feat(scanner): add bounded incarnation-scoped ACK receiver (#7182)

* chore(deps): refresh SDKs and pin clock skew regression coverage Refresh compatible dependencies for Scanner/Heal V2 batch 1 and verify the production S3 retry/signing path with a deterministic clock. Co-Authored-By: heihutu <heihutu@gmail.com> Co-Authored-By: zhi22915 <qiuzgang@gmail.com> * feat(scanner): add bounded incarnation-scoped ACK receiver Refs rustfs/backlog#2265 and rustfs/backlog#2240. Co-Authored-By: heihutu <heihutu@gmail.com> Co-Authored-By: zhi22915 <qiuzgang@gmail.com> --------- Co-authored-by: heihutu <heihutu@gmail.com> Co-authored-by: zhi22915 <qiuzgang@gmail.com>
2026-09-07 04:25:54 +00:00 · 2026-09-05 16:54:55 +08:00
parent 0d1b312673
commit acfeef55ab
17 changed files with 1097 additions and 24 deletions
@@ -196,15 +196,16 @@ pub mod bucket {
        pub use crate::bucket::metadata_sys::ConfigWriteLockProbe;
        pub use crate::bucket::metadata_sys::{
            BucketMetadataMutationGuard, BucketMetadataSys, ObjectLockConfigState, acquire_bucket_metadata_transaction_lock,
-            acquire_bucket_metadata_transaction_lock_for_incarnation, capture_bucket_metadata_incarnation, delete,
-            delete_if_incarnation, delete_under_transaction_lock, get, get_accelerate_config, get_bucket_policy,
-            get_bucket_policy_raw, get_bucket_targets_config, get_config_from_disk, get_cors_config, get_durability_config,
-            get_global_bucket_metadata_sys, get_lifecycle_config, get_logging_config, get_notification_config,
-            get_object_lock_config, get_object_lock_config_state, get_on_demand_migration_config, get_public_access_block_config,
-            get_quota_config, get_replication_config, get_request_payment_config, get_sse_config, get_tagging_config,
-            get_versioning_config, get_website_config, init_bucket_metadata_sys, list_bucket_targets, reload_bucket_metadata,
-            remove_bucket_metadata, set_bucket_metadata, update, update_bucket_targets_under_transaction_lock,
-            update_config_with, update_if_incarnation, update_quota_if_incarnation, update_under_transaction_lock,
+            acquire_bucket_metadata_transaction_lock_for_incarnation, acquire_scanner_bucket_incarnation_fence,
+            capture_bucket_metadata_incarnation, delete, delete_if_incarnation, delete_under_transaction_lock, get,
+            get_accelerate_config, get_bucket_policy, get_bucket_policy_raw, get_bucket_targets_config, get_config_from_disk,
+            get_cors_config, get_durability_config, get_global_bucket_metadata_sys, get_lifecycle_config, get_logging_config,
+            get_notification_config, get_object_lock_config, get_object_lock_config_state, get_on_demand_migration_config,
+            get_public_access_block_config, get_quota_config, get_replication_config, get_request_payment_config, get_sse_config,
+            get_tagging_config, get_versioning_config, get_website_config, init_bucket_metadata_sys, list_bucket_targets,
+            reload_bucket_metadata, remove_bucket_metadata, set_bucket_metadata, update,
+            update_bucket_targets_under_transaction_lock, update_config_with, update_if_incarnation, update_quota_if_incarnation,
+            update_under_transaction_lock,
        };
    }

@@ -655,6 +655,12 @@ pub struct BucketMetadataMutationGuard {
 }

 impl BucketMetadataMutationGuard {
+    /// Returns the storage-verified identity while both incarnation fences remain valid.
+    pub fn checked_bucket_incarnation(&self) -> Result<(&str, Uuid)> {
+        self.ensure_valid(&self.bucket)?;
+        Ok((&self.bucket, self.incarnation_id))
+    }
+
    fn ensure_valid(&self, bucket: &str) -> Result<()> {
        if self.bucket != bucket {
            return Err(Error::other("bucket metadata mutation guard does not match bucket"));
@@ -674,6 +680,29 @@ async fn acquire_config_write_guard_for_incarnation(
    sys: Arc<RwLock<BucketMetadataSys>>,
    bucket: &str,
    expected_incarnation_id: Option<Uuid>,
+) -> Result<BucketMetadataMutationGuard> {
+    acquire_config_write_guard_with_migration(sys, bucket, expected_incarnation_id, true).await
+}
+
+/// Scanner probes must not create an incarnation to make a capability available.
+pub async fn acquire_scanner_bucket_incarnation_fence(
+    bucket: &str,
+    expected_incarnation_id: Uuid,
+    expected_owner_id: Uuid,
+) -> Result<BucketMetadataMutationGuard> {
+    super::utils::check_valid_bucket_name(bucket)?;
+    let sys = get_bucket_metadata_sys()?;
+    if expected_owner_id.is_nil() || sys.read().await.api.id != expected_owner_id || expected_incarnation_id.is_nil() {
+        return Err(Error::other("scanner bucket incarnation owner does not match"));
+    }
+    acquire_config_write_guard_with_migration(sys, bucket, Some(expected_incarnation_id), false).await
+}
+
+async fn acquire_config_write_guard_with_migration(
+    sys: Arc<RwLock<BucketMetadataSys>>,
+    bucket: &str,
+    expected_incarnation_id: Option<Uuid>,
+    migrate: bool,
 ) -> Result<BucketMetadataMutationGuard> {
    let metadata_sys = sys.read().await.clone();
    let lifecycle_guard = metadata_sys.api.acquire_bucket_lifecycle_read_lock(bucket).await?;
@@ -681,13 +710,15 @@ async fn acquire_config_write_guard_for_incarnation(
    // Legacy buckets are migrated while the lifecycle fence prevents a
    // same-name replacement. The second read under the write transaction is
    // the CAS source of truth for the actual rewrite.
-    await_bucket_namespace_operation(
-        Some(&lifecycle_guard),
-        bucket,
-        "bucket config incarnation migration",
-        metadata_sys.get_bucket_incarnation_id(bucket),
-    )
-    .await?;
+    if migrate {
+        await_bucket_namespace_operation(
+            Some(&lifecycle_guard),
+            bucket,
+            "bucket config incarnation migration",
+            metadata_sys.get_bucket_incarnation_id(bucket),
+        )
+        .await?;
+    }
    let transaction_guard = await_bucket_namespace_operation(
        Some(&lifecycle_guard),
        bucket,
@@ -3176,6 +3207,82 @@ mod tests {
        );
    }

+    #[tokio::test]
+    async fn scoped_dirty_usage_incarnation_probe_does_not_migrate_legacy_metadata() {
+        let (dirs, store) = isolated_store_over_temp_disks().await;
+        let sys = Arc::new(RwLock::new(BucketMetadataSys::new(store.clone())));
+        let bucket = "scoped-ack-legacy";
+        for dir in &dirs {
+            std::fs::create_dir_all(dir.path().join(bucket)).expect("create legacy bucket");
+        }
+        let mut metadata = BucketMetadata::new(bucket);
+        metadata.bucket_incarnation_id = Uuid::nil();
+        sys.read()
+            .await
+            .persist_and_set(metadata)
+            .await
+            .expect("persist legacy metadata");
+        assert!(
+            acquire_config_write_guard_with_migration(sys.clone(), bucket, Some(Uuid::new_v4()), false)
+                .await
+                .is_err()
+        );
+        assert!(load_bucket_incarnation(store, bucket).await.expect("read sidecar").is_none());
+        assert!(
+            sys.read()
+                .await
+                .get_config_from_disk(bucket)
+                .await
+                .expect("read metadata")
+                .bucket_incarnation_id
+                .is_nil()
+        );
+    }
+
+    #[tokio::test(flavor = "multi_thread", worker_threads = 4)]
+    #[serial]
+    async fn scoped_dirty_usage_incarnation_rejects_deleted_and_recreated_bucket() {
+        let (_dirs, store) = isolated_store_over_temp_disks().await;
+        init_bucket_metadata_sys(store.clone(), Vec::new()).await;
+        let sys = bucket_metadata_sys_of(&store.ctx).expect("metadata owner");
+        let bucket = "scoped-ack-recreated";
+        store
+            .make_bucket(bucket, &MakeBucketOptions::default())
+            .await
+            .expect("create bucket");
+        let old = store.bucket_incarnation_id_from_disk(bucket).await.expect("old incarnation");
+        let guard = acquire_config_write_guard_with_migration(sys.clone(), bucket, Some(old), false)
+            .await
+            .expect("trusted incarnation fence");
+        assert_eq!(guard.checked_bucket_incarnation().expect("valid fences"), (bucket, old));
+        drop(guard);
+        store
+            .delete_bucket(bucket, &DeleteBucketOptions::default())
+            .await
+            .expect("delete bucket");
+        assert!(
+            acquire_config_write_guard_with_migration(sys.clone(), bucket, Some(old), false)
+                .await
+                .is_err()
+        );
+        store
+            .make_bucket(bucket, &MakeBucketOptions::default())
+            .await
+            .expect("recreate bucket");
+        let new = store.bucket_incarnation_id_from_disk(bucket).await.expect("new incarnation");
+        assert_ne!(old, new);
+        assert!(
+            acquire_config_write_guard_with_migration(sys.clone(), bucket, Some(old), false)
+                .await
+                .is_err()
+        );
+        assert!(
+            acquire_config_write_guard_with_migration(sys, bucket, Some(new), false)
+                .await
+                .is_ok()
+        );
+    }
+
    #[tokio::test]
    async fn old_node_metadata_rewrite_cannot_replace_bucket_incarnation_sidecar() {
        let (dirs, ecstore) = isolated_store_over_temp_disks().await;
@@ -30,6 +30,7 @@ use rustfs_protos::{
    ChannelClass, create_new_channel, get_channel_for_class,
    proto_gen::node_service::{
        heal_control_service_client::HealControlServiceClient, node_service_client::NodeServiceClient,
+        scanner_control_service_client::ScannerControlServiceClient,
        tier_mutation_control_service_client::TierMutationControlServiceClient,
    },
 };
@@ -60,6 +61,24 @@ pub async fn node_service_time_out_client(
    node_service_time_out_client_for_class(addr, interceptor, ChannelClass::Control).await
 }

+pub(crate) async fn scanner_control_time_out_client(
+    addr: &str,
+    interceptor: TonicInterceptor,
+) -> crate::error::Result<ScannerControlServiceClient<InterceptedService<AuthenticatedChannel, TonicInterceptor>>> {
+    let interceptor = interceptor.with_rpc_audience(addr)?;
+    let channel = match runtime_sources::cached_node_channel(addr).await {
+        Some(channel) => channel,
+        None => create_new_channel(addr)
+            .await
+            .map_err(|err| crate::error::Error::other(err.to_string()))?,
+    };
+    let channel = ReplayScopeChannel::new(channel, interceptor.replay_scope_audience());
+    let limit = rustfs_protos::scoped_dirty_usage::SCOPED_DIRTY_USAGE_MAX_REQUEST_BYTES as usize;
+    Ok(ScannerControlServiceClient::with_interceptor(channel, interceptor)
+        .max_decoding_message_size(limit)
+        .max_encoding_message_size(limit))
+}
+
 pub async fn heal_control_time_out_client(
    addr: &str,
    interceptor: TonicInterceptor,
@@ -2050,6 +2050,53 @@ impl PeerRestClient {
        .await
    }

+    /// Probe only: scoped ACK production requires a durable per-bucket proof.
+    pub async fn scanner_scoped_dirty_usage_capability(
+        &self,
+        owner_id: String,
+        instance_id: String,
+        entries: Vec<rustfs_protos::proto_gen::node_service::ScannerScopedDirtyUsageEntry>,
+    ) -> Result<bool> {
+        use rustfs_protos::scoped_dirty_usage::*;
+        let payload = rustfs_protos::proto_gen::node_service::ScannerScopedDirtyUsageAckRequest {
+            challenge: Uuid::new_v4().as_bytes().to_vec().into(),
+            protocol_version: SCOPED_DIRTY_USAGE_PROTOCOL_VERSION,
+            owner_id,
+            instance_id,
+            scope: SCOPED_DIRTY_USAGE_BUCKET_SCOPE,
+            probe_only: true,
+            entries,
+        };
+        let canonical = canonical_scoped_dirty_usage_request(&payload).map_err(|err| Error::other(err.to_string()))?;
+        self.finalize_result(
+            async {
+                let mut client = super::client::scanner_control_time_out_client(
+                    &self.grid_host,
+                    TonicInterceptor::Signature(gen_tonic_signature_interceptor()),
+                )
+                .await?;
+                let mut request = Request::new(payload.clone());
+                set_tonic_canonical_body_digest(&mut request, &canonical)?;
+                let response = client.scanner_scoped_dirty_usage_ack(request).await?.into_inner();
+                let body = canonical_scoped_dirty_usage_response(&canonical, &response)
+                    .map_err(|_| Error::other("scoped dirty usage capability response is too large"))?;
+                verify_tonic_rpc_response_proof(&body, response.response_proof.as_ref())?;
+                if response.protocol_version != SCOPED_DIRTY_USAGE_PROTOCOL_VERSION
+                    || response.owner_id != payload.owner_id
+                    || response.instance_id != payload.instance_id
+                    || response.max_entries != SCOPED_DIRTY_USAGE_MAX_ENTRIES
+                    || response.max_request_bytes != SCOPED_DIRTY_USAGE_MAX_REQUEST_BYTES
+                    || response.cleared != 0
+                {
+                    return Err(Error::other("scoped dirty usage capability response does not match request"));
+                }
+                Ok(response.supported)
+            }
+            .await,
+        )
+        .await
+    }
+
    pub async fn acknowledge_scanner_dirty_usage(&self, instance_id: String, generation: u64) -> Result<ScannerPeerActivity> {
        let result = self
            .scanner_activity_request_with_protocol(instance_id.clone(), generation, SCANNER_ACTIVITY_PROTOCOL_VERSION)