From 6869d92926f3782507940ba64ec91c9b3257d233 Mon Sep 17 00:00:00 2001
From: Sean Klein <sean@oxide.computer>
Date: Thu, 27 Jun 2024 17:51:15 -0700
Subject: [PATCH] Zone bundler

---
 sled-agent/src/long_running_tasks.rs |  2 +-
 sled-agent/src/sled_agent.rs         |  5 ++++-
 sled-agent/src/zone_bundle.rs        | 15 +++++++++++++++
 3 files changed, 20 insertions(+), 2 deletions(-)

diff --git a/sled-agent/src/long_running_tasks.rs b/sled-agent/src/long_running_tasks.rs
index e42c26591f..e920ffc3fc 100644
--- a/sled-agent/src/long_running_tasks.rs
+++ b/sled-agent/src/long_running_tasks.rs
@@ -46,7 +46,7 @@ pub struct LongRunningTaskHandles {
     /// for establishing zpools on disks and managing their datasets.
     pub storage_manager: StorageHandle,
 
-    /// A mehcanism for talking to the [`StorageMonitor`], which reacts to disk
+    /// A mechanism for talking to the [`StorageMonitor`], which reacts to disk
     /// changes and updates the dump devices.
     pub storage_monitor_handle: StorageMonitorHandle,
 
diff --git a/sled-agent/src/sled_agent.rs b/sled-agent/src/sled_agent.rs
index 4e2efa5657..e596dbe049 100644
--- a/sled-agent/src/sled_agent.rs
+++ b/sled-agent/src/sled_agent.rs
@@ -836,7 +836,10 @@ impl SledAgent {
         // to start using new disks and stop using old ones.
         self.inner.storage_monitor.await_generation(*our_gen).await?;
 
-        // - TODO: Update Zone bundles?
+        // Ensure that the ZoneBundler, if it was creating a bundle referencing
+        // the old U.2s, has stopped using them.
+        self.inner.zone_bundler.await_completion_of_prior_bundles().await;
+
         // - TODO: Mark probes failed?
         // - TODO: Mark instances failed?
 
diff --git a/sled-agent/src/zone_bundle.rs b/sled-agent/src/zone_bundle.rs
index 16147e5957..cd9f36fe3a 100644
--- a/sled-agent/src/zone_bundle.rs
+++ b/sled-agent/src/zone_bundle.rs
@@ -256,6 +256,9 @@ impl Inner {
     // exist; and returns those.
     async fn bundle_directories(&self) -> Vec<Utf8PathBuf> {
         let resources = self.storage_handle.get_latest_disks().await;
+        // NOTE: These bundle directories are always stored on M.2s, so we don't
+        // need to worry about synchronizing with U.2 disk expungement at the
+        // callsite.
         let expected = resources.all_zone_bundle_directories();
         let mut out = Vec::with_capacity(expected.len());
         for each in expected.into_iter() {
@@ -426,6 +429,10 @@ impl ZoneBundler {
         zone: &RunningZone,
         cause: ZoneBundleCause,
     ) -> Result<ZoneBundleMetadata, BundleError> {
+        // NOTE: [Self::await_completion_of_prior_bundles] relies on this lock
+        // being held across this whole function. If we want more concurrency,
+        // we'll need to add a barrier-like mechanism to let callers know when
+        // prior bundles have completed.
         let inner = self.inner.lock().await;
         let storage_dirs = inner.bundle_directories().await;
         let resources = inner.storage_handle.get_latest_disks().await;
@@ -443,6 +450,14 @@ impl ZoneBundler {
         create(&self.log, zone, &context).await
     }
 
+    /// Awaits the completion of all prior calls to [ZoneBundler::create].
+    ///
+    /// This is critical for disk expungement, which wants to ensure that the
+    /// Sled Agent is no longer using devices after they have been expunged.
+    pub async fn await_completion_of_prior_bundles(&self) {
+        let _ = self.inner.lock().await;
+    }
+
     /// Return the paths for all bundles of the provided zone and ID.
     pub async fn bundle_paths(
         &self,