Fixes#13139
Using a bare channel to signal dependent goroutines means that we can only signal success, not failure, which leads to deadlock if we fail in a way that doesn't cause the whole `apiConnRoutineManager` to tear down routines.
Instead, we use a new object called a `checkpoint` that signals success or failure, so that dependent routines get unblocked if the routine they depend on fails.
Removes a check for `context.Canceled` inside the `handleManifest` routine. This checking is handled in the `apiConnRoutineManager`, so checking inside the handler is redundant.
Currently, importing `codersdk` just to interact with the API requires
importing tailscale, which causes builds to fail unless manually using
our fork.
Changes the agent to use the new v2 API for sending logs, via the logSender component.
We keep the PatchLogs function around, but deprecate it so that we can test the v1 endpoint.
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.
The agent is extended with a `--script-data-dir` flag, defaulting to the
OS temp dir. This dir is used for storing `coder-script-data/bin` and
`coder-script/[script uuid]`. The former is a place for all scripts to
place executable binaries that will be available by other scripts, SSH
sessions, etc. The latter is a place for the script to store files.
Since we default to OS temp dir, files are ephemeral by default. In the
future, we may consider adding new env vars or changing the default
storage location. Workspace startup speed could potentially benefit from
scripts being able to skip steps that require downloading software. We
may also extend this with more env variables (e.g. persistent storage in
HOME).
Fixes#11131
This commit refactors where custom environment variables are set in the
workspace and decouples agent specific configs from the `agentssh.Server`.
To reproduce all functionality, `agentssh.Config` is introduced.
The custom environment variables are now configured in `agent/agent.go`
and the agent retains control of the final state. This will allow for
easier extension in the future and keep other modules decoupled.
`agentsdk` depends on `agent/proto` because it needs to get the version to dial.
Therefore, the conversion routines need to live in `agentsdk` so that we can convert to and from the Manifest.
I briefly considered refactoring the agent to only reference `proto.Manifest`, but decided against it because we might have multiple protocol versions in the future, its useful to have a protocol-independent data structure.
This one is huge, and I'm sorry.
The problem is that once I change `tailnet.Conn` to start doing v2 behavior, I kind of have to change it everywhere, including in CoderSDK (CLI), the agent, wsproxy, and ServerTailnet.
There is still a bit more cleanup to do, and I need to add code so that when we lose connection to the Coordinator, we mark all peers as LOST, but that will be in a separate PR since this is big enough!
- An opt-in feature has been added to the agent to allow
deprioritizing non coder-related processes for CPU by setting their
niceness level to 10.
- Opting in to the feature requires setting CODER_PROC_PRIO_MGMT to a non-empty value.
* chore: add /v2 to import module path
go mod requires semantic versioning with versions greater than 1.x
This was a mechanical update by running:
```
go install github.com/marwan-at-work/mod/cmd/mod@latest
mod upgrade
```
Migrate generated files to import /v2
* Fix gen
I forgot that waiting on the cond releases the lock so it was possible
to get pty output after writing the buffer but before adding the pty to
the map. To fix, add the pty to the map while under the same lock where
we read from the buffer.
The rest does not need to be behind the lock so I moved it out of
doAttach, and that also means we no longer need
waitForStateOrContextLocked.
Also, this can hit a logger error saying the attach failed which fails
the tests however it is not that the attach failed, just that the
process already ran and exited, so when the process exits do not
set an error, instead for now assume this is an expected close.
* Add screen backend for reconnecting ptys
The screen portion is a port from wsep. There is an interface that lets
you choose between screen and the previous method. By default it will
choose screen if it is installed but this can be overidden (mostly for
tests).
The tests use a scanner instead of a reader now because the reader will
loop infinitely at the end of a stream.
Replace /bin/bash with bash since bash is not always in /bin.
* Remove connection_id from reconnecting PTY logger
This serves multiple connections so it makes no sense to scope it to a
single connection.
Also lets us use "connection_id" when logging write errors instead of
"other_conn_id".
* Use PATH to test buffered reconnecting pty
