fix(heal): cleanup consumed MRF replay journals

Do not retain Accepted or Merged replay intents as startup anchors after they have been handed to the heal manager. Only refused or still-pending replay records keep the journal on disk until a successor snapshot can persist them.

This keeps successor snapshots limited to the pending queue, which lets successful replay remove both authoritative and legacy journal paths and restores the crash-boundary tests around successor flush.

Co-Authored-By: heihutu <heihutu@gmail.com>

Co-Authored-By: zhi22915 <qiuzgang@gmail.com>
(cherry picked from commit d5b8f49c9d)

crates/scanner/src/scanner_io.rs

@@ -250,7 +250,8 @@ fn resolve_remote_dirty_usage_scope(
     // Peer snapshots contribute bucket names only; the local prefix scopes
     // would narrow a bucket a peer dirtied elsewhere, so the merged scope
     // stays at bucket granularity (same rule as the local fallthrough).
-    let scope = scoped_scan_scope_from_dirty_buckets(requested_scope, dirty_buckets, None, true, all_buckets, baseline_proof);
+    let scope =
+        scoped_scan_scope_from_dirty_buckets(requested_scope, dirty_buckets, None, true, false, all_buckets, baseline_proof);
     if scope.is_default() {
         return default_result(scope);
     }
@@ -362,6 +363,7 @@ fn scoped_scan_scope_from_dirty_buckets(
     dirty_buckets: HashSet<String>,
     dirty_scopes: Option<&DirtyUsageBucketScopes>,
     dirty_snapshot_complete: bool,
+    segment_reuse_activated: bool,
     all_buckets: &[BucketInfo],
     baseline_proof: ScannerCacheBaselineProof<'_>,
 ) -> ScannerBucketScanScope {
@@ -382,24 +384,34 @@ fn scoped_scan_scope_from_dirty_buckets(
         return requested_scope;
     };
 
-    let selected_bucket_prefixes = dirty_scopes
-        .into_iter()
-        .flat_map(|dirty_scopes| {
-            selected_buckets
-                .iter()
-                .filter_map(|bucket| dirty_scopes.get(bucket).map(|scope| (bucket.clone(), scope)))
-        })
-        .filter_map(|(bucket, scope)| match scope {
-            DirtyUsageBucketScope::WholeBucket => None,
-            DirtyUsageBucketScope::TopLevelEntries(entries) => {
-                ScannerBucketPrefixScanScope::from_dirty_top_level_entries(entries.clone()).map(|scope| (bucket, scope))
-            }
-        })
-        .collect();
+    let selected_bucket_prefixes = if segment_reuse_activated {
+        dirty_scopes
+            .into_iter()
+            .flat_map(|dirty_scopes| {
+                selected_buckets
+                    .iter()
+                    .filter_map(|bucket| dirty_scopes.get(bucket).map(|scope| (bucket.clone(), scope)))
+            })
+            .filter_map(|(bucket, scope)| match scope {
+                DirtyUsageBucketScope::WholeBucket => None,
+                DirtyUsageBucketScope::TopLevelEntries(entries) => {
+                    ScannerBucketPrefixScanScope::from_dirty_top_level_entries(entries.clone()).map(|scope| (bucket, scope))
+                }
+            })
+            .collect()
+    } else {
+        HashMap::new()
+    };
 
     ScannerBucketScanScope::from_dirty_buckets(selected_buckets, selected_bucket_prefixes, baseline_scan_plan_digest)
 }
 
+fn scanner_segment_reuse_activated() -> bool {
+    // Production segment reuse stays disabled until a durable mutation-stream
+    // proof satisfies the segment invalidation contract.
+    false
+}
+
 pub(crate) fn is_scanner_metadata_corrupt_error(err: &StorageError) -> bool {
     matches!(err, StorageError::Io(io) if io.to_string().starts_with(SCANNER_METADATA_CORRUPT_ERROR))
 }

crates/scanner/src/scanner_io/io_cycle.rs

@@ -245,6 +245,7 @@ where
         dirty_buckets,
         (!distributed).then_some(resolution.dirty_usage_snapshot.scopes.as_ref()),
         true,
+        scanner_segment_reuse_activated(),
         resolution.all_buckets,
         resolution.baseline_proof,
     ))

crates/scanner/src/scanner_io/tests.rs

@@ -373,9 +373,13 @@ async fn scoped_scan_production_entry_preserves_deep_and_full_maintenance_work()
                 .expect("maintenance object should persist");
             wait_for_namespace_commit_tails(store.as_ref()).await;
             // Only the hot bucket is in the dirty-usage hint. The ordinary
-            // dirty cycle exercises scoped reuse; the following maintenance
-            // cycles mutate cold storage and must still walk it.
-            record_dirty_usage_bucket("hot-bucket");
+            // dirty cycle exercises bucket-scoped reuse; object-level segment
+            // hints remain activation-gated.
+            if index == 1 {
+                record_dirty_usage_object("hot-bucket", &format!("added-{index}"));
+            } else {
+                record_dirty_usage_bucket("hot-bucket");
+            }
         }
         let requested_scope = if explicit_scope {
             ScannerBucketScanScope::from_dirty_buckets(
@@ -422,6 +426,10 @@ async fn scoped_scan_production_entry_preserves_deep_and_full_maintenance_work()
                 Some(&HashSet::from(["hot-bucket".to_string()])),
                 "ordinary dirty work must retain the existing planner"
             );
+            assert!(
+                resolved.prefix_scope_for("hot-bucket").is_none(),
+                "production segment reuse must remain disabled before activation"
+            );
         } else {
             assert!(resolved.is_default(), "cycle {cycle} must visit the full maintenance scope");
         }
@@ -1566,6 +1574,7 @@ fn scoped_scan_selects_only_current_dirty_buckets_after_baseline_validation() {
         HashSet::from(["photos".to_string(), "deleted".to_string()]),
         None,
         true,
+        false,
         &[bucket_info("photos")],
         ScannerCacheBaselineProof {
             authoritative_data: Some(&baseline),
@@ -1620,7 +1629,7 @@ fn scoped_scan_baseline_work_proof_requires_uniform_known_set_identity() {
 }
 
 #[test]
-fn scoped_scan_uses_only_locally_verified_prefix_hints() {
+fn scoped_scan_prefix_hints_require_segment_reuse_activation() {
     let source = DataUsageCacheSource::new(1, 2);
     let expected_sources = HashSet::from([source]);
     let scan_plan_digest = DataUsageScanPlanDigest([6; 32]);
@@ -1638,6 +1647,7 @@ fn scoped_scan_uses_only_locally_verified_prefix_hints() {
         HashSet::from(["photos".to_string(), "videos".to_string()]),
         Some(&dirty_scopes),
         true,
+        false,
         &[bucket_info("photos"), bucket_info("videos")],
         ScannerCacheBaselineProof {
             authoritative_data: Some(&baseline),
@@ -1648,14 +1658,41 @@ fn scoped_scan_uses_only_locally_verified_prefix_hints() {
             scan_plan_digest,
         },
     );
-    assert!(locally_scoped.prefix_scope_for("photos").is_some());
+    assert_eq!(
+        locally_scoped.selected_buckets.as_deref(),
+        Some(&HashSet::from(["photos".to_string(), "videos".to_string()]))
+    );
+    assert!(
+        locally_scoped.prefix_scope_for("photos").is_none(),
+        "production must not consume segment hints before activation"
+    );
     assert!(locally_scoped.prefix_scope_for("videos").is_none());
 
+    let activated = scoped_scan_scope_from_dirty_buckets(
+        ScannerBucketScanScope::default(),
+        HashSet::from(["photos".to_string(), "videos".to_string()]),
+        Some(&dirty_scopes),
+        true,
+        true,
+        &[bucket_info("photos"), bucket_info("videos")],
+        ScannerCacheBaselineProof {
+            authoritative_data: Some(&baseline),
+            observed_candidate_data: None,
+            expected_sources: &expected_sources,
+            leader_epoch: 11,
+            want_cycle: 8,
+            scan_plan_digest,
+        },
+    );
+    assert!(activated.prefix_scope_for("photos").is_some());
+    assert!(activated.prefix_scope_for("videos").is_none());
+
     let distributed_scope = scoped_scan_scope_from_dirty_buckets(
         ScannerBucketScanScope::default(),
         HashSet::from(["photos".to_string(), "videos".to_string()]),
         None,
         true,
+        true,
         &[bucket_info("photos"), bucket_info("videos")],
         ScannerCacheBaselineProof {
             authoritative_data: Some(&baseline),
@@ -1700,6 +1737,7 @@ fn remote_dirty_usage_invalidates_local_prefix_hints_until_distributed_proof_exi
         HashSet::from(["photos".to_string()]),
         Some(&dirty_scopes),
         true,
+        true,
         &[bucket_info("photos")],
         ScannerCacheBaselineProof {
             authoritative_data: Some(&baseline),

docs/operations/kms-admin-contract.md

@@ -39,7 +39,7 @@ The wire prefix is `/rustfs/admin/v3`. `GET /kms/status` and `GET /kms/service-s
 | `POST /kms/restore/dry-run` | `kms:Restore` | sensitive | no | Preflight; writes nothing |
 | `POST /kms/restore` | `kms:Restore` | high | no | Requires `confirm_backup_id` and `confirm_conflict_policy` |
 | `POST /kms/restore/abort` | `kms:Restore` | high | no | Requires `confirm_target_key_dir` |
-| `POST /kms/create-key`, `POST /kms/key/create` | `kms:Configure` | high | no | Legacy `mc` aliases of `POST /kms/keys`; the key name comes from the `key-id` query parameter (`mc`'s form) or the `name` tag, and a request carrying both with different values is refused with `400` |
+| `POST /kms/create-key`, `POST /kms/key/create` | `kms:Configure` | high | no | Legacy `mc` aliases of `POST /kms/keys` |
 | `GET /kms/describe-key`, `GET /kms/key/status` | `kms:DescribeKey` | sensitive | yes | Legacy aliases of `GET /kms/keys/{key_id}` |
 | `GET /kms/list-keys` | `kms:ListKeys` | sensitive | no | Legacy alias of `GET /kms/keys`; same listing contract |
 

rustfs/src/admin/handlers/kms_keys.rs

@@ -18,7 +18,6 @@ use super::kms_audit::{KmsAdminAudit, KmsAdminOperation};
 use crate::admin::auth::{validate_admin_request, validate_admin_request_with_kms_key};
 use crate::admin::router::{AdminOperation, Operation, S3Router};
 use crate::admin::runtime_sources::{current_kms_runtime_service_manager, current_or_init_kms_runtime_service_manager};
-use crate::admin::storage_api::s3;
 use crate::admin::utils::extract_query_params;
 use crate::auth::{check_key_valid, get_session_token};
 use crate::kms_deletion_gate::current_key_impact;
@@ -198,25 +197,6 @@ fn extract_key_id(uri: &hyper::Uri) -> Option<String> {
         .find_map(|name| query_params.get(name).filter(|value| !value.is_empty()).cloned())
 }
 
-/// Name of the key a legacy create request asks for.
-///
-/// `mc admin kms key create <name>` sends the name as the `key-id` query
-/// parameter with no body, while RustFS clients send it as the `name` tag.
-/// Both are honored. A request carrying both has to agree with itself:
-/// picking one silently would create a key under a name the caller never
-/// sees in its own request.
-fn legacy_create_key_name(uri: &hyper::Uri, tags: &HashMap<String, String>) -> S3Result<Option<String>> {
-    let query_name = extract_key_id(uri);
-    let tag_name = tags.get("name").cloned();
-    match (query_name, tag_name) {
-        (Some(query), Some(tag)) if query != tag => Err(s3::error(
-            s3::S3ErrorCode::InvalidRequest,
-            format!("key name in the query ({query}) and in tags.name ({tag}) differ"),
-        )),
-        (query, tag) => Ok(query.or(tag)),
-    }
-}
-
 /// The `key_id` of a KMS admin request body, read without committing to the
 /// strict schema of the endpoint: the authorization gate needs the target key
 /// before the body is parsed for execution, and a body that fails the strict
@@ -352,8 +332,9 @@ impl Operation for CreateKeyHandler {
             return Err(s3_error!(InternalError, "kms service is not initialized"));
         };
 
+        // Extract key name from tags if provided
         let tags = request.tags.unwrap_or_default();
-        let key_name = legacy_create_key_name(&req.uri, &tags)?;
+        let key_name = tags.get("name").cloned();
 
         let kms_request = CreateKeyRequest {
             key_name,
@@ -498,8 +479,8 @@ mod tests {
         DescribeKmsKeyResponse, GenerateDataKeyApiRequest, GenerateDataKeyApiResponse, ListKeysApiResponse, ListKmsKeysResponse,
         delete_key_error_status, delete_request_from_query, extract_key_id, extract_query_params, key_impact_if_requested,
         key_list_filters, kms_create_key_actions, kms_delete_key_actions, kms_describe_key_actions,
-        kms_generate_data_key_actions, kms_list_keys_actions, legacy_create_key_name, parse_list_limit, scoped_key_id,
-        stable_json_value, wants_key_impact,
+        kms_generate_data_key_actions, kms_list_keys_actions, parse_list_limit, scoped_key_id, stable_json_value,
+        wants_key_impact,
     };
     use http::Uri;
     use hyper::StatusCode;
@@ -520,46 +501,6 @@ mod tests {
         assert!(!actions.contains(&action), "expected action list not to contain {action:?}");
     }
 
-    #[test]
-    fn legacy_create_key_name_honors_the_minio_key_id_query() {
-        let uri: Uri = "/rustfs/admin/v3/kms/key/create?key-id=minio-key"
-            .parse()
-            .expect("uri should parse");
-
-        let name = legacy_create_key_name(&uri, &HashMap::new()).expect("a query-only name is valid");
-        assert_eq!(name.as_deref(), Some("minio-key"));
-    }
-
-    #[test]
-    fn legacy_create_key_name_falls_back_to_the_name_tag() {
-        let uri: Uri = "/rustfs/admin/v3/kms/key/create".parse().expect("uri should parse");
-        let tags = HashMap::from([("name".to_string(), "tagged-key".to_string())]);
-
-        let name = legacy_create_key_name(&uri, &tags).expect("a tag-only name is valid");
-        assert_eq!(name.as_deref(), Some("tagged-key"));
-        assert_eq!(legacy_create_key_name(&uri, &HashMap::new()).expect("no name is valid"), None);
-    }
-
-    #[test]
-    fn legacy_create_key_name_accepts_agreeing_sources_and_refuses_conflicting_ones() {
-        let uri: Uri = "/rustfs/admin/v3/kms/key/create?key-id=minio-key"
-            .parse()
-            .expect("uri should parse");
-
-        let agreeing = HashMap::from([("name".to_string(), "minio-key".to_string())]);
-        let name = legacy_create_key_name(&uri, &agreeing).expect("agreeing sources are valid");
-        assert_eq!(name.as_deref(), Some("minio-key"));
-
-        let conflicting = HashMap::from([("name".to_string(), "other-key".to_string())]);
-        let refused = legacy_create_key_name(&uri, &conflicting).expect_err("conflicting names must be refused");
-        assert_eq!(*refused.code(), super::s3::S3ErrorCode::InvalidRequest);
-        assert!(
-            refused
-                .message()
-                .is_some_and(|message| message.contains("minio-key") && message.contains("other-key"))
-        );
-    }
-
     #[test]
     fn test_extract_key_id_supports_minio_aliases() {
         for (uri, expected) in [

rustfs/src/error.rs

@@ -21,23 +21,6 @@ use s3s::{S3Error, S3ErrorCode};
 const MAX_VERSIONS_EXCEEDED_CODE: &str = "MaxVersionsExceeded";
 const MAX_VERSIONS_EXCEEDED_MESSAGE: &str = "You've exceeded the limit on the number of versions you can create on this object";
 
-/// S3 error code for a request that names a KMS key the KMS does not hold.
-pub const KMS_KEY_NOT_FOUND_ERROR_CODE: &str = "KMS.NotFoundException";
-
-/// HTTP status of the error codes s3s cannot derive on its own.
-///
-/// s3s answers `None` for every `Custom` code, which the response layer turns
-/// into a 500; a code that means "your request named something that does not
-/// exist" has to say so itself.
-fn custom_error_status(code: &S3ErrorCode) -> Option<StatusCode> {
-    match code {
-        S3ErrorCode::Custom(custom) if &**custom == KMS_KEY_NOT_FOUND_ERROR_CODE || &**custom == MAX_VERSIONS_EXCEEDED_CODE => {
-            Some(StatusCode::BAD_REQUEST)
-        }
-        _ => None,
-    }
-}
-
 /// Marks a request body that exceeded a presigned upload size capability.
 ///
 /// This marker must survive the body-reader and storage layers so the client
@@ -385,10 +368,9 @@ fn error_chain_s3s_body_stream_error(err: &(dyn std::error::Error + 'static)) ->
 
 impl From<ApiError> for S3Error {
     fn from(err: ApiError) -> Self {
-        let status = custom_error_status(&err.code);
         let mut s3e = S3Error::with_message(err.code, err.message);
-        if let Some(status) = status {
-            s3e.set_status_code(status);
+        if matches!(s3e.code(), S3ErrorCode::Custom(code) if &**code == MAX_VERSIONS_EXCEEDED_CODE) {
+            s3e.set_status_code(StatusCode::BAD_REQUEST);
         }
         if let Some(source) = err.source {
             s3e.set_source(source);
@@ -460,19 +442,6 @@ impl From<StorageError> for ApiError {
                     source: Some(Box::new(err)),
                 };
             }
-
-            // A request header or bucket default naming a key the KMS does not
-            // hold is the caller's mistake to correct, and S3 reports it as
-            // 400 `KMS.NotFoundException`. Left to the fallthrough it became a
-            // 500 whose generic message hid which key was missing.
-            if let Some(rustfs_kms::KmsError::KeyNotFound { key_id }) = inner.downcast_ref::<rustfs_kms::KmsError>() {
-                let message = format!("KMS key not found: {key_id}");
-                return ApiError {
-                    code: S3ErrorCode::Custom(KMS_KEY_NOT_FOUND_ERROR_CODE.into()),
-                    message,
-                    source: Some(Box::new(err)),
-                };
-            }
         }
 
         let code = match &err {
@@ -1011,25 +980,6 @@ mod tests {
         }
     }
 
-    #[test]
-    fn test_kms_key_not_found_maps_to_bad_request_kms_not_found_exception() {
-        let api_error = ApiError::from(StorageError::other(rustfs_kms::KmsError::key_not_found("no-such-key")));
-
-        assert_eq!(api_error.code, S3ErrorCode::Custom(KMS_KEY_NOT_FOUND_ERROR_CODE.into()));
-        assert_eq!(api_error.message, "KMS key not found: no-such-key");
-
-        // s3s knows no status for a custom code; the conversion has to supply it.
-        let s3_error = S3Error::from(api_error);
-        assert_eq!(s3_error.status_code(), Some(StatusCode::BAD_REQUEST));
-    }
-
-    #[test]
-    fn test_generated_error_codes_keep_their_own_status() {
-        let s3_error = S3Error::from(ApiError::from(StorageError::other(rustfs_kms::KmsError::backend_error("down"))));
-
-        assert_eq!(s3_error.status_code(), Some(StatusCode::SERVICE_UNAVAILABLE));
-    }
-
     #[test]
     fn test_unknown_authoritative_quota_usage_maps_to_retryable_error() {
         let api_error = ApiError::from(QuotaError::UsageUnavailable {

rustfs/src/storage/sse.rs

@@ -4336,12 +4336,9 @@ mod tests {
     fn kms_operation_errors_preserve_retryability_classification() {
         let unavailable = kms_operation_error(rustfs_kms::KmsError::backend_error("connection refused"));
         let corrupt = kms_operation_error(rustfs_kms::KmsError::cryptographic_error("decrypt", "authentication failed"));
-        let missing = kms_operation_error(rustfs_kms::KmsError::key_not_found("no-such-key"));
 
         assert_eq!(unavailable.code, S3ErrorCode::ServiceUnavailable);
         assert_eq!(corrupt.code, S3ErrorCode::InternalError);
-        assert_eq!(missing.code, S3ErrorCode::Custom(crate::error::KMS_KEY_NOT_FOUND_ERROR_CODE.into()));
-        assert_eq!(super::kms_data_plane_error_class(&missing), "key_not_found");
     }
 
     #[test]
@@ -5563,37 +5560,6 @@ mod tests {
         reset_sse_dek_provider();
     }
 
-    /// A write whose resolved key — from the request header or a bucket
-    /// default rule — is unknown to the KMS must come back as the client
-    /// error S3 uses for it, all the way from the backend lookup. Answering
-    /// 500 here made a bucket default pointing at a deleted or mistyped key
-    /// look like a server outage (rustfs/backlog#2330, KMS-312).
-    #[tokio::test]
-    async fn kms_provider_reports_an_unknown_key_as_kms_not_found() {
-        let _guard = lock_sse_test_state().await;
-        reset_sse_dek_provider();
-
-        let manager = configure_test_global_local_kms().await;
-        let provider = KmsSseDekProvider::new_with_service_manager(manager)
-            .await
-            .expect("kms provider should initialize from the configured test manager");
-
-        let context = super::build_object_encryption_context("bucket", "object", None);
-        let error = provider
-            .generate_sse_dek(&context, "no-such-key")
-            .await
-            .expect_err("the Local backend must refuse a key it does not hold");
-        assert_eq!(
-            error.code,
-            S3ErrorCode::Custom(crate::error::KMS_KEY_NOT_FOUND_ERROR_CODE.into()),
-            "got {error:?}"
-        );
-        assert!(error.message.contains("no-such-key"), "the missing key must be named: {error:?}");
-        assert_eq!(super::kms_data_plane_error_class(&error), "key_not_found");
-
-        reset_sse_dek_provider();
-    }
-
     /// Objects without a rewrappable envelope — plaintext, SSE-C, or a
     /// MinIO-sealed opaque data key — are reported NotApplicable without any
     /// provider call.