Scott Cao

This commit introduces **periodic checks** within the **Buck2 daemon** to monitor the accessibility of the project's working directory. It logs **soft errors** when the directory becomes inaccessible or recovers, providing crucial data on the frequency and recoverability of such issues. This **new capability** in `app/buck2_daemon/src/daemon.rs` (specifically in `run` and `check_daemon_dirs_thread`) aims to gather insights into stale working directories, particularly with temporary checkouts. The collected data will inform future decisions on whether to implement immediate daemon shutdowns for these scenarios, ultimately enhancing the **daemon's robustness and error handling**.

This commit introduces a **new test case** to the **Buck2 daemon testing suite**, specifically `test_daemon_survives_checkout_removal.py`, to verify that the **Buck2 daemon** remains active even after its associated source code checkout is removed. This **maintenance** work confirms an existing and expected behavior of the daemon's lifecycle, ensuring its stability under various conditions. Additionally, the `tests/e2e_util/buck_workspace.py` utility is **refactored** to extract Eden checkout removal logic into a dedicated `eden_remove` function, improving the modularity and reusability of test cleanup operations. This enhances the robustness and clarity of **Buck2's end-to-end testing infrastructure** by providing a clearer separation of concerns in test utilities.

This commit **refactors and enhances the installer tests** for the `buck install` command by **migrating them to a sandboxed environment** (`inplace=False`) within the `tests/core/install` directory. Previously, these tests ran `inplace=True` due to challenges in building GRPC-enabled installers with a basic prelude. The change introduces a **new testing strategy** where the installer is pre-built as an artifact and then passed to the end-to-end tests, ensuring proper isolation and accurate testing of `buck install` functionality. This significantly **improves the robustness and correctness** of the **Buck installer testing suite** by providing a more realistic test setup.

This commit **improves the reliability and timeliness of Out-Of-Memory (OOM) detection** for the **Buck2 daemon** by the Buck2 client. Instead of waiting for the daemon to report its cgroup path, the client now proactively determines the daemon's cgroup path directly from its process ID. This **feature enhancement** centralizes the OOM detection logic within the `buck2_client_ctx` module, specifically in `EventsCtx`, allowing for earlier and more independent OOM event handling. The change primarily impacts the **daemon client connection, event streaming, and OOM subscriber subsystems**, ensuring the client can react to daemon OOMs more promptly.

This commit introduces a **new debug command**, `buck2 debug crash oom`, designed to **simulate Out-Of-Memory (OOM) conditions** within the Buck2 daemon for testing purposes. It extends the `buck2_cli_proto/daemon.proto` with an `Oom` crash type and implements both client-side and server-side logic in `buck2_cmd_debug_client` and `buck2_server` respectively, to allocate a specified amount of memory and generate pressure. This **testing utility** allows developers to validate Buck2's OOM detection and handling mechanisms, and includes new OS-specific utilities for retrieving system page size.

This commit introduces a **new capability** for the `buck2` client to accurately detect and report when its daemon process is terminated by an OOM killer. Previously, such events resulted in a generic daemon disconnect error; now, the **`buck2_client_ctx`** module includes new logic to parse `dmesg` logs and identify OOM-related daemon kills via the new `oom.rs` subscriber. This enhancement to **error reporting** updates the **`buck2_data`** and **`buck2_error`** modules with a specific `DAEMON_OOM_KILLED` tag and `FatalOom` exit code. The change provides clearer diagnostic messages to users and enables better telemetry for memory-related daemon failures.

This commit **introduces a new `pagable_transition_alias` rule** and applies it to the `fbcode//buck2:buck2` target, resolving a critical build configuration conflict. This **build system enhancement** ensures that `buck2` is always built with pagable enabled, even when its parent target (like an `antlir` VM test) has pagable disabled. The change in `defs.bzl` adds the necessary Starlark functions to define this new alias type, allowing **E2e VM tests** that depend on both `buck2` and `antlir` to compile successfully. This provides a robust mechanism for enforcing specific build configurations on critical targets, preventing compilation failures due to conflicting dependency requirements.

Refactor graph traversal for sketches

This commit introduces a new **feature** by adding the `parse_legacy_config_list_property` asynchronous function to the `HasLegacyConfigs` trait within the **`buck2_common`** crate's `legacy_configs` subsystem. This helper specifically enables the parsing of **legacy configuration properties that are lists of values**, complementing existing functionality for single-value properties. It provides a convenient and dedicated utility for handling list-based configurations, simplifying future development and improving code clarity when interacting with legacy settings.

This commit introduces a **new capability** to the **`ArtifactGroupValues`** component within `buck2_build_api` by adding the `iter_many` method. This method enables **efficient iteration over multiple artifact groups** (specifically `tsets`), automatically handling the deduplication of shared child artifacts. The change improves the performance and resource utilization of artifact processing workflows by providing a streamlined way to process aggregated artifact data. This enhancement is foundational for future optimizations, such as efficiently computing artifact size and count sketches across multiple `ArtifactGroupValues`.

This commit provides a **bug fix** to prevent the **action graph sketch** from unintentionally generating **detailed top-level metrics** and **command-level overall sketches** when they are not explicitly requested. It **refactors** the `build` command within `app/buck2_server_commands/src/build.rs` to decouple sketch computation from aggregated metrics, ensuring these are only produced on demand. Additionally, the `bxl` and `test` commands in `app/buck2_bxl/src/command.rs` and `app/buck2_test/src/command.rs` are updated to correctly suppress this unrequested metric generation. This change improves efficiency by avoiding unnecessary metric computation across key Buck2 operations.

This commit introduces a **new capability** to expose the `allprocs_cgroup_path` through the `buck2` daemon's status reporting. It **modifies the `StatusResponse` protobuf message** in `app/buck2_cli_proto/daemon.proto` to include an optional `allprocs_cgroup_path` field. The `buck2_server` daemon now populates this field by retrieving the cgroup path from its memory tracker, enabling the `buck2` client to access this information upon daemon startup. Consequently, users can now view the `allprocs_cgroup_path` directly in the output of the `buck2 status` command, enhancing **diagnosability** and client-side awareness of the daemon's cgroup configuration.

This commit **refactors** the `materializer` subsystem by correcting the behavior of the `get_artifact_entries_for_materialized_paths` function within `buck2_execute` and `buck2_execute_impl`. The parameter `fetch_projected_artifact_entries`, which was intended to fetch root artifact entries for subpaths, was incorrectly designed to fetch projected artifact entries. It has been **renamed** to `fetch_root_artifact_entries_for_subpaths` to accurately reflect its purpose. This **correction** ensures that the `materializer` now retrieves the correct root artifact entries for projected artifacts, simplifying the internal code path and properly laying the groundwork for future features like artifact count and size sketches.

This commit addresses a **minor textual error** by correcting a typo within the codebase. It represents a **maintenance fix** aimed at improving the overall clarity and correctness of documentation, comments, or user-facing messages. While the specific location of the typo is not detailed, this change contributes to the general quality and readability of the project. The impact is primarily on textual accuracy, with no expected functional changes to any specific module or feature.

This commit introduces a **new capability** to the **materializer subsystem**, extending the `get_artifact_entries_for_materialized_paths` API to support fetching entries for **projected artifact subpaths**. With the new `fetch_projected_artifact_entries` parameter, the materializer can now traverse materialized directories to locate and return entries for specific subpaths (e.g., `src/subdir/file.txt`) within a larger artifact, rather than just the root artifact. This enhancement is crucial for **correctly implementing artifact path sketching for projected artifacts**, enabling more precise resolution of individual files within materialized directory structures. The `run` action implementation is updated to accommodate this API change, explicitly setting the new parameter to `false` for its current usage.

This commit **adds a new unit test** to verify the functionality of `get_artifact_entries_for_materialized_paths` within the **deferred materializer's command processor** in `buck2_execute_impl`. To facilitate this, new testing utility methods, `testing_declare_existing_full` and `testing_get_artifact_entries_for_materialized_paths`, are introduced to the `command_processor.rs` file. This **new test** ensures the correct retrieval of artifact entries for materialized paths, thereby **improving the reliability** and test coverage of the materialization subsystem.

This commit introduces **unit tests** for the `calc_output_count_and_bytes` function within the **`buck2_execute` module**, specifically in `output_size.rs`. These tests cover various scenarios, including multiple files, nested directories, symlinks, and accurate physical disk size calculations, ensuring the correctness of output size estimations. This **quality assurance** work improves the reliability of Buck2's execution phase by verifying a critical utility function. Additionally, it updates the `app/buck2_execute/BUCK` file to add `tempfile` as a necessary test dependency for these Rust tests.

This commit **enhances** the **Buck2 execution and build API** by modifying the `calc_output_count_and_bytes` function to optionally account for symlinks. A new `include_symlinks` parameter is introduced to the `OutputSize` trait and its implementations, allowing for more precise control over how output counts and bytes are calculated, particularly for directories. This **feature addition** and **refactoring** updates all existing call sites within `buck2_build_api` and `buck2_execute_impl` to explicitly set `include_symlinks = false`, maintaining current behavior while laying the groundwork for future artifact path sketch calculations that require symlink awareness.

This commit implements a **bug fix** within the **remote execution (RE) logging system** of `buck2_execute`. It modifies the `to_proto` function in `app/buck2_execute/src/execute/kind.rs` to **always include `use_case` and `platform` details** when converting RE action data to its protobuf representation. Previously, these critical details were being omitted, appearing as null, which hindered **non-deterministic action analysis**. By ensuring these details are preserved, the change significantly improves the **observability and debuggability** of remote execution actions, providing essential context for future analysis.

This commit introduces a **new observability feature** by logging the daemon's cgroup path. It **extends the `SystemInfo` protocol buffer message** in `app/buck2_data/data.proto` with an optional `daemon_cgroup_path` field. The **`buck2_server` daemon** then populates this field within its `system_info` method, ensuring that the cgroup path is included in the `SystemInfo` event sent to clients at the beginning of each command. This enhancement provides crucial context for **monitoring and debugging daemon resource management**.