fix(runtime-core): name a node job's macOS boot by its session UUID, not calendar boot time

packages/runtime-core/src/node/journal.rs

@@ -1124,4 +1124,39 @@ mod tests {
         drop(inspect);
         std::fs::remove_dir_all(root).unwrap();
     }
+
+    #[test]
+    fn a_rollback_journal_deleted_between_check_and_open_reads_as_absent() {
+        use super::super::test_support::BeforeOpen;
+        let root = test_root("journal-race");
+        drop(Store::open(&root, true).unwrap());
+        let journal = root.join("journal.sqlite-journal");
+        let rollback_journal = || {
+            std::fs::write(&journal, b"").unwrap();
+            std::fs::set_permissions(
+                &journal,
+                std::os::unix::fs::PermissionsExt::from_mode(0o600),
+            )
+            .unwrap();
+        };
+        // A running supervisor's commit deletes its rollback journal after any earlier
+        // check saw it and just before the restarted node's open (HACK-1418).
+        rollback_journal();
+        let race = BeforeOpen::arm(&journal, |path| std::fs::remove_file(path).unwrap());
+        let store = Store::open(&root, false).unwrap();
+        assert!(race.fired());
+        assert!(journal.symlink_metadata().is_err());
+        drop((store, race));
+        // The same instant cannot swap in an alias: the O_NOFOLLOW open still refuses.
+        rollback_journal();
+        let outside = root.join("outside");
+        let race = BeforeOpen::arm(&journal, move |path| {
+            std::fs::remove_file(path).unwrap();
+            std::os::unix::fs::symlink(&outside, path).unwrap();
+        });
+        assert!(Store::open(&root, false).is_err());
+        assert!(race.fired());
+        drop(race);
+        std::fs::remove_dir_all(root).unwrap();
+    }
 }

packages/runtime-core/src/node/mod.rs

@@ -66,6 +66,8 @@ pub(crate) fn existing_private_file(path: &Path) -> Result<Option<File>> {
 }
 
 fn open_private_file(path: &Path, create: bool) -> std::io::Result<File> {
+    #[cfg(test)]
+    test_support::before_open(path);
     OpenOptions::new()
         .read(true)
         .write(true)
@@ -103,6 +105,85 @@ pub(crate) fn try_lock(path: &Path) -> Result<Option<File>> {
     }
 }
 
+#[cfg(test)]
+pub(crate) mod test_support {
+    use super::{Receipt, Store};
+    use std::cell::RefCell;
+    use std::path::{Path, PathBuf};
+
+    type Hook = (PathBuf, Box<dyn FnOnce(&Path)>);
+    thread_local! {
+        static BEFORE_OPEN: RefCell<Option<Hook>> = const { RefCell::new(None) };
+    }
+
+    /// Runs this thread's one-shot hook when the node opens `path`, just before the
+    /// `open` system call: the instant another writer (SQLite committing) can act.
+    pub(crate) fn before_open(path: &Path) {
+        let hook = BEFORE_OPEN.with(|cell| {
+            let mut slot = cell.borrow_mut();
+            if slot.as_ref().is_some_and(|(target, _)| target == path) {
+                slot.take()
+            } else {
+                None
+            }
+        });
+        if let Some((target, hook)) = hook {
+            hook(&target);
+        }
+    }
+
+    /// Arms a one-shot hook for `path` on this thread; dropping it disarms the hook.
+    pub(crate) struct BeforeOpen;
+    impl BeforeOpen {
+        pub(crate) fn arm(path: &Path, hook: impl FnOnce(&Path) + 'static) -> Self {
+            BEFORE_OPEN.with(|cell| cell.replace(Some((path.to_owned(), Box::new(hook)))));
+            Self
+        }
+        pub(crate) fn fired(&self) -> bool {
+            BEFORE_OPEN.with(|cell| cell.borrow().is_none())
+        }
+    }
+    impl Drop for BeforeOpen {
+        fn drop(&mut self) {
+            BEFORE_OPEN.with(|cell| cell.replace(None));
+        }
+    }
+
+    /// Writes one non-terminal job receipt directly, bypassing admission, so supervisor
+    /// and recovery tests can start from an exact journal state.
+    pub(crate) fn job(store: &Store, id: &str, state: &str, timeout_ms: u64) -> Receipt {
+        let receipt = Receipt {
+            job_id: id.into(),
+            fixture: "tree".into(),
+            source: None,
+            state: state.into(),
+            accepted_at_ms: super::now(),
+            queue_deadline_ms: super::now() + 60_000,
+            execution_timeout_ms: timeout_ms,
+            starts: 0,
+            supervisor_pid: None,
+            child_pid: None,
+            supervisor_identity: None,
+            child_identity: None,
+            cancel_requested: false,
+            stdout: String::new(),
+            stderr: String::new(),
+            truncated: false,
+            exit_code: None,
+            signal: None,
+            detail: None,
+        };
+        store
+            .connection
+            .execute(
+                "INSERT INTO jobs VALUES (?,?)",
+                rusqlite::params![id, serde_json::to_string(&receipt).unwrap()],
+            )
+            .unwrap();
+        receipt
+    }
+}
+
 #[cfg(test)]
 mod tests {
     use super::*;