packets to the other nodes in
the region. Packets are forwarded only one hop within the region. There is no
routing between regions. The assumption is that the mesh TCP connections are
over a VPC that's very fast, low latency, and not charged per byte. The
coordination server assigns the list of nodes in a region as a function of the
tailnet, so all nodes within a tailnet should generally be on the same node and
not require forwarding. Only after a failure do clients of a particular tailnet
get split between nodes in a region and require inter-node forwarding. But over
time it balances back out. There's also an admin-only DERP frame type to force
close the TCP connection of a particular client to force them to reconnect to
their primary if the operator wants to force things to balance out sooner.
(Using the (*derphttp.Client).ClosePeer method, as used by Tailscale's
internal rarely-used cmd/derpprune maintenance tool)
We generally run a minimum of three nodes in a region not for quorum reasons (there's no voting) but just because two is too uncomfortably few for cascading failure reasons: if you're running two nodes at 51% load (CPU, memory, etc) and then one fails, that makes the second one fail. With three or more nodes, you can run each node a bit hotter.