fc-common: add failed paths cache infrastructure

Signed-off-by: NotAShelf <raf@notashelf.dev>
Change-Id: I35f9bfb044160151cf73c43ed9ada3476a6a6964

crates/common/migrations/016_failed_paths_cache.sql

@@ -0,0 +1,9 @@
+-- Failed paths cache: prevents rebuilding known-failing derivations
+CREATE TABLE failed_paths_cache (
+    drv_path TEXT PRIMARY KEY,
+    source_build_id UUID,
+    failure_status TEXT,
+    failed_at TIMESTAMP WITH TIME ZONE NOT NULL DEFAULT NOW()
+);
+
+CREATE INDEX idx_failed_paths_cache_failed_at ON failed_paths_cache(failed_at);

crates/common/src/config.rs

@@ -79,6 +79,14 @@ pub struct QueueRunnerConfig {
   /// retrying.
   #[serde(default)]
   pub strict_errors: bool,
+
+  /// Cache failed derivation paths to skip known-failing builds.
+  #[serde(default = "default_true")]
+  pub failed_paths_cache: bool,
+
+  /// TTL in seconds for failed paths cache entries (default 24h).
+  #[serde(default = "default_failed_paths_ttl")]
+  pub failed_paths_ttl: u64,
 }
 
 #[derive(Debug, Clone, Serialize, Deserialize)]
@@ -405,6 +413,10 @@ const fn default_true() -> bool {
   true
 }
 
+const fn default_failed_paths_ttl() -> u64 {
+  86400
+}
+
 const fn default_check_interval() -> i32 {
   60
 }
@@ -521,11 +533,13 @@ impl Default for EvaluatorConfig {
 impl Default for QueueRunnerConfig {
   fn default() -> Self {
     Self {
-      workers:       4,
-      poll_interval: 5,
-      build_timeout: 3600,
-      work_dir:      PathBuf::from("/tmp/fc-queue-runner"),
-      strict_errors: false,
+      workers:            4,
+      poll_interval:      5,
+      build_timeout:      3600,
+      work_dir:           PathBuf::from("/tmp/fc-queue-runner"),
+      strict_errors:      false,
+      failed_paths_cache: true,
+      failed_paths_ttl:   86400,
     }
   }
 }

crates/common/src/repo/builds.rs

@@ -276,13 +276,13 @@ pub async fn cancel_cascade(pool: &PgPool, id: Uuid) -> Result<Vec<Build>> {
 }
 
 /// Restart a build by resetting it to pending state.
-/// Only works for failed, succeeded, or cancelled builds.
+/// Only works for failed, succeeded, cancelled, or cached_failure builds.
 pub async fn restart(pool: &PgPool, id: Uuid) -> Result<Build> {
-  sqlx::query_as::<_, Build>(
+  let build = sqlx::query_as::<_, Build>(
     "UPDATE builds SET status = 'pending', started_at = NULL, completed_at = \
      NULL, log_path = NULL, build_output_path = NULL, error_message = NULL, \
      retry_count = retry_count + 1 WHERE id = $1 AND status IN ('failed', \
-     'succeeded', 'cancelled') RETURNING *",
+     'succeeded', 'cancelled', 'cached_failure') RETURNING *",
   )
   .bind(id)
   .fetch_optional(pool)
@@ -291,7 +291,15 @@ pub async fn restart(pool: &PgPool, id: Uuid) -> Result<Build> {
     CiError::NotFound(format!(
       "Build {id} not found or not in a restartable state"
     ))
-  })
+  })?;
+
+  if let Err(e) =
+    super::failed_paths_cache::invalidate(pool, &build.drv_path).await
+  {
+    tracing::warn!(build_id = %id, "Failed to invalidate failed paths cache: {e}");
+  }
+
+  Ok(build)
 }
 
 /// Mark a build's outputs as signed.

crates/common/src/repo/failed_paths_cache.rs

@@ -0,0 +1,64 @@
+use sqlx::PgPool;
+use uuid::Uuid;
+
+use crate::{
+  error::{CiError, Result},
+  models::BuildStatus,
+};
+
+pub async fn is_cached_failure(pool: &PgPool, drv_path: &str) -> Result<bool> {
+  let row: Option<(bool,)> =
+    sqlx::query_as("SELECT true FROM failed_paths_cache WHERE drv_path = $1")
+      .bind(drv_path)
+      .fetch_optional(pool)
+      .await
+      .map_err(CiError::Database)?;
+
+  Ok(row.is_some())
+}
+
+pub async fn insert(
+  pool: &PgPool,
+  drv_path: &str,
+  failure_status: BuildStatus,
+  source_build_id: Uuid,
+) -> Result<()> {
+  let status_str = failure_status.to_string();
+  sqlx::query(
+    "INSERT INTO failed_paths_cache (drv_path, source_build_id, \
+     failure_status, failed_at) VALUES ($1, $2, $3, NOW()) ON CONFLICT \
+     (drv_path) DO UPDATE SET source_build_id = $2, failure_status = $3, \
+     failed_at = NOW()",
+  )
+  .bind(drv_path)
+  .bind(source_build_id)
+  .bind(&status_str)
+  .execute(pool)
+  .await
+  .map_err(CiError::Database)?;
+
+  Ok(())
+}
+
+pub async fn invalidate(pool: &PgPool, drv_path: &str) -> Result<()> {
+  sqlx::query("DELETE FROM failed_paths_cache WHERE drv_path = $1")
+    .bind(drv_path)
+    .execute(pool)
+    .await
+    .map_err(CiError::Database)?;
+
+  Ok(())
+}
+
+pub async fn cleanup_expired(pool: &PgPool, ttl_seconds: u64) -> Result<u64> {
+  let result = sqlx::query(
+    "DELETE FROM failed_paths_cache WHERE failed_at < NOW() - \
+     make_interval(secs => $1)",
+  )
+  .bind(ttl_seconds as f64)
+  .execute(pool)
+  .await
+  .map_err(CiError::Database)?;
+
+  Ok(result.rows_affected())
+}

crates/common/src/repo/mod.rs

@@ -6,6 +6,7 @@ pub mod build_steps;
 pub mod builds;
 pub mod channels;
 pub mod evaluations;
+pub mod failed_paths_cache;
 pub mod jobset_inputs;
 pub mod jobsets;
 pub mod notification_configs;