Gossip persistence

Cargo.toml

@@ -60,8 +60,8 @@ signature = "2"
 sled-hardware-types = { git = "https://github.com/oxidecomputer/omicron", tag = "rel/v21/rc1" }
 slog = "2"
 slog-term = "2"
-sprockets-tls = { git = "https://github.com/oxidecomputer/sprockets.git", rev = "b8782742d1bd91546887fbc1757a7d024a5469ef", default-features = false }
-sprockets-tls-test-utils = { git = "https://github.com/oxidecomputer/sprockets.git", rev = "b8782742d1bd91546887fbc1757a7d024a5469ef" }
+sprockets-tls = { git = "https://github.com/oxidecomputer/sprockets.git", rev = "98615b86d51f74ce220198eb16c57932fd24e7cb", default-features = false }
+sprockets-tls-test-utils = { git = "https://github.com/oxidecomputer/sprockets.git", rev = "98615b86d51f74ce220198eb16c57932fd24e7cb" }
 ssh-key = { version = "0.6", features = ["ed25519", "p256", "serde"] }
 tempfile = "3"
 thiserror = "1"

client/src/commands.rs

@@ -872,10 +872,29 @@ async fn authenticate<E>(
     if let Some(via) = via {
         request = request.via(via.to_string());
     }
-    let identity = request.send().await?.into_inner();
+    let identity = match request.send().await {
+        Ok(identity) => identity.into_inner(),
+        Err(error) if error.status() == Some(StatusCode::UNAUTHORIZED) => {
+            let server = client.version().send().await.ok().map(|v| v.into_inner());
+            return Err(login_refused(server, error.into()));
+        }
+        Err(error) => return Err(error.into()),
+    };
     Ok((identity, Authz::new(credentials, key)))
 }
 
+/// A refused login usually indicates build skew rather than a bad key.
+/// Blame the builds when they differ, otherwise return the server error.
+fn login_refused(server: Option<VersionInfo>, error: CommandError) -> CommandError {
+    let client = VersionInfo::current();
+    match server {
+        Some(server) if server.commit != client.commit => {
+            CommandError::AuthnBuildMismatch { client, server }
+        }
+        _ => error,
+    }
+}
+
 /// Make a request as someone who is logged in, logging in if needed.
 /// The client's pre-send hook signs each attempt.
 async fn with_login<T, E, Req>(
@@ -2357,6 +2376,15 @@ pub enum CommandError {
     AmbiguousSerial(String),
     #[error("❌ Authentication error")]
     Authn(#[from] AuthnError),
+    #[error(
+        "❌ Authentication failed, client and server are different builds\n   \
+            Client: {client}\n   \
+            Server: {server}"
+    )]
+    AuthnBuildMismatch {
+        client: VersionInfo,
+        server: VersionInfo,
+    },
     #[error("❌ Canceled")]
     Canceled,
     #[error("❌ Certificate for `{0}` is outside its validity window")]
@@ -2554,3 +2582,29 @@ impl From<ClientError<Upgraded>> for CommandError {
         }
     }
 }
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[test]
+    fn login_refused_blames_build_skew() {
+        let refused = || CommandError::Client(String::from("Authentication required, try `iam`"));
+        let skewed = VersionInfo {
+            version: String::from("0.1.0"),
+            commit: String::from("not-this-build"),
+        };
+        assert!(matches!(
+            login_refused(Some(skewed), refused()),
+            CommandError::AuthnBuildMismatch { .. }
+        ));
+        assert!(matches!(
+            login_refused(Some(VersionInfo::current()), refused()),
+            CommandError::Client(_)
+        ));
+        assert!(matches!(
+            login_refused(None, refused()),
+            CommandError::Client(_)
+        ));
+    }
+}

common/src/jobs.rs

@@ -422,6 +422,8 @@ pub enum ProcessError {
     OutputLimitExceeded { stream: JobOutputStream, limit: u64 },
     #[error("Unable to join job process: {0}")]
     Join(String),
+    #[error("The server restarted while the job was running")]
+    Interrupted,
 }
 
 impl ProcessError {

server/Cargo.toml

@@ -16,6 +16,7 @@ embedded = []
 test-support = []
 
 [dependencies]
+atomicwrites.workspace = true
 bytes.workspace = true
 camino.workspace = true
 bytesize.workspace = true