huskies: merge 1098 bug Shadow drift: set_retry_count / bump_retry_count write CRDT register without updating pipeline_items.retry_count

server/src/db/mod.rs

@@ -592,6 +592,42 @@ mod tests {
         );
     }
 
+    /// `shadow_write::init` spawns its background task on the calling runtime,
+    /// which under `#[tokio::test]` is per-test and dies when the test ends.
+    /// Park the init on a leaked multi-thread runtime so the bg task lives for
+    /// the whole test process; mirrors `db::ops::tests::ensure_shadow_db`.
+    #[cfg(test)]
+    static SHADOW_RT: std::sync::OnceLock<tokio::runtime::Runtime> = std::sync::OnceLock::new();
+
+    #[cfg(test)]
+    async fn ensure_shadow_db() {
+        static INIT: std::sync::OnceLock<()> = std::sync::OnceLock::new();
+        if INIT.get().is_some() {
+            return;
+        }
+        let rt = SHADOW_RT.get_or_init(|| {
+            tokio::runtime::Builder::new_multi_thread()
+                .worker_threads(1)
+                .enable_all()
+                .build()
+                .expect("shadow rt")
+        });
+        rt.spawn(async {
+            static INNER: std::sync::OnceLock<()> = std::sync::OnceLock::new();
+            if INNER.get().is_some() {
+                return;
+            }
+            let tmp = tempfile::tempdir().expect("tmp");
+            let db_path = tmp.path().join("pipeline.db");
+            std::mem::forget(tmp);
+            shadow_write::init(&db_path).await.expect("shadow init");
+            let _ = INNER.set(());
+        })
+        .await
+        .expect("shadow init task");
+        let _ = INIT.set(());
+    }
+
     /// Regression for story 1095: `set_name` must propagate the new name to the
     /// SQLite shadow table via `sync_item_name`.  Before the fix, the CRDT
     /// register was updated but `pipeline_items.name` stayed stale.
@@ -599,10 +635,7 @@ mod tests {
     async fn set_name_updates_shadow_name_column() {
         crate::crdt_state::init_for_test();
         ensure_content_store();
-
-        let tmp = tempfile::tempdir().unwrap();
-        let db_path = tmp.path().join("pipeline.db");
-        shadow_write::init(&db_path).await.expect("db init");
+        ensure_shadow_db().await;
 
         let story_id = "9095_story_set_name_shadow";
         write_item_with_content(
@@ -612,17 +645,29 @@ mod tests {
             ItemMeta::named("Original Name"),
         );
 
+        // Wait for the initial insert to land.
+        tokio::time::sleep(std::time::Duration::from_millis(50)).await;
+
         // Rename via the CRDT setter — now also triggers sync_item_name.
         crate::crdt_state::set_name(story_id, Some("Updated Name"));
 
         // Wait for the background write task to flush.
         tokio::time::sleep(std::time::Duration::from_millis(100)).await;
 
-        let pool = shadow_write::get_shared_pool().expect("pool must be initialised");
+        // Open a fresh pool on this test's runtime — sqlx pools are not safe
+        // to share across runtimes, so we can't reuse `get_shared_pool()`
+        // (which was created on the leaked shadow-write runtime).
+        let path = shadow_write::SHADOW_DB_PATH
+            .get()
+            .expect("SHADOW_DB_PATH set by init");
+        let opts = sqlx::sqlite::SqliteConnectOptions::new()
+            .filename(path)
+            .create_if_missing(false);
+        let pool = sqlx::SqlitePool::connect_with(opts).await.unwrap();
         let row: (Option<String>,) =
             sqlx::query_as("SELECT name FROM pipeline_items WHERE id = ?1")
                 .bind(story_id)
-                .fetch_one(pool)
+                .fetch_one(&pool)
                 .await
                 .unwrap();
 
@@ -633,6 +678,58 @@ mod tests {
         );
     }
 
+    /// Bug 1098: `bump_retry_count` must mirror the new value to the SQLite
+    /// shadow table, not only to the CRDT register.
+    ///
+    /// Before the fix, calling `bump_retry_count` updated the CRDT but left
+    /// `pipeline_items.retry_count` stale.
+    #[tokio::test]
+    async fn bump_retry_count_updates_shadow_table() {
+        crate::crdt_state::init_for_test();
+        ensure_content_store();
+        ensure_shadow_db().await;
+
+        let story_id = "9899_story_retry_shadow_1098";
+
+        // Insert the story into both CRDT and the shadow table.
+        write_item_with_content(
+            story_id,
+            "2_current",
+            "# Retry shadow test\n",
+            ItemMeta::named("Retry Shadow Test"),
+        );
+
+        // Let the background write task process the initial insert.
+        tokio::time::sleep(std::time::Duration::from_millis(50)).await;
+
+        // Three bumps → retry_count must reach 3 in SQLite.
+        crate::crdt_state::bump_retry_count(story_id);
+        crate::crdt_state::bump_retry_count(story_id);
+        crate::crdt_state::bump_retry_count(story_id);
+
+        // Let the background write task process all three updates.
+        tokio::time::sleep(std::time::Duration::from_millis(100)).await;
+
+        let path = shadow_write::SHADOW_DB_PATH
+            .get()
+            .expect("SHADOW_DB_PATH set by init");
+        let opts = sqlx::sqlite::SqliteConnectOptions::new()
+            .filename(path)
+            .create_if_missing(false);
+        let pool = sqlx::SqlitePool::connect_with(opts).await.unwrap();
+        let (count,): (i64,) =
+            sqlx::query_as("SELECT retry_count FROM pipeline_items WHERE id = ?1")
+                .bind(story_id)
+                .fetch_one(&pool)
+                .await
+                .unwrap();
+
+        assert_eq!(
+            count, 3,
+            "retry_count must be 3 after three bump_retry_count calls"
+        );
+    }
+
     /// Story 1087, AC2: the split-stage migration projects every supported
     /// wire-form `stage` string into the canonical `(pipeline, status)` pair.
     /// The fixture covers each Stage variant (and the legacy numeric-prefix