Expose propolis' serial console channel in nexus #1766
Conversation
lifning
force-pushed
the
passme
branch
5 times, most recently
from
7516059
to
6a6d5a8
Compare
lifning
force-pushed
the
passme
branch
8 times, most recently
from
2d30ac3
to
8e11bff
Compare
| /// Connect to an instance's serial console | ||
| #[channel { | ||
| protocol = WEBSOCKETS, | ||
| path = "/organizations/{organization_name}/projects/{project_name}/instances/{instance_name}/serial-console/stream", |
There was a problem hiding this comment.
Is the idea we'll have both the streaming and non-streaming endpoint in the limit? I assume the latter for the buffer playback behaviour?
There was a problem hiding this comment.
the streaming endpoint will eventually get params to start from an arbitrary byte offset (within the buffer) as well once i've moved the longer buffering into propolis-server, at which point the main reasons to keep the non-streaming one around would be
- wanting to separate out permission to read it vs. read/write
- wanting to provide an API for accessing it that doesn't require websocket support
nexus/src/app/instance.rs
Outdated
| } | ||
| } | ||
| } | ||
| // pass along Close messages |
There was a problem hiding this comment.
Should we also try to gracefully send a close to the other side if one unexpectedly drops?
| @@ -25,7 +25,7 @@ use uuid::Uuid; | |||
|
|
|||
| use super::sled_agent::SledAgent; | |||
|
|
|||
| use propolis_client::api::VolumeConstructionRequest; | |||
| use crucible_client_types::VolumeConstructionRequest; | |||
There was a problem hiding this comment.
Will the generated client not re-export types it uses?
There was a problem hiding this comment.
kind of a rough edge i think: progenitor re-declares a copy of them based on the openapi spec it's given (which is of course generated using the derive(JsonSchema) on the original crucible_client_types types). unless there's a fancy feature of progenitor i'm missing, this means we end up needing boilerplate to convert between the two
There was a problem hiding this comment.
ah right i know you added the From impls for the generated-migration feature but I wonder if its worth just having in general too
There was a problem hiding this comment.
i get the distinct feeling it might be, yeah.
- Includes updates to progenitor and reqwest 0.11.12 for client support for Dropshot websocket channels. - Literally only passes through the websocket, does not yet include the serial output buffering done by its parent GET endpoint. - Adds support for ensuring a websocket endpoint gets upgraded in nexus integration tests (i.e. test_unauthorized). - Begins migrating sled-agent's propolis-client usage to the new progenitor-generated version (oxidecomputer/propolis#206)
I'll update how-to-run.adoc after either oxidecomputer/cli#268 lands (or if we implement it in oxide-sdk-rust instead)