feat(server): model remote compute driver endpoints

Signed-off-by: Evan Lezar <elezar@nvidia.com>

Author: elezar

architecture/compute-runtimes.md

@@ -33,8 +33,8 @@ when a sandbox create request asks for GPU resources.
 | Docker | Local development with Docker available. | Container plus nested sandbox namespace. | Uses host networking so loopback gateway endpoints work from the supervisor. |
 | Podman | Rootless or single-machine deployments. | Container plus nested sandbox namespace. | Uses the Podman REST API, OCI image volumes, and CDI GPU devices when available. |
 | Kubernetes | Cluster deployment through Helm. | Pod plus nested sandbox namespace. | Uses Kubernetes API objects, service accounts, secrets, PVC-backed workspace storage, and GPU resources. |
-| VM | Experimental microVM isolation. | Per-sandbox libkrun VM. | Gateway spawns `openshell-driver-vm` as a subprocess over a private, state-local Unix socket. The VM driver boots a cached bootstrap `rootfs.ext4`, prepares requested OCI images inside a bootstrap VM with `umoci`, attaches the prepared image disk read-only, and gives each sandbox a writable `overlay.ext4` for merged-root changes and runtime material. The driver persists each accepted launch request beside the overlay and restarts those VMs on driver startup without recreating the overlay. |
-| External | Out-of-tree drivers operated alongside the gateway. | Whatever boundary the driver implements. | Activated by `--compute-driver-socket=<path>` (env `OPENSHELL_COMPUTE_DRIVER_SOCKET`). The gateway connects to a UDS the operator already provisioned, runs `GetCapabilities`, logs the advertised `driver_name`, and dispatches all sandbox lifecycle calls through the same `compute_driver.proto` surface as the in-tree drivers. The driver process and socket lifecycle are operator-owned; the gateway does not spawn, supervise, or remove the driver. The trust boundary is the socket's filesystem permissions — the operator must ensure only the gateway uid can read/write it. |
+| VM | Experimental microVM isolation. | Per-sandbox libkrun VM. | Managed endpoint-backed driver. The gateway spawns `openshell-driver-vm`, waits for its Unix socket, and then consumes it through the same remote `compute_driver.proto` path used by unmanaged endpoint drivers. The VM driver boots a cached bootstrap `rootfs.ext4`, prepares requested OCI images inside a bootstrap VM with `umoci`, attaches the prepared image disk read-only, and gives each sandbox a writable `overlay.ext4` for merged-root changes and runtime material. The driver persists each accepted launch request beside the overlay and restarts those VMs on driver startup without recreating the overlay. |
+| Extension | Out-of-tree drivers operated alongside the gateway. | Whatever boundary the driver implements. | Selected by a non-reserved custom `compute_drivers = ["<name>"]` entry with `[openshell.drivers.<name>].socket_path`, or by `--compute-driver-socket=<path>` as launch-time shorthand for the `extension` driver ID. Reserved built-in names such as `vm`, `docker`, `podman`, and `kubernetes` cannot be used as unmanaged socket endpoints. The gateway connects to a UDS the operator already provisioned, runs `GetCapabilities`, logs the advertised `driver_name`, and dispatches all sandbox lifecycle calls through `compute_driver.proto`. The driver process and socket lifecycle are operator-owned; the gateway does not spawn, supervise, or remove unmanaged extension drivers. The trust boundary is the socket's filesystem permissions: the operator must ensure only the gateway uid can read/write it. |
 
 Per-sandbox CPU and memory values currently enter the driver layer through
 template resource limits. Docker and Podman apply them as runtime limits.
@@ -79,7 +79,7 @@ The supervisor must be available inside each sandbox workload:
 | Podman | Read-only OCI image volume containing the supervisor binary. |
 | Kubernetes | Sandbox pod image or pod template configuration. |
 | VM | Embedded in the guest rootfs bundle. |
-| External | Defined by the out-of-tree driver. |
+| Extension | Defined by the out-of-tree driver. |
 
 Driver-controlled environment variables must override sandbox image or template
 values for sandbox ID, sandbox name, gateway endpoint, relay socket path, TLS

crates/openshell-core/src/config.rs

@@ -4,6 +4,7 @@
 //! Configuration management for `OpenShell` components.
 
 use serde::{Deserialize, Serialize};
+use std::collections::BTreeMap;
 use std::fmt;
 #[cfg(unix)]
 use std::io::{Read, Write};
@@ -69,6 +70,27 @@ impl ComputeDriverKind {
     }
 }
 
+/// Normalize a configured compute driver name.
+///
+/// Built-in driver names and custom remote driver names share the same
+/// selection namespace. The normalized value is lowercase ASCII and may contain
+/// letters, digits, `-`, and `_`.
+pub fn normalize_compute_driver_name(value: &str) -> Result<String, String> {
+    let value = value.trim();
+    if value.is_empty() {
+        return Err("compute driver name cannot be empty".to_string());
+    }
+    if !value
+        .bytes()
+        .all(|b| b.is_ascii_alphanumeric() || matches!(b, b'-' | b'_'))
+    {
+        return Err(format!(
+            "invalid compute driver name '{value}'. use ASCII letters, digits, '-' or '_'"
+        ));
+    }
+    Ok(value.to_ascii_lowercase())
+}
+
 impl fmt::Display for ComputeDriverKind {
     fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
         f.write_str(self.as_str())
@@ -358,13 +380,14 @@ pub struct Config {
     /// The config shape allows multiple drivers so the gateway can evolve
     /// toward multi-backend routing. Current releases require exactly one
     /// configured driver.
-    pub compute_drivers: Vec<ComputeDriverKind>,
+    pub compute_drivers: Vec<String>,
 
-    /// When set, the gateway dispatches sandbox lifecycle to an out-of-tree
-    /// compute driver process listening on this Unix domain socket and
-    /// speaking `compute_driver.proto`. Takes precedence over
-    /// `compute_drivers` and the auto-detection probe.
-    pub external_compute_driver_socket: Option<PathBuf>,
+    /// Operator-provided endpoints for named remote compute drivers.
+    ///
+    /// This is populated by CLI/env inputs such as `--compute-driver-socket`.
+    /// TOML-authored endpoints live under `[openshell.drivers.<name>]` and are
+    /// resolved by the gateway config loader.
+    pub compute_driver_endpoints: BTreeMap<String, PathBuf>,
 
     /// TTL for SSH session tokens, in seconds. 0 disables expiry.
     pub ssh_session_ttl_secs: u64,
@@ -565,7 +588,7 @@ impl Config {
             gateway_jwt: None,
             database_url: String::new(),
             compute_drivers: vec![],
-            external_compute_driver_socket: None,
+            compute_driver_endpoints: BTreeMap::new(),
             ssh_session_ttl_secs: default_ssh_session_ttl_secs(),
             grpc_rate_limit_requests: None,
             grpc_rate_limit_window_secs: None,
@@ -621,18 +644,27 @@ impl Config {
 
     /// Create a new configuration with the configured compute drivers.
     #[must_use]
-    pub fn with_compute_drivers<I>(mut self, drivers: I) -> Self
+    pub fn with_compute_drivers<I, D>(mut self, drivers: I) -> Self
     where
-        I: IntoIterator<Item = ComputeDriverKind>,
+        I: IntoIterator<Item = D>,
+        D: ToString,
     {
-        self.compute_drivers = drivers.into_iter().collect();
+        self.compute_drivers = drivers
+            .into_iter()
+            .map(|driver| driver.to_string())
+            .collect();
         self
     }
 
-    /// Pin an external compute driver by Unix domain socket path.
+    /// Register a Unix domain socket endpoint for a named remote driver.
     #[must_use]
-    pub fn with_external_compute_driver_socket(mut self, socket: Option<PathBuf>) -> Self {
-        self.external_compute_driver_socket = socket;
+    pub fn with_compute_driver_endpoint(
+        mut self,
+        name: impl Into<String>,
+        socket: impl Into<PathBuf>,
+    ) -> Self {
+        self.compute_driver_endpoints
+            .insert(name.into(), socket.into());
         self
     }
 
@@ -780,8 +812,8 @@ mod tests {
     use super::is_reachable_unix_socket;
     use super::{
         ComputeDriverKind, Config, DEFAULT_SERVICE_ROUTING_DOMAIN, GatewayJwtConfig, detect_driver,
-        docker_host_unix_socket_path, is_unix_socket, podman_socket_candidates_from_env,
-        podman_socket_responds,
+        docker_host_unix_socket_path, is_unix_socket, normalize_compute_driver_name,
+        podman_socket_candidates_from_env, podman_socket_responds,
     };
     #[cfg(unix)]
     use std::io::{Read as _, Write as _};
@@ -817,6 +849,18 @@ mod tests {
         assert!(err.contains("unsupported compute driver 'firecracker'"));
     }
 
+    #[test]
+    fn compute_driver_name_normalization_accepts_builtin_and_custom_names() {
+        assert_eq!(normalize_compute_driver_name(" VM ").unwrap(), "vm");
+        assert_eq!(
+            normalize_compute_driver_name("Kyma_GPU-1").unwrap(),
+            "kyma_gpu-1"
+        );
+
+        let err = normalize_compute_driver_name("kyma/gpu").unwrap_err();
+        assert!(err.contains("invalid compute driver name"));
+    }
+
     #[test]
     fn config_defaults_to_loopback_bind_address() {
         let expected: SocketAddr = "127.0.0.1:17670".parse().expect("valid address");

crates/openshell-server/src/cli.rs

@@ -109,15 +109,15 @@ struct RunArgs {
         value_delimiter = ',',
         value_parser = parse_compute_driver
     )]
-    drivers: Vec<ComputeDriverKind>,
+    drivers: Vec<String>,
 
-    /// Path to a Unix domain socket served by an out-of-tree compute driver
+    /// Path to a Unix domain socket served by a remote compute driver
     /// implementing `compute_driver.proto`.
     ///
-    /// When set, the gateway dispatches sandbox lifecycle to that driver
-    /// instead of one of the in-tree backends, skipping both the `--drivers`
-    /// list and the auto-detection probe. The driver name advertised in
-    /// `GetCapabilities` is logged for diagnostics.
+    /// When set, the socket is associated with the single configured driver
+    /// name. If no driver name is configured, the gateway uses `extension`.
+    /// Reserved built-in driver names such as Docker, Podman, Kubernetes, and
+    /// VM do not accept socket endpoints.
     #[arg(long, env = "OPENSHELL_COMPUTE_DRIVER_SOCKET")]
     compute_driver_socket: Option<PathBuf>,
 
@@ -245,6 +245,7 @@ async fn run_from_args(mut args: RunArgs, matches: ArgMatches) -> Result<()> {
     if let Some(file) = file.as_ref() {
         merge_file_into_args(&mut args, &file.openshell.gateway, &matches);
     }
+    normalize_compute_driver_socket_args(&mut args)?;
 
     let local_tls = apply_runtime_defaults(&mut args)?;
     let local_jwt = defaults::complete_local_jwt_config()?;
@@ -375,13 +376,19 @@ async fn run_from_args(mut args: RunArgs, matches: ArgMatches) -> Result<()> {
             args.grpc_rate_limit_requests,
             args.grpc_rate_limit_window_seconds,
         )
-        .with_external_compute_driver_socket(args.compute_driver_socket.clone())
         .with_server_sans(args.server_sans.clone())
         .with_loopback_service_http(args.enable_loopback_service_http);
     validate_grpc_rate_limit_args(
         args.grpc_rate_limit_requests,
         args.grpc_rate_limit_window_seconds,
     )?;
+    if let Some(socket) = args.compute_driver_socket.clone() {
+        let driver = args
+            .drivers
+            .first()
+            .expect("normalize_compute_driver_socket_args sets a driver for socket endpoints");
+        config = config.with_compute_driver_endpoint(driver.clone(), socket);
+    }
 
     if let Some(ttl) = file
         .as_ref()
@@ -468,8 +475,8 @@ async fn run_from_args(mut args: RunArgs, matches: ArgMatches) -> Result<()> {
     .into_diagnostic()
 }
 
-fn parse_compute_driver(value: &str) -> std::result::Result<ComputeDriverKind, String> {
-    value.parse()
+fn parse_compute_driver(value: &str) -> std::result::Result<String, String> {
+    openshell_core::config::normalize_compute_driver_name(value)
 }
 
 fn resolve_config_path(args: &RunArgs) -> Result<Option<PathBuf>> {
@@ -668,16 +675,51 @@ fn validate_grpc_rate_limit_args(requests: Option<u64>, window_seconds: Option<u
     Ok(())
 }
 
-fn effective_single_driver(args: &RunArgs) -> Option<ComputeDriverKind> {
-    // An external-driver socket pins dispatch to the out-of-tree path and
-    // bypasses both the `--drivers` list and auto-detection probe; callers
-    // that key off the in-tree `ComputeDriverKind` get `None` here.
-    if args.compute_driver_socket.is_some() {
-        return None;
+fn normalize_compute_driver_socket_args(args: &mut RunArgs) -> Result<()> {
+    let Some(socket) = args.compute_driver_socket.as_ref() else {
+        return Ok(());
+    };
+    if socket.as_os_str().is_empty() {
+        return Err(miette::miette!(
+            "--compute-driver-socket must not be an empty path"
+        ));
+    }
+
+    match args.drivers.as_slice() {
+        [] => {
+            args.drivers.push("extension".to_string());
+            Ok(())
+        }
+        [driver] => {
+            let driver = openshell_core::config::normalize_compute_driver_name(driver)
+                .map_err(|err| miette::miette!("{err}"))?;
+            if matches!(
+                driver.parse::<ComputeDriverKind>().ok(),
+                Some(
+                    ComputeDriverKind::Docker
+                        | ComputeDriverKind::Podman
+                        | ComputeDriverKind::Kubernetes
+                        | ComputeDriverKind::Vm
+                )
+            ) {
+                return Err(miette::miette!(
+                    "--compute-driver-socket cannot be combined with reserved built-in compute driver '{driver}'"
+                ));
+            }
+            args.drivers[0] = driver;
+            Ok(())
+        }
+        drivers => Err(miette::miette!(
+            "--compute-driver-socket requires exactly one compute driver name, got: {}",
+            drivers.join(",")
+        )),
     }
+}
+
+fn effective_single_driver(args: &RunArgs) -> Option<ComputeDriverKind> {
     match args.drivers.as_slice() {
         [] => openshell_core::config::detect_driver(),
-        [driver] => Some(*driver),
+        [driver] => driver.parse().ok(),
         _ => None,
     }
 }
@@ -1585,41 +1627,67 @@ ssh_session_ttl_secs = 1234
             .unwrap_or_else(std::sync::PoisonError::into_inner);
         let _g = EnvVarGuard::remove("OPENSHELL_COMPUTE_DRIVER_SOCKET");
 
-        let (args, _) = parse_with_args(&[
+        let (mut args, _) = parse_with_args(&[
             "openshell-gateway",
             "--db-url",
             "sqlite::memory:",
             "--compute-driver-socket",
-            "/run/openshell/external.sock",
+            "/run/openshell/extension.sock",
         ]);
+        super::normalize_compute_driver_socket_args(&mut args).unwrap();
         assert_eq!(
             args.compute_driver_socket.as_deref(),
-            Some(std::path::Path::new("/run/openshell/external.sock"))
+            Some(std::path::Path::new("/run/openshell/extension.sock"))
         );
-        // External socket pins dispatch off the in-tree enum, so the
-        // single-driver helper must return None even when no --drivers given.
+        assert_eq!(args.drivers, ["extension"]);
         assert!(super::effective_single_driver(&args).is_none());
     }
 
     #[test]
-    fn compute_driver_socket_overrides_drivers_flag() {
+    fn compute_driver_socket_rejects_reserved_builtin_drivers() {
         let _lock = ENV_LOCK
             .lock()
             .unwrap_or_else(std::sync::PoisonError::into_inner);
         let _g = EnvVarGuard::remove("OPENSHELL_COMPUTE_DRIVER_SOCKET");
 
-        let (args, _) = parse_with_args(&[
+        let (mut args, _) = parse_with_args(&[
             "openshell-gateway",
             "--db-url",
             "sqlite::memory:",
             "--drivers",
             "docker",
             "--compute-driver-socket",
-            "/run/openshell/external.sock",
+            "/run/openshell/extension.sock",
+        ]);
+        let err = super::normalize_compute_driver_socket_args(&mut args).unwrap_err();
+        assert!(
+            err.to_string()
+                .contains("cannot be combined with reserved built-in compute driver 'docker'"),
+            "unexpected error: {err}"
+        );
+    }
+
+    #[test]
+    fn compute_driver_socket_rejects_vm_endpoint() {
+        let _lock = ENV_LOCK
+            .lock()
+            .unwrap_or_else(std::sync::PoisonError::into_inner);
+        let _g = EnvVarGuard::remove("OPENSHELL_COMPUTE_DRIVER_SOCKET");
+
+        let (mut args, _) = parse_with_args(&[
+            "openshell-gateway",
+            "--db-url",
+            "sqlite::memory:",
+            "--drivers",
+            "vm",
+            "--compute-driver-socket",
+            "/run/openshell/vm.sock",
         ]);
+        let err = super::normalize_compute_driver_socket_args(&mut args).unwrap_err();
         assert!(
-            super::effective_single_driver(&args).is_none(),
-            "external socket must short-circuit --drivers"
+            err.to_string()
+                .contains("cannot be combined with reserved built-in compute driver 'vm'"),
+            "unexpected error: {err}"
         );
     }
 
@@ -1630,14 +1698,16 @@ ssh_session_ttl_secs = 1234
             .unwrap_or_else(std::sync::PoisonError::into_inner);
         let _g = EnvVarGuard::set(
             "OPENSHELL_COMPUTE_DRIVER_SOCKET",
-            "/var/run/openshell/external.sock",
+            "/var/run/openshell/extension.sock",
         );
 
-        let (args, _) = parse_with_args(&["openshell-gateway", "--db-url", "sqlite::memory:"]);
+        let (mut args, _) = parse_with_args(&["openshell-gateway", "--db-url", "sqlite::memory:"]);
+        super::normalize_compute_driver_socket_args(&mut args).unwrap();
         assert_eq!(
             args.compute_driver_socket.as_deref(),
-            Some(std::path::Path::new("/var/run/openshell/external.sock"))
+            Some(std::path::Path::new("/var/run/openshell/extension.sock"))
         );
+        assert_eq!(args.drivers, ["extension"]);
     }
 
     #[test]