Allow setting a fault attribute for attributing whether guest or host to blame

Signed-off-by: Ryan Levick <ryan.levick@fermyon.com>

Author: rylev

crates/factor-key-value/src/host.rs

@@ -2,7 +2,7 @@ use super::{Cas, SwapError};
 use anyhow::{Context, Result};
 use spin_core::{async_trait, wasmtime::component::Resource};
 use spin_resource_table::Table;
-use spin_telemetry::OpenTelemetrySpanExt as _;
+use spin_telemetry::traces::{self, Fault};
 use spin_world::v2::key_value;
 use spin_world::wasi::keyvalue as wasi_keyvalue;
 use std::{collections::HashSet, sync::Arc};
@@ -72,10 +72,7 @@ impl KeyValueDispatch {
     pub fn get_store<T: 'static>(&self, store: Resource<T>) -> anyhow::Result<&Arc<dyn Store>> {
         let res = self.stores.get(store.rep()).context("invalid store");
         if let Err(err) = &res {
-            let current_span = tracing::Span::current();
-            current_span.set_status(spin_telemetry::opentelemetry::trace::Status::error(
-                err.to_string(),
-            ));
+            traces::mark_as_error(err, Some(Fault::Host));
         }
         res
     }
@@ -190,12 +187,11 @@ impl key_value::HostStore for KeyValueDispatch {
 
 /// Make sure that infrastructure related errors are tracked in the current span.
 fn track_error_on_span(err: Error) -> Error {
-    if let Error::Other(_) | Error::StoreTableFull = &err {
-        let current_span = tracing::Span::current();
-        current_span.set_status(spin_telemetry::opentelemetry::trace::Status::error(
-            err.to_string(),
-        ));
-    }
+    let fault = match err {
+        Error::NoSuchStore | Error::AccessDenied => Fault::Guest,
+        Error::StoreTableFull | Error::Other(_) => Fault::Host,
+    };
+    traces::mark_as_error(&err, Some(fault));
     err
 }
 

crates/factor-variables/src/host.rs

@@ -1,5 +1,5 @@
 use spin_factors::anyhow;
-use spin_telemetry::OpenTelemetrySpanExt as _;
+use spin_telemetry::traces::{self, Fault};
 use spin_world::{v1, v2::variables, wasi::config as wasi_config};
 use tracing::instrument;
 
@@ -73,19 +73,18 @@ impl wasi_config::store::Host for InstanceState {
     }
 }
 
+/// Convert a `spin_expressions::Error` to a `variables::Error`, setting the current span's status and fault attribute.
 fn expressions_to_variables_err(err: spin_expressions::Error) -> variables::Error {
     use spin_expressions::Error;
+    let fault = match err {
+        Error::InvalidName(_) | Error::InvalidTemplate(_) | Error::Undefined(_) => Fault::Guest,
+        Error::Provider(_) => Fault::Host,
+    };
+    traces::mark_as_error(&err, Some(fault));
     match err {
         Error::InvalidName(msg) => variables::Error::InvalidName(msg),
         Error::Undefined(msg) => variables::Error::Undefined(msg),
-        other @ Error::InvalidTemplate(_) => variables::Error::Other(format!("{other}")),
-        Error::Provider(err) => {
-            // This error may not be caused by bad user input, so set the span status to error.
-            let current_span = tracing::Span::current();
-            current_span.set_status(spin_telemetry::opentelemetry::trace::Status::error(
-                err.to_string(),
-            ));
-            variables::Error::Provider(err.to_string())
-        }
+        Error::InvalidTemplate(_) => variables::Error::Other(format!("{err}")),
+        Error::Provider(err) => variables::Error::Provider(err.to_string()),
     }
 }

crates/telemetry/src/lib.rs

@@ -13,10 +13,7 @@ mod env;
 pub mod logs;
 pub mod metrics;
 mod propagation;
-mod traces;
-
-pub use opentelemetry;
-pub use tracing_opentelemetry::OpenTelemetrySpanExt;
+pub mod traces;
 
 #[cfg(feature = "testing")]
 pub mod testing;

crates/telemetry/src/traces.rs

@@ -5,6 +5,7 @@ use opentelemetry_sdk::{
     Resource,
 };
 use tracing::Subscriber;
+use tracing_opentelemetry::OpenTelemetrySpanExt as _;
 use tracing_subscriber::{registry::LookupSpan, EnvFilter, Layer};
 
 use crate::detector::SpinResourceDetector;
@@ -61,3 +62,32 @@ pub(crate) fn otel_tracing_layer<S: Subscriber + for<'span> LookupSpan<'span>>(
         .with_threads(false)
         .with_filter(env_filter))
 }
+
+/// Indicate whether the error is more likely caused by the guest or the host.
+///
+/// This can be used to filter errors in telemetry or logging systems.
+#[derive(Debug)]
+pub enum Fault {
+    /// The error is most likely caused by the guest.
+    Guest,
+    /// The error might have been caused by the host.
+    Host,
+}
+
+impl Fault {
+    fn as_str(&self) -> &'static str {
+        match self {
+            Fault::Guest => "guest",
+            Fault::Host => "host",
+        }
+    }
+}
+
+/// Marks the current span as an error with the given error message and optional fault type.
+pub fn mark_as_error<E: std::fmt::Display>(err: &E, fault: Option<Fault>) {
+    let current_span = tracing::Span::current();
+    current_span.set_status(opentelemetry::trace::Status::error(err.to_string()));
+    if let Some(fault) = fault {
+        current_span.set_attribute("error.fault", fault.as_str());
+    }
+}