fixes#14715
Configures agents to use an address both in the Tailscale service prefix and the new Coder service prefix. Also modifies the Coordinator auth to allow the new prefix.
Updates `coder/tailscale` to include https://github.com/coder/tailscale/pull/62 which fixes a bug around forwarding TCP connections to localhost. This functionality is tested in the modifications to `TestAgent_Dial`.
re: #14715
This PR introduces the Coder service prefix: `fd60:627a:a42b::/48` and refactors our existing code as calling the Tailscale service prefix explicitly (rather than implicitly).
Removes the unused `Addresses` agent option. All clients today assume they can compute the Agent's IP address based on its UUID, so an agent started with a custom address would break things.
Fixes#12560
When gracefully disconnecting from the coordinator, we would send the Disconnect message and then close the dRPC stream. However, closing the dRPC stream can cause the server not to process the Disconnect message, since we use the stream context in a `select` while sending it to the coordinator.
This is a product bug uncovered by the flake, and probably results in us failing graceful disconnect some minority of the time.
Instead, the `remoteCoordination` (and `inMemoryCoordination` for consistency) should send the Disconnect message and then wait for the coordinator to hang up (on some graceful disconnect timer, in the form of a context).
Drops support for v1 of the tailnet API, which was the original coordination protocol where we only sent node updates, never marked them lost or disconnected.
v2 of the tailnet API went GA for CLI clients in Coder 2.8.0, so clients older than that would stop working.
Removes the support for the Agent's "legacy IP" which was a hardcoded IP address all agents used to use, before we introduced "single tailnet". Single tailnet went GA in 2.7.0.
fixes#14365
I bet what's going on is that in `connectToCoordinatorAndFetchResumeToken()` we call `Coordinate()`, send a message on the `Coordinate` client and then close it in rapid succession. We don't wait around for a response from the coordinator, so dRPC is likely aborting the call `Coordinate()` in the backend because the stream is closed before it even gets a chance.
Instead of using the Coordinator to record the peer ID assigned on the API call, we can wrap the resume token provider, since we call that API _and_ wait for a response. This also affords the opportunity to directly assert we get called with the right token.
First PR to address #14244.
Adds common potential reasons as to why a direct connection to the workspace agent couldn't be established to `coder ping`:
- If the Coder deployment administrator has blocked direction connections (`CODER_BLOCK_DIRECT`).
- If the client has no STUN servers within it's DERP map.
- If the client or agent appears to be behind a hard NAT, as per Tailscale `netInfo.MappingVariesByDestIP`
Also adds a warning if the client or agent has a network interface below the 'safe' MTU for tailnet. This warning is always displayed at the end of a `coder ping`.
* fix: avoid deleting peers on graceful close
- Fixes an issue where a coordinator deletes all
its peers on shutdown. This can cause disconnects
whenever a coderd is redeployed.
Code that was in `/clock` has been moved to github.com/coder/quartz. This PR refactors our use of the clock library to point to the external Quartz repo.
- `DERPForceWebSockets`: Test that DERP over WebSocket (as well as DERPForceWebSockets works). This does not test the actual DERP failure detection code and automatic fallback.
- `DERPFallbackWebSockets`: Test that falling back to DERP over WebSocket works.
Also:
- Rearranges some test code and refactors `TestTopology.StartServer` to be `TestTopology.ServerOptions` and take a struct instead of a function
Closes#13045
I initially made this change when hacking wgengine to also capture wireguard packets going into the magicsock, so that we could capture the initial wireguard handshake.
I don't think we should ship that additional capture logic, but... it seems generally useful to capture packets from the get go on speedtest, so that you can see disco and pings before the TCP speedtest session starts.
When an agent receives a node, it responds with an ACK which is relayed
to the client. After the client receives the ACK, it's allowed to begin
pinging.
Currently, importing `codersdk` just to interact with the API requires
importing tailscale, which causes builds to fail unless manually using
our fork.
This adds the ability for `TunnelAuth` to also authorize incoming wireguard node IPs, preventing agents from reporting anything other than their static IP generated from the agent ID.
Beginnings of a solution to #12297
Doesn't cover disco or definitively display whether we successfully connected to DERP, but shows some checklist diagnostics for connecting to an agent.
For this first PR, I just added it to `coder ping` to see how we like it, but could be incorporated into `coder ssh` _et al._ after a timeout.
```
$ coder ping dogfood2
p2p connection established in 147ms
pong from dogfood2 p2p via 95.217.xxx.yyy:42631 in 147ms
pong from dogfood2 p2p via 95.217.xxx.yyy:42631 in 140ms
pong from dogfood2 p2p via 95.217.xxx.yyy:42631 in 140ms
✔ preferred DERP region 999 (Council Bluffs, Iowa)
✔ sent local data to Coder networking coodinator
✔ received remote agent data from Coder networking coordinator
preferred DERP 10013 (Europe Fly.io (Paris))
endpoints: 95.217.xxx.yyy:42631, 95.217.xxx.yyy:37576, 172.17.0.1:37576, 172.20.0.10:37576
✔ Wireguard handshake 11s ago
```
In anticipation of needing the `LogSender` to run on a context that doesn't get immediately canceled when you `Close()` the agent, I've undertaken a little refactor to manage the goroutines that get run against the Tailnet and Agent API connection.
This handles controlling two contexts, one that gets canceled right away at the start of graceful shutdown, and another that stays up to allow graceful shutdown to complete.
I noticed a possible race where tailnet.Conn can try to dial the embedded region before we've set our custom dialer that send the DERP in-memory. This closes that race and adds a test case for servertailnet with no STUN and an embedded relay
I noticed in testing that the CLI wasn't correctly sending the disconnect message when it shuts down, and thus agents are seeing this as a "lost" peer, rather than a "disconnected" one.
What was happening is that we just used a single context for everything from the netconn to the RPCs, and when the context was canceled we failed to send the disconnect message due to canceled context.
So, this PR splits things into two contexts, with a graceful one set to last up to 1 second longer than the main one.
Jon Ayers