mirror of
https://github.com/OPCFoundation/UA-.NETStandard.git
synced 2026-10-29 07:07:24 +08:00
fuzzing
3
Commits
| Author | SHA1 | Message | Date | |
|---|---|---|---|---|
|
|
0fbc900919 |
Add opt-in SubscriptionRecoveryPolicy for unsolicited Good_SubscriptionTransferred (#3540) (#3847)
# Description Adds an opt-in `SubscriptionRecoveryPolicy` for both the Classic V1 and V2 subscription engines that lets the client recover from server-emitted `Good_SubscriptionTransferred` notifications which arrive against a freshly created subscription (observed against Kepware after a server re-initialise, very likely a leak from the pre-restart session's pending status notifications onto the new session — abetted by Kepware re-using `SubscriptionId = 1` across generations). Per [OPC UA Part 4 §5.14.7](https://reference.opcfoundation.org/Core/Part4/v105/docs/5.14.7), `Good_SubscriptionTransferred` is supposed to be delivered to the *old* Session whose subscription was transferred away via `TransferSubscriptions`. The existing V1 behaviour of stopping the per-subscription publish loop on receipt is therefore **spec-compliant**; the issue is that some servers deliver the notification on a Session that is still actively using the subscription, leaving the per-subscription dispatch silently disabled. The fix is **opt-in** and **non-breaking**: - New `SubscriptionRecoveryPolicy` enum (`ReportOnly` default, `RecreateOnUnsolicitedTransfer`). - V1 Classic `Subscription.RecoveryPolicy` property; V2 `Subscriptions.SubscriptionOptions.RecoveryPolicy` init property. - When the policy is `RecreateOnUnsolicitedTransfer`, the SDK drops queued acks for the dead subscription id (avoiding `BadSubscriptionIdInvalid` ack errors on servers that re-use ids) and recreates the subscription in place on the same Session. ## What changed **Public API additions (non-breaking, defaults preserve current behaviour):** - `SubscriptionRecoveryPolicy` enum (`ReportOnly`, `RecreateOnUnsolicitedTransfer`). - V1 `Subscription.RecoveryPolicy` property. - V2 `Subscriptions.SubscriptionOptions.RecoveryPolicy` init property. - Internal `IMessageAckQueue.DropPendingForSubscription(uint)` and protected `MessageProcessor.AckQueue` accessor. **V1 Classic — `Libraries/Opc.Ua.Client/Subscription/Classic/Subscription.cs`:** - Replaces the inline `Good_SubscriptionTransferred` branch with `HandleGoodSubscriptionTransferred`. `ReportOnly` preserves the original behaviour (stop publish timer/worker); `RecreateOnUnsolicitedTransfer` triggers `TriggerUnsolicitedTransferRecovery`, guarded by `m_recreateAfterTransferInProgress` for idempotency. - Recovery drains pending acks via the new `ClassicSubscriptionEngine.RemoveAcknowledgementsForSubscription` **before** resetting the subscription, then runs `DeleteSubscription()` + `CreateAsync()` on the same session so the publish worker restarts via `StartKeepAliveTimer`. - Discriminator (`Session != null && !Session.Reconnecting && Created`) defers to the reconnect pipeline if it is already running. **V2 — `Libraries/Opc.Ua.Client/Subscription/Subscription.cs`:** - Completes the `OnStatusChangeNotificationAsync` TODO: now always fires `OnPublishStateChanged(mask)` so `PublishState.Transferred` actually reaches handlers (previously the no-op override silently swallowed the flag). - Under `RecreateOnUnsolicitedTransfer`, calls `AckQueue.DropPendingForSubscription(deadId)` then existing `ResetToRecreateAsync` on a worker task — no service calls from the dispatch path. - V2 `SubscriptionManager` implements `DropPendingForSubscription` via drain-and-refill on the prioritised ack channel. **Tests — `Tests/Opc.Ua.Client.Tests/Subscription`:** - V1 in `Classic/SubscriptionUnitTests.cs`: `ReportOnlyPolicyDoesNotRecreateOnUnsolicitedTransferAsync`, `RecreatePolicyRecreatesSubscriptionOnUnsolicitedTransferAsync`, `RecreatePolicyCollapsesConcurrentNotificationsAsync`, `RecreatePolicyDoesNotRecreateWhileSessionReconnectingAsync`. - V2 in `SubscriptionTests.cs`: `GoodSubscriptionTransferredRaisesPublishStateTransferredAsync`, `GoodSubscriptionTransferredRecreatesSubscriptionWhenPolicyEnabledAsync`, `GoodSubscriptionTransferredCollapsesConcurrentRecreatesAsync`. - `FakeMessageAckQueue` gained a `DroppedSubscriptions` recorder so the V2 test can assert ack pruning ran. **Docs — `Docs/TransferSubscription.md`:** - New "Recovering from unsolicited `Good_SubscriptionTransferred`" section with spec context, usage examples for both engines, and the caveat that recovery is recreate-based — **not** lossless failover; the server-side retransmission queue and triggering relationships tied to the invalidated subscription id are not preserved. ## Validation | Scope | Result | | --- | --- | | `Libraries/Opc.Ua.Client` build (`net10.0`, `net48`) | 0 warnings, 0 errors | | `Tests/Opc.Ua.Client.Tests` build (`net10.0`, `net472`) | 0 warnings, 0 errors | | New tests on `net10.0` and `net472` | 7/7 pass on each | | Full subscription unit-test suite (`SubscriptionUnitTests` + `SubscriptionTests` + `MessageProcessorTests` + `SubscriptionManagerTests`) | 110/110 pass on each TFM | | Entire `Opc.Ua.Client.Tests` suite (`net10.0`) | 1268/1268 pass | | Classic transfer integration tests (`Tests/Opc.Ua.Subscriptions.Classic.Tests`, `net10.0`) | 20/20 pass | | V2 transfer compliance tests (`Tests/Opc.Ua.Subscriptions.Tests`, `net10.0`) | 9/9 pass | > Marked as **draft** because the maintainer wanted to first confirm whether the stack is spec-compliant before deciding which knobs to ship; this PR documents the investigation (stack is compliant), and surfaces an opt-in recovery policy that addresses the field symptom without changing the default behaviour. Convert to ready-for-review once the design intent is acknowledged. ## Related Issues - Fixes #3540 ## Checklist - [x] I have signed the [CLA](https://opcfoundation.org/license/cla/ContributorLicenseAgreementv1.0.pdf) and read the [CONTRIBUTING](https://github.com/OPCFoundation/UA-.NETStandard/blob/master/CONTRIBUTING.md) doc. - [x] I have added tests that prove my fix is effective or that my feature works and increased code coverage. - [x] I have added all necessary documentation. - [x] I have verified that my changes do not introduce (new) build or analyzer warnings. - [x] I ran **all** tests locally using the **UA.slnx** solution against at least .net **framework** and .net **10**, and all passed. - [ ] I fixed **all** failing and flaky tests in the CI pipelines and **all** CodeQL warnings. - [ ] I have addressed **all** PR feedback received. --------- Co-authored-by: marcschier <marc.schier@github.com> |
||
|
|
239a40b183 |
GC reduction via IPooledEncodeable activator pooling (#3793)
* GC reduction via IPooledEncodeable activator pooling
Add opt-in activator-level pooling for IEncodeable instances decoded
during V2 subscription publish dispatch. The design replaces the
earlier IAllocator-on-IServiceMessageContext approach (which had a
cross-contamination hazard for non-publish decodes) with a clean
factory-based model where pooling lives on the encodeable activator.
New public API:
- IPooledEncodeable : IEncodeable - opt-in marker with Reuse() method
- PooledEncodeableType<T> : EncodeableType<T> - activator base with
BoundedObjectPool<T> (lock-free, no new package dependency)
- ISubscriptionManager.PoolNotifications { get; set; }
- ManagedSessionOptions.PoolNotifications { get; init; }
- ManagedSessionBuilder.WithPoolNotifications(bool)
Source-gen:
- All concrete non-abstract structure types (including service
request/response types) are generated with PooledEncodeableType<T>
activators and IPooledEncodeable.Reuse() bodies by default
- Derived types use 'new' to hide inherited Reuse/ClearPooledSentinel
so each concrete type routes through its own activator's pool
- No changes to external design XML files (UA Core Services.xml etc.)
- poolable selection is hardcoded in the generator
Benchmarks (MonitoredItemNotification, 1000 ops):
- Allocations: 128 KB -> 408 B (~315x reduction)
- Gen0 collections: 29.75 -> 0.09 per 1000 ops
No interface changes to IEncodeable, IDecoder, IServiceMessageContext,
or ISubscriptionNotificationHandler. Handlers retain by copy when
pooling is enabled.
Call IPooledEncodeable.Reuse() on decoded IServiceRequest and
IServiceResponse after the server is done with them:
- TcpTransportListener.OnRequestReceivedAsync: reuse both request
and response in a finally block after ProcessRequestAsync +
SendResponse complete
- EndpointBase.InvokeServiceAsync (HTTPS path): same pattern after
ProcessRequestAsync + BinaryEncoder.EncodeMessage
Walk acknowledged, overflow-evicted, and disposed NotificationMessage
objects and call IPooledEncodeable.Reuse() on their notification
payloads (MonitoredItemNotification, DataChangeNotification,
EventFieldList, EventNotificationList).
Insertion points in Subscription.cs:
- Acknowledge(): after RemoveAt from m_sentMessages
- InnerPublish() overflow: before RemoveRange/Clear on m_sentMessages
- Dispose(): before m_sentMessages.Clear()
- InnerPublish() unsent-message overflow: before RemoveRange on
the messages list
Replace new T() with XxxActivator.Instance.CreateInstance() for all
server-side notification payload types so instances are rented from
the activator pool instead of heap-allocated:
New static helper ReuseNotificationPayloads() walks the
NotificationMessage.NotificationData, extracts DataChangeNotification
and EventNotificationList via TryGetValue, iterates their items, and
calls Reuse() on each pooled payload.
|
||
|
|
bd4543e961 |
ManagedSession, Pluggable Subscription engine, and fluent builder + DI (#3730)
- Add new managed session, wrapping session and providing connection management and failover handling - Add new Subscription engine that can act as alternative to current engine, but is based on channels with worker tasks - Add a fluent builder for managed session clients - Replace NodeCache with LruNodeCache, rename ILruNodeCache and LruNodeCache to INodeCache and NodeCache - Update migration guide and document new functionality. |