Plumb transceiver error through API as a string

Author: bnaecker

dpd/src/transceivers/mod.rs

@@ -94,11 +94,12 @@ impl QsfpDevice {
 }
 
 /// The cause of a fault on a transceiver.
-#[derive(Clone, Copy, Debug, Deserialize, JsonSchema, Serialize)]
+#[derive(Clone, Debug, Deserialize, JsonSchema, Serialize)]
 #[serde(rename_all = "snake_case")]
 pub enum FaultReason {
-    /// An error occurred accessing the transceiver.
-    Failed,
+    /// An error occurred accessing the transceiver, with details about the
+    /// failure.
+    Failed { details: String },
     /// Power was enabled, but did not come up in the requisite time.
     PowerTimeout,
     /// Power was enabled and later lost.

dpd/src/transceivers/tofino_impl.rs

@@ -1109,8 +1109,16 @@ impl Switch {
                     port.as_qsfp_mut().unwrap().transceiver =
                         Some(Transceiver::Faulted(FaultReason::DisabledBySp));
                 } else if status.failures.modules.is_set(index).unwrap() {
+                    let details = status
+                        .failures
+                        .errors
+                        .get(usize::from(index))
+                        .map(|err| err.to_string())
+                        .unwrap_or_else(|| String::from("Unknown error"));
                     port.as_qsfp_mut().unwrap().transceiver =
-                        Some(Transceiver::Faulted(FaultReason::Failed));
+                        Some(Transceiver::Faulted(FaultReason::Failed {
+                            details,
+                        }));
                 } else {
                     // At this point, the transceiver is present, and so we
                     // _can_ check it for support. To do so requires that it be
@@ -1201,7 +1209,13 @@ impl Switch {
                 let new_transceiver = if unsupported.is_set(index).unwrap() {
                     Transceiver::Unsupported
                 } else if new_modules.failed.is_set(index).unwrap() {
-                    Transceiver::Faulted(FaultReason::Failed)
+                    let details = status
+                        .failures
+                        .errors
+                        .get(usize::from(index))
+                        .map(|err| err.to_string())
+                        .unwrap_or_else(|| String::from("Unknown error"));
+                    Transceiver::Faulted(FaultReason::Failed { details })
                 } else {
                     continue;
                 };
@@ -1327,11 +1341,28 @@ impl Switch {
                         transceiver.in_reset = None;
                         transceiver.interrupt_pending = None;
                     }
-                } else if power.failures.modules.is_set(index).unwrap()
-                    || vendor_info.failures.modules.is_set(index).unwrap()
-                {
+                } else if power.failures.modules.is_set(index).unwrap() {
+                    let details = power
+                        .failures
+                        .errors
+                        .get(usize::from(index))
+                        .map(|err| err.to_string())
+                        .unwrap_or_else(|| String::from("Unknown error"));
+                    port_lock.lock().await.as_qsfp_mut().unwrap().transceiver =
+                        Some(Transceiver::Faulted(FaultReason::Failed {
+                            details,
+                        }));
+                } else if vendor_info.failures.modules.is_set(index).unwrap() {
+                    let details = vendor_info
+                        .failures
+                        .errors
+                        .get(usize::from(index))
+                        .map(|err| err.to_string())
+                        .unwrap_or_else(|| String::from("Unknown error"));
                     port_lock.lock().await.as_qsfp_mut().unwrap().transceiver =
-                        Some(Transceiver::Faulted(FaultReason::Failed));
+                        Some(Transceiver::Faulted(FaultReason::Failed {
+                            details,
+                        }));
                 }
             }
 

openapi/dpd.json

@@ -5594,11 +5594,25 @@
         "description": "The cause of a fault on a transceiver.",
         "oneOf": [
           {
-            "description": "An error occurred accessing the transceiver.",
-            "type": "string",
-            "enum": [
+            "description": "An error occurred accessing the transceiver, with details about the failure.",
+            "type": "object",
+            "properties": {
+              "failed": {
+                "type": "object",
+                "properties": {
+                  "details": {
+                    "type": "string"
+                  }
+                },
+                "required": [
+                  "details"
+                ]
+              }
+            },
+            "required": [
               "failed"
-            ]
+            ],
+            "additionalProperties": false
           },
           {
             "description": "Power was enabled, but did not come up in the requisite time.",

swadm/src/switchport.rs

@@ -364,7 +364,13 @@ fn print_faulted_transceiver_row(
     port_id: &PortId,
     reason: &types::FaultReason,
 ) -> anyhow::Result<()> {
-    writeln!(tw, "{}\tfaulted ({:?})\t\t\t\t\t\t\t", port_id, reason)
+    let reason = match reason {
+        types::FaultReason::Failed { details } => details.as_str(),
+        types::FaultReason::PowerTimeout => "Power timeout",
+        types::FaultReason::PowerLost => "Power lost",
+        types::FaultReason::DisabledBySp => "Disabled by SP",
+    };
+    writeln!(tw, "{}\tfaulted ({})\t\t\t\t\t\t\t", port_id, reason)
         .map_err(|e| e.into())
 }