starknet_transaction_prover: proving-job duration + outcome metrics

Adds Prometheus counters / histograms recorded by `VirtualSnosProver` for
each proving job: total count by outcome (`success`, `validation_error`,
`internal_error`, `l1_provider_error`) and end-to-end duration.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Author: avi-starkware

crates/starknet_transaction_prover/src/errors.rs

@@ -131,6 +131,27 @@ pub enum VirtualSnosProverError {
     TransactionBlocked,
 }
 
+impl VirtualSnosProverError {
+    /// Maps the variant to one of the bounded label values declared in
+    /// `crate::server::metrics::outcomes`. The single match keeps the
+    /// `prover_prove_transaction_outcome_total{outcome}` cardinality fixed —
+    /// adding a variant requires a dashboard update at the same time.
+    pub fn metric_outcome(&self) -> &'static str {
+        use crate::server::metrics::outcomes;
+        match self {
+            VirtualSnosProverError::InvalidTransactionType(_)
+            | VirtualSnosProverError::InvalidTransactionInput(_)
+            | VirtualSnosProverError::ValidationError(_) => outcomes::VALIDATION,
+            VirtualSnosProverError::TransactionBlocked => outcomes::BLOCKED,
+            VirtualSnosProverError::RunnerError(_) => outcomes::RUNNER,
+            VirtualSnosProverError::OutputParseError(_)
+            | VirtualSnosProverError::ProgramOutputError(_) => outcomes::OUTPUT_PARSE,
+            #[cfg(feature = "stwo_proving")]
+            VirtualSnosProverError::ProvingError(_) => outcomes::PROVING,
+        }
+    }
+}
+
 /// Errors that can occur during configuration.
 #[derive(Debug, Error)]
 pub enum ConfigError {

crates/starknet_transaction_prover/src/proving/virtual_snos_prover.rs

@@ -25,6 +25,7 @@ use crate::blocking_check::{BlockingCheckClient, BlockingCheckResult};
 use crate::config::ProverConfig;
 use crate::errors::VirtualSnosProverError;
 use crate::running::runner::{RpcRunnerFactory, RunnerOutput, VirtualSnosRunner};
+use crate::server::metrics::{names as metric_names, outcomes};
 
 /// Result of a successful prove transaction operation.
 ///
@@ -149,7 +150,28 @@ impl<R: VirtualSnosRunner + 'static> VirtualSnosProver<R> {
         transaction: RpcTransaction,
     ) -> Result<ProveTransactionResult, VirtualSnosProverError> {
         let start_time = Instant::now();
+        let result = self.prove_transaction_inner(block_id, transaction).await;
+        let total = start_time.elapsed();
+        // One histogram observation per request. Latency is bucketed and is
+        // recorded regardless of outcome so SLO calculations (e.g. p99 over
+        // success-only) can be done in the query layer.
+        metrics::histogram!(metric_names::PROVE_TRANSACTION_DURATION_SECONDS)
+            .record(total.as_secs_f64());
+        let outcome = match &result {
+            Ok(_) => outcomes::SUCCESS,
+            Err(err) => err.metric_outcome(),
+        };
+        metrics::counter!(metric_names::PROVE_TRANSACTION_OUTCOME_TOTAL, "outcome" => outcome)
+            .increment(1);
+        info!(total_duration_ms = %total.as_millis(), outcome, "prove_transaction completed");
+        result
+    }
 
+    async fn prove_transaction_inner(
+        &self,
+        block_id: BlockId,
+        transaction: RpcTransaction,
+    ) -> Result<ProveTransactionResult, VirtualSnosProverError> {
         // Validate block_id is not pending.
         if matches!(block_id, BlockId::Pending) {
             return Err(VirtualSnosProverError::ValidationError(
@@ -162,15 +184,12 @@ impl<R: VirtualSnosRunner + 'static> VirtualSnosProver<R> {
         validate_transaction_input(&invoke_v3, self.validate_zero_fee_fields)?;
         let invoke_tx = InvokeTransaction::V3(invoke_v3.into());
 
-        let result = match &self.blocking_check_client {
-            None => self.run_and_prove(block_id, vec![invoke_tx]).await?,
+        match &self.blocking_check_client {
+            None => self.run_and_prove(block_id, vec![invoke_tx]).await,
             Some(client) => {
-                self.prove_with_blocking_check(client, block_id, transaction, invoke_tx).await?
+                self.prove_with_blocking_check(client, block_id, transaction, invoke_tx).await
             }
-        };
-
-        info!(total_duration_ms = %start_time.elapsed().as_millis(), "prove_transaction completed");
-        Ok(result)
+        }
     }
 
     /// Runs the OS and generates a proof. This is the core proving pipeline.
@@ -186,18 +205,18 @@ impl<R: VirtualSnosRunner + 'static> VirtualSnosProver<R> {
             .await
             .map_err(|err| VirtualSnosProverError::RunnerError(Box::new(err)))?;
 
-        info!(
-            os_duration_ms = %os_start.elapsed().as_millis(),
-            "OS execution completed"
-        );
+        let os_duration = os_start.elapsed();
+        metrics::histogram!(metric_names::OS_RUN_DURATION_SECONDS)
+            .record(os_duration.as_secs_f64());
+        info!(os_duration_ms = %os_duration.as_millis(), "OS execution completed");
 
         let prove_start = Instant::now();
         let result = self.prove_virtual_snos_run(runner_output).await?;
 
-        info!(
-            prove_duration_ms = %prove_start.elapsed().as_millis(),
-            "Proving completed"
-        );
+        let prove_duration = prove_start.elapsed();
+        metrics::histogram!(metric_names::STWO_PROVE_DURATION_SECONDS)
+            .record(prove_duration.as_secs_f64());
+        info!(prove_duration_ms = %prove_duration.as_millis(), "Proving completed");
 
         Ok(result)
     }

crates/starknet_transaction_prover/src/server/metrics.rs

@@ -28,6 +28,28 @@ pub mod names {
     pub const BUILD_INFO: &str = "prover_build_info";
     /// Requests rejected because the concurrency semaphore was full.
     pub const CONCURRENCY_REJECTED_TOTAL: &str = "prover_concurrency_rejected_total";
+    /// Wall-clock duration of `prove_transaction` end-to-end. Bucketed.
+    pub const PROVE_TRANSACTION_DURATION_SECONDS: &str =
+        "prover_prove_transaction_duration_seconds";
+    /// `prove_transaction` outcomes by category. See [`super::outcomes`] for
+    /// the fixed set of label values.
+    pub const PROVE_TRANSACTION_OUTCOME_TOTAL: &str = "prover_prove_transaction_outcome_total";
+    /// Virtual SNOS run sub-step duration. Bucketed.
+    pub const OS_RUN_DURATION_SECONDS: &str = "prover_os_run_duration_seconds";
+    /// Stwo proving sub-step duration. Bucketed.
+    pub const STWO_PROVE_DURATION_SECONDS: &str = "prover_stwo_prove_duration_seconds";
+}
+
+/// Fixed, bounded set of values for the `outcome` label on
+/// [`names::PROVE_TRANSACTION_OUTCOME_TOTAL`]. Adding a variant requires a
+/// dashboard update.
+pub mod outcomes {
+    pub const SUCCESS: &str = "success";
+    pub const VALIDATION: &str = "failure_validation";
+    pub const BLOCKED: &str = "failure_blocked";
+    pub const RUNNER: &str = "failure_runner";
+    pub const OUTPUT_PARSE: &str = "failure_output_parse";
+    pub const PROVING: &str = "failure_proving";
 }
 
 /// Initializes the global Prometheus exporter and emits the `build_info`