Parvez Shaikh

This commit **enhances observability** within the **FBOSS agent** by introducing a new `fb303` counter, `rib_route_programming_time.us`. This **new feature** exports the duration of RIB route programming, measured in microseconds, each time `SwSwitchRouteUpdateWrapper::updateStats()` is called. The counter provides detailed performance metrics, including p50, p95, p99, and p100 percentiles, allowing for better monitoring of route update efficiency within the `SwitchStats` subsystem. A corresponding unit test was added to verify the proper export of this new statistic.

This commit introduces a **new observability capability** by adding a `rib_route_programming_time.us` counter to the **`SwitchStats`** module within `fboss/agent`. This `QuantileStatWrapper` will track **RIB route programming latency** in microseconds, providing p50, p95, p99, p100 percentiles, along with count and average. The change involves declaring the `ribRouteProgrammingTimeUs_` member and its accessor `ribRouteProgrammingTimeUs` in `SwitchStats.h` and initializing it in `SwitchStats.cpp`. This is the initial step to enable detailed monitoring of RIB programming performance, with the actual wiring to the update path planned for a subsequent commit.

This commit significantly **improves documentation** within the **FBOSS agent's SAI platform implementations** by adding detailed comments to the `getSaiProfileVendorExtensionValues()` function in `SaiMinipack3NPlatform.cpp` and `SaiYangraPlatform.cpp`. These comments explain the purpose, impact on SAI SDK behavior, and FBOSS's rationale for setting various **SAI vendor extension key-value pairs**, verified against NVIDIA SAI SDK source. This **maintenance** task enhances understanding of the SAI profile configuration for both the **Minipack3N and Yangra platforms**. A notable finding is also documented, indicating that `SAI_KEY_ROUTE_METADATA_FIELD_SIZE` is silently ignored by the SDK, with the actual metadata size derived from other keys, providing crucial context for **route metadata configuration**.

This commit **enables** the `SAI_KEY_NO_HQOS_QUEUE_TO_SUBGROUP` key within the **SAI platform configuration** for the `SaiYangraPlatform` by modifying the `getSaiProfileVendorExtensionValues` function. This crucial **configuration update** ensures **correct Quality of Service (QoS) treatment for Priority Flow Control (PFC) frames**, allowing them to be handled per Traffic Class (TC) rather than per port. Previously, High-Quality of Service (HQoS) operating at the port level would cause PFC frames to backpressure all packets on a port. Now, only the specific TC for which a PFC XOFF is received will be affected, **enhancing network stability and performance** by preventing unnecessary traffic halts.

This commit introduces a **refactoring** to the **SAI object management subsystem** by adding a destructor to `SaiObjectEventAggregateSubscriber` in `SaiObjectEventSubscriber.h`. This new destructor automatically calls `resetObject()` to ensure proper cleanup of the managed SAI object when a subscriber is destroyed. This change **eliminates boilerplate code** in concrete subscriber destructors, specifically removing manual `resetObject()` calls from `ManagedNextHop::~ManagedNextHop()` and `ManagedSaiNextHopGroupMember::~SaiNextHopGroupMember()`. The `resetObject()` calls in runtime event handlers like `removeObject()` and `handleLinkDown()` remain unaffected, ensuring correct behavior during dynamic object lifecycle events.

This commit **refactors** the **publisher key configuration** within the `fboss/agent/hw/sai/store` module by **removing the `detail::PublisherKeyInternal` helper class**. Instead of inheriting, each `PublisherKey<T>` specialization now directly defines its `type` and `custom_type` aliases, making them self-contained. This change **simplifies `Traits.h`** to only essential declarations and centralizes publisher key configurations in `SaiObjectEventPublisher.h`. The primary goal is to improve the **readability and maintainability** of type traits by making publisher key definitions immediately clear and self-contained.

This commit **refactors** the **SAI object event publishing system** by reorganizing trait declarations within `fboss/agent/hw/sai/store/SaiObjectEventPublisher.h`. It groups `IsObjectPublisher` and `PublisherKey` specializations for each publisher type, significantly improving **code readability** and making it easier to understand a publisher's full configuration at a glance. A new comment block was added to clarify the two-declaration pattern and the distinction between `AdapterHostKey` and custom key publishers. This is a **non-functional change** primarily focused on **improving maintainability** and clarity within the `fboss/agent/hw/sai/store` module.

This commit performs a significant **refactoring** within the **SAI hardware abstraction layer** by **removing the `IsPublisherKeyCustomType` trait**. The `PublisherKeyInternal` template in `fboss/agent/hw/sai/store/Traits.h` now automatically derives whether a publisher uses a **custom key** by comparing its `Attr` type with `ObjectTraits::AdapterHostKey`. This change simplifies the **type system**, eliminating SFINAE machinery and reducing boilerplate related to custom key identification. Consequently, methods like `SaiObject::getPublisherKey()` and `SaiStore::setObject()` are updated to leverage this new `custom_type` derivation for key resolution.

This commit **adapts the SAI SDK's serdes attribute handling** specifically for the **Chenab ASIC** to prevent critical warm boot issues. It **removes the `Preemphasis` attribute** from the `SaiPortSerdesTraits::CreateAttributes` tuple and guards all related code paths with `#if !defined(CHENAB_SAI_SDK)` preprocessor checks. This **bug fix** resolves a state mismatch where the Chenab ASIC uses `TX_FIR_PRE1` instead of `Preemphasis`, which previously led to serdes recreation and potential **traffic disruption during warm boot**. The changes affect the **`PortApi` definition**, **NPU serdes management** in `SaiPortManager.cpp`, API tracing, and various test utilities, ensuring **platform-specific stability** for Chenab.

This commit **refactors** the **`SaiObjectEventSubscriber`** within the **SAI agent's event handling subsystem** by replacing `std::any` with `std::shared_ptr<void>` for storing subscription handles. This change **improves debuggability** and code clarity while maintaining the crucial lifetime management and circular dependency breaking behavior. The `std::shared_ptr<void>` correctly calls the subscription's destructor and provides better introspection, addressing a long-standing TODO. This is a **maintenance improvement** that enhances the robustness of the **SAI agent's event subscription mechanism**.

This commit **adds new unit tests** to validate the publish/subscribe (pub/sub) behavior of `ManagedVlanRifNeighbor` within the **SAI agent's hardware layer**. Specifically, these tests verify the correct creation, removal, and recreation of neighbors based on notifications from the `SaiFdbTraits` publisher. It also confirms that link-down events correctly cascade from FDB changes to the neighbor state. This enhancement improves the **reliability and correctness of network neighbor resolution** by ensuring the `ManagedVlanRifNeighbor` accurately reflects FDB state changes.

This commit **adds new pub/sub tests** for the **ManagedSaiNextHopGroupMember** component within the **SAI switch agent**. These tests specifically verify the full pub/sub cascade, ensuring that next hop group members are correctly created when a neighbor resolves (via FDB, neighbor, and nexthop traits) and properly removed when a neighbor unresolves or a link goes down. This **enhances the robustness and reliability** of next hop group management by validating its dynamic response to network state changes, crucial for accurate routing and forwarding. The new test suite is located in `ManagedNextHopGroupMemberPubSubTest.cpp`.

This commit **refactors** the **SAI object event subscription system** by introducing a default no-op `handleLinkDown()` method within the `SaiObjectEventAggregateSubscriber` base class. This **code quality improvement** eliminates the need for concrete subscribers, such as `ManagedSaiNextHopGroupMember` and `ManagedVlanMember`, to implement empty `handleLinkDown()` stubs when no specific link-down logic is required. Consequently, it **simplifies the implementation of derived classes** and reduces boilerplate. This change primarily impacts the `fboss/agent/hw/sai/store` module, making the codebase cleaner and more maintainable by centralizing default behavior.

This commit **refactors** the **`SaiObject`** and **`SaiObjectEventPublisher`** components to simplify template metaprogramming. It replaces the SFINAE-based `IsPublisherKeyAdapterHostKey` trait and its partial `PublisherKey` specialization with direct, explicit `PublisherKey<T>` specializations for `SaiIpNextHopTraits`, `SaiMplsNextHopTraits`, `SaiSrv6SidlistNextHopTraits`, and `SaiNeighborTraits`. This change unifies the template specialization pattern across all `PublisherKey` types, aligning them with the approach already used by CustomType publishers. Additionally, the `SaiObject::getPublisherKey()` static assertion is updated to use `std::is_same_v`, resulting in a **more consistent and less complex template specialization mechanism** within the **SAI hardware abstraction layer**.

This commit performs a **refactoring** by **removing the unused `IsPublisherKeyCreateAttributes` trait** and its associated logic from the **`SaiObject` subsystem**. The trait was identified as dead code because it was never specialized for any publisher type, leading to unnecessary complexity. Specifically, this change removes the trait declaration from `Traits.h`, two SFINAE partial specializations of `PublisherKey` that depended on it, and a dead `else` branch within `SaiObject::getPublisherKey()`. This **code cleanup** simplifies the remaining `PublisherKey` specialization and the overall codebase by eliminating unneeded conditional logic and dead code paths.

This commit **adds comprehensive unit tests** for the **`ManagedNextHop`'s publish/subscribe mechanism** within the **`fboss/agent/hw/sai/switch`** component. These tests verify critical behaviors, including immediate next hop creation upon neighbor existence, deferred creation when a neighbor resolves later, correct non-creation when a neighbor is absent, and proper removal or reset of next hops in response to `SaiNeighborTraits` notifications and cascading events like link down. This work significantly **improves the test coverage and reliability** of next hop management logic, ensuring its correct interaction with dynamic neighbor state changes in the SAI agent.

This commit introduces **new edge case tests** for the **SAI object event publisher** within the **pub/sub framework**, specifically targeting publisher notification paths. It adds `notifyDeleteWithNoSubscribers` to verify that `notifyDelete` handles scenarios with no active subscribers without crashing, and `publisherReCreationAfterRemoval` to validate the full create-delete-re-create cycle, ensuring subscribers receive correct notifications and `isReady()` states. These **tests** are crucial for **closing coverage gaps** and enhancing the **robustness** and **reliability** of the event system by confirming correct behavior under less common conditions.

This commit introduces **new tests** for the **pub/sub framework's** `SaiObjectEventAggregateSubscriber` to validate its core behavior. The tests specifically verify that managed objects are only created when **all required publishers are alive**, and correctly removed when a publisher dies. Scenarios covered include preventing object creation with partial publishers, successful creation upon all publishers becoming active, and proper object removal when a publisher is destroyed. This **testing work** enhances the **robustness and reliability** of the aggregate subscriber pattern, ensuring its correct functioning within the system.

This commit **adds new tests** for the `ManagedFdbEntry`'s publish/subscribe behavior within the `fboss/agent/hw/sai/switch` component. These **new tests** specifically verify that FDB entries are correctly created, removed, and recreated in response to events from the `SaiBridgePortTraits` publisher. This ensures the reliability of FDB management as bridge ports are dynamically added or removed, improving the overall stability and correctness of the system's forwarding database.

This commit **introduces new tests** to validate the **publish/subscribe behavior of `ManagedVlanMember`** within the FBOSS agent's SAI layer. These tests specifically verify the **correct lifecycle management of VLAN members**, ensuring they are accurately created, removed, and subsequently recreated based on notifications from the `SaiBridgePortTraits` publisher. This work enhances the **robustness and reliability of the VLAN management subsystem**, guaranteeing consistent network configuration even with dynamic changes to bridge port presence. The added tests improve **quality assurance** for how VLAN memberships are dynamically maintained.