Ashutosh Grewal

This commit introduces a **new capability** to read the actual switching mode directly from **hardware**, primarily to aid in latency studies for scale-up switches. It adds a `getSwitchingModeFromHw()` **Thrift handler** to the `AgentHwTestCtrl` service and a platform-agnostic **hardware test utility** (`HwTestSwitchingModeUtils`) that retrieves the `SAI_SWITCH_ATTR_SWITCHING_MODE` attribute from the **SAI hardware layer**. This enhancement allows **Agent tests** (`AgentSwitchingModeTests.cpp`) to verify that the configured switching mode is correctly applied and reflected in the hardware, significantly improving the **testing infrastructure** by enabling direct hardware state validation for critical network devices. While fully implemented for SAI, a stub is provided for the **BCM hardware layer** as this attribute is not supported there.

This commit **introduces new hardware tests** to the **FBOSS agent's test suite** for validating network switching modes. It adds `AgentSwitchingModeTests.cpp` to verify the functionality and persistence of **cut-through and store-and-forward modes** across warm boots, as well as correct toggling between them. This **new test development** also includes adding `CUT_THROUGH` as a `ProductionFeature` in `production_features.thrift` and updating the CMake and Buck build configurations. These tests are crucial for **evaluating latency reduction** in scale-up switches, specifically targeting ASICs that support `CUT_THROUGH_FORWARDING`.

This commit performs a **documentation refactoring** by relocating the `MULTIPLANE_TESTING.md` file. The document, which details multiplane testing procedures, has been moved from `fboss/fsdb/tests/utils/` to the more appropriate `host_reach_tracker/` directory. This change clarifies that the testing documentation is specifically relevant to the **Host Reach Tracker** project, not the FSDB subsystem. The move improves the **organizational structure** of the project's documentation, enhancing discoverability and logical grouping of related content for the `host_reach_tracker` component.

This commit **fixes a bug** in the **multi-plane testing instructions** for **FSDB**. Previously, starting a second FSDB instance with `--readConfigFile=false` prevented it from loading necessary ID paths, causing it to reject patch type subscriptions used by `FsdbCowStateSubManager`. By **removing this flag**, the FSDB instance now correctly reads its default configuration, enabling proper handling of these subscriptions and ensuring **multi-plane testing** operates as intended.

This commit **implements a new capability** to **dynamically configure the packet forwarding mode** (e.g., cut-through or store-and-forward) on network switches at runtime. It establishes the **config-to-hardware propagation pipeline** within the **SAI hardware agent**, specifically impacting the `SaiSwitch` and `SaiSwitchManager` components. The `SaiSwitch` now detects changes in `cfg::PacketForwardingMode` and invokes `SaiSwitchManager::setSwitchingMode`, which translates the configuration to the corresponding SAI attribute and applies it to the underlying hardware. This **enhancement** allows for **runtime adjustment of switch latency characteristics**, aiming to reduce latency for scale-up switches.

This commit introduces a **new capability** by adding the `SAI_SWITCH_ATTR_SWITCHING_MODE` attribute to the **SAI Switch API**, enabling control over the **ASIC's packet forwarding mechanism** (store-and-forward vs. cut-through). This foundational change in the **hardware abstraction layer** is crucial for studying and optimizing network latency in scale-up switches. Support for this attribute is integrated across the **SAI ecosystem**, including its definition in `SwitchApi.h`, implementation in the **Fake SAI** for simulation, registration in the **SAI Tracer**, and proper accommodation during **platform initialization** in `SaiPlatform.cpp` and `SaiPhyRetimer.cpp`. Unit tests in `SwitchApiTest.cpp` validate the attribute's functionality.

This commit introduces the **`CUT_THROUGH_FORWARDING`** feature flag to the `HwAsic::Feature` enum, allowing the system to query and manage this hardware capability. It **enables cut-through forwarding** specifically for the **Tomahawk5 ASIC** by updating its `isSupported` method to return true, leveraging `SAI_SWITCH_ATTR_SWITCHING_MODE`, while explicitly disabling it for all other 18 ASIC types. This **new capability** is a foundational step within the **hardware abstraction layer** to study and potentially reduce latency in scale-up switches. The change impacts various ASIC-specific feature detection methods, paving the way for future performance optimizations.

This commit introduces the **PacketForwardingMode** concept, defining an enum (STORE_AND_FORWARD, CUT_THROUGH) in `switch_config.thrift` and adding a corresponding state field to `switch_state.thrift`. This **new capability** establishes the foundational layer for **cut-through forwarding support**, which is crucial for reducing latency in scale-up switches. The **FBOSS agent's state management** is updated to handle this new setting, including getter/setter methods in `SwitchSettings.h` and logic in `ApplyThriftConfig.cpp` to process configuration changes. Furthermore, the **FSDB thriftpath** is regenerated to enable pub/sub access to this new `packetForwardingMode` field, ensuring it's fully integrated into the system's monitoring and configuration pipeline.

This commit **fixes a build failure** within the **`fboss/lib/thrift_service_client`** library by adding the essential `#include <gflags/gflags_declare.h>` to `ConnectionOptions.h`. The absence of this include previously caused compilation errors, specifically on `aarch64 opt` builds, due to the undeclared `gflags` macros like `DECLARE_string`. To fully address this, the `BUCK` build file for the `thrift_service_client` module was also updated to explicitly export `gflags` as an external dependency. This ensures the proper declaration of `gflags` macros and **resolves critical compilation issues** for the module.

This commit **updates the testing documentation** by adding a new "Profile Filtering" section to the `docs/docs/testing/test_categories.md` file. It explains the **`--profile` tag convention** (`t` for traditional, `s` for scale-up) for **T0/T1/T2 test categories**, detailing how to execute **scale-up tests** using `--profile=s`. This **documentation enhancement** clarifies the process for running profile-filtered tests and emphasizes the requirement to tag new test patterns appropriately, improving developer understanding and adherence to testing standards.

This commit **enhances the test infrastructure** by introducing **scale-up profile tags** (`s`) to the **T0 agent hardware and SAI test configuration files**. Specifically, `fboss/oss/hw_sanity_tests/t0_agent_hw_tests.conf` and `fboss/oss/hw_sanity_tests/t0_sai_tests.conf` are updated to categorize tests relevant for scale-up scenarios. This **new capability** allows for the selective execution of scale-up related tests using the `--profile=s` flag, improving efficiency during validation. While T0-only patterns like `*MacLearning*` remain untagged, a new `*AgentEmpty*` pattern for init sanity is introduced and marked for scale-up profiles. This change streamlines test execution and improves the relevance of test runs for different deployment profiles.

This commit **updates the configuration** for **T2 agent hardware tests** and **SAI tests** by introducing `s` (scale-up) and `t` (traditional) profile tags. It annotates existing test patterns in `fboss/oss/hw_sanity_tests/t2_agent_hw_tests.conf` and `fboss/oss/hw_sanity_tests/t2_sai_tests.conf` with these tags, ensuring traditional patterns default to `t` and scale-up patterns receive `s`. Additionally, new discard counter patterns are added as scale-up-only (`s`), while specific scale-out-only SAI tests are explicitly marked as `t`-only. This **maintenance** change enables more precise test selection and execution based on hardware profiles, improving test relevance for different T2 environments.

This commit introduces a **new capability** to the **`run_test.py` script** by adding a `--profile` command-line argument. This argument, when used in conjunction with `--filter_file`, enables **selective filtering of tests** from the configuration file based on specified profile tags. This enhancement allows users to efficiently execute specific subsets of tests, such as those designated for **scale-up scenarios**, without modifying the filter file itself. The change is fully **backward compatible**, ensuring existing test execution workflows remain unaffected when the `--profile` argument is omitted.

This commit **refactors** the **hardware sanity test configurations** for T1 platforms by introducing and applying "scale-up" profile tags. It **moves the `*IngressBuffer*` test pattern** from T2 to `t1_agent_hw_tests.conf`, aligning it with the `*Pfc*` feature it supports and ensuring its execution within T1 profiles. Additionally, a new **`*NetworkAIQos*` pattern is added as a scale-up-only test**, replacing a broader `*Qos*` match and refining the scope of QoS testing. These changes improve the **categorization and targeted execution of tests** within `fboss/oss/hw_sanity_tests/t1_agent_hw_tests.conf` and `fboss/oss/hw_sanity_tests/t1_sai_tests.conf`, enhancing the relevance of test runs for specific hardware and scale-up scenarios.

This commit performs **maintenance and refactoring** by **cleaning up and refining gtest filter patterns** across various hardware test configurations. It **removes dead or misplaced patterns** from **T1 and T2 agent hardware test configurations** (`t1_agent_hw_tests.conf`, `t2_agent_hw_tests.conf`), addressing coverage gaps by updating `*HwIngressBufferTest*` to `*IngressBuffer*` for a renamed class. Additionally, patterns like `*HwInPause*` are **relocated to T1 SAI tests** (`t1_sai_tests.conf`), and `*RouteNeighbor*` is broadened to `*HwRoute*` in `t0_sai_tests.conf` to improve **test coverage** for related hardware route tests. The **documentation** in `agent_hw_test.md` is also updated to reflect these pattern changes and correct a typo, ensuring accurate descriptions for the **agent hardware test suite**.

This commit provides a **bug fix** by correcting a case-sensitivity issue in the gtest filter for **PRBS tests** within the **T0 agent hardware tests configuration**. Previously, the `*PRBS*` filter failed to match `AgentPrbsTest` due to gtest's case-sensitive globbing, preventing these critical tests from running. The filter in `fboss/oss/hw_sanity_tests/t0_agent_hw_tests.conf` has been updated to `*Prbs*` to correctly identify and enable the **PRBS tests**, ensuring proper execution and improved test coverage for the T0 agent. Additionally, the **testing documentation** in `docs/docs/testing/agent_hw_test.md` has been updated to reflect this correct filter string.

This commit performs **maintenance cleanup** by **removing several dead gtest filter patterns** from the **`t0_agent_hw_tests.conf` configuration file**. Patterns such as `L2ClassID`, `HwRoute`, `HwTest_PROFILE`, and `FlextPort` are eliminated because they do not match any existing test fixtures or methods within the `sai_agent_hw_test` binary, making them effectively useless. This change improves the clarity and accuracy of the **test configuration** by removing irrelevant entries that could cause confusion. Additionally, the **`agent_hw_test.md` documentation** is updated to remove references to these defunct patterns, ensuring consistency and reducing cruft in the testing documentation.

This commit **refactors** and **cleans up test code** within the **FBOSS Agent hardware test suite**. It **simplifies calls** to the `verifyAcrossWarmBoots` utility function by migrating from a verbose two-parameter invocation (e.g., `verifyAcrossWarmBoots([] {}, verify)`) to a more concise one-parameter overload (`verifyAcrossWarmBoots(verify)`). This change removes unnecessary boilerplate across several hardware test files, including `AgentHwUdfTests.cpp`, `AgentMirroringScaleTests.cpp`, `AgentOlympicQosSchedulerTests.cpp`, `AgentRouteTests.cpp`, `AgentVoqSwitchTests.cpp`, and `AgentVoqSwitchWithFabricPortsTests.cpp`. The update improves the **readability and maintainability** of these tests without altering their functional behavior or test coverage.

This commit **refactors test code** within the **`AgentFabricSwitchTests`** by migrating calls to `verifyAcrossWarmBoots`. It replaces verbose two-parameter calls, `verifyAcrossWarmBoots([] {}, verify)`, with the more concise one-parameter overload, `verifyAcrossWarmBoots(verify)`. This **simplifies the test suite's implementation** by leveraging an existing utility function overload that internally handles the empty setup lambda, thereby **reducing boilerplate** and improving readability of the **hardware agent tests**. The change has no impact on test functionality or product behavior.

This commit **refactors the build system** by **extracting ARS/Flowlet tests** into a new, dedicated library named `agent_ars_test_src`. This change primarily affects the **`fboss/agent` test infrastructure**, specifically the `agent_hw_test_src` module, by moving relevant source files and their build definitions from `cmake/AgentTestAgentHwTests.cmake` and `fboss/agent/test/agent_hw_tests/BUCK`. The main goal is to **optimize build times** for these specific tests, enabling faster iteration. To ensure **backward compatibility**, the newly created `agent_ars_test_src` library is added as a dependency of the original `agent_hw_test_src`.