Brian Vincent

This commit **refactors** the **`buck2_file_watcher`** module to address a **concurrency issue** and improve its reliability. It modifies the `sync` method within `app/buck2_file_watcher/src/notify.rs` to significantly **reduce the scope of a mutex lock**. By holding the lock only for the `mem::replace` operation, it prevents the lock from blocking the notify callback thread, which previously caused filesystem events to be missed. This **performance optimization** aims to make Buck2's file watching more robust and less prone to overlooking changes.

This commit **fixes a critical bug** in the **remote execution gRPC client** that caused the application to panic during `ByteStream` uploads. Previously, an I/O error encountered within the upload loop would trigger an unrecoverable crash due to the use of `.unwrap()`. By replacing `.unwrap()` with the `?` operator in `remote_execution/oss/re_grpc/src/client.rs`, I/O errors are now properly propagated up the call stack. This **bug fix** ensures graceful error handling instead of a fatal panic, significantly improving the **robustness of remote execution uploads** and preventing unexpected application termination.

This commit provides a **bug fix** by correcting a misleading error message within the **remote execution client**'s gRPC implementation. Specifically, it updates the error message from "for writing failed" to "for reading failed" when a file fails to open during the `batch_update_blobs` operation in `remote_execution/oss/re_grpc/src/client.rs`. This **maintenance** improvement ensures that users receive accurate diagnostic information when encountering issues with file access during **batch blob uploads**, making debugging more straightforward.

This commit **optimizes remote execution performance** by **increasing the default Content Addressable Storage (CAS) Time-To-Live (TTL)**. It modifies the `cas_ttl_secs` parameter within the `REClientBuilder` in `remote_execution/oss/re_grpc/src/client.rs`, extending its default value from 60 seconds to 3 hours. This **configuration tuning** addresses an issue where the previous aggressive 60-second TTL led to frequent, unnecessary rechecks of blob availability. The change significantly **reduces redundant network requests and improves the efficiency of remote execution operations** by assuming a more realistic retention period for CAS blobs.

This commit provides a **bug fix** for the **Open Source Remote Execution (OSS RE) client**, specifically within its gRPC implementation. Previously, the `execute_with_progress` method in `remote_execution/oss/re_grpc/src/client.rs` incorrectly hardcoded the `skip_cache_lookup` parameter to `false`, causing the client to ignore requests to bypass cache reads. This change ensures that the `skip_cache_lookup` value from the `ExecuteRequest` is now correctly forwarded, allowing users of the **OSS RE client** to effectively utilize the `--skip-cache-read` flag and induce cache misses as intended.

This commit introduces a **new capability** to manage resource usage by adding a configurable `execution_concurrency_limit` to the **Buck2 remote execution client**. This **enhancement** specifically integrates a new execution semaphore into the `buck2_execute` client's `execute` method, which is configured via the `execution_concurrency_limit` option in `buck2_re_configuration`. By limiting the number of concurrent remote execution requests, this feature helps **prevent overwhelming remote execution services** and improves overall build stability and efficiency. This provides a crucial mechanism for **resource management** within the remote execution subsystem.

This commit **fixes a panic** that occurred when executing `buck2 debug chrome-trace` in debug mode, caused by **duplicate `clap` argument IDs**. Such duplication could also lead to subtle issues in release builds, so this change acts as a **bug fix** and **preventative refactoring**. Specifically, the `path` argument within the **`buck2_cmd_debug_client` module**'s `chrome_trace.rs` file has been renamed to `event_log_path` to ensure unique identification. This **improves the stability and reliability** of the `buck2` debug tooling by preventing runtime errors related to command-line argument parsing.

This commit **fixes a critical memory consumption bug** within the **remote execution client** by correcting the LRU cache size for `find_missing_cache`. Previously, the cache in `remote_execution/oss/re_grpc/src/client.rs` was mistakenly configured for approximately 52 million entries, leading to an excessive memory footprint of 8-10GB instead of the intended 50MB. This **bug fix** reduces the cache to a more appropriate 500,000 entries, which will significantly **reduce memory usage** to around 100MB and prevent potential out-of-memory issues, thereby improving the stability and performance of the **remote execution subsystem** during large builds.

This commit **updates the `rust-overlay` input** within the `flake.lock` file to a newer revision. This **maintenance fix** resolves a critical issue where the **Nix development environment** failed to provide the specific **Rust nightly toolchain `nightly-2025-12-05`**, which was explicitly required by the project's `rust-toolchain` file. Previously, this mismatch caused the Nix dev shell to fail with an "Nightly is not available" error. By updating the pinned `rust-overlay` revision, the Nix environment can now correctly find and utilize the necessary Rust version, ensuring stable environment setup for developers.

This commit **introduces a new capability** to specify a **priority for remote executions** within Buck2. A `priority` field has been added to the `CommandExecutorConfig` Starlark function, allowing users to define a priority for their remote build actions. This priority is then propagated through the **remote execution subsystem**, including `RemoteEnabledExecutorOptions` and various client and manager components, and ultimately included in the gRPC `ExecuteRequest` sent to the remote execution server. This **feature enhancement** enables users to influence the scheduling of their remote tasks, although the exact interpretation of the priority value remains server-dependent.

This commit implements a **bug fix** to enhance the consistency of the build system's internal state with the file system. It modifies the **file watcher** (`buck2_file_watcher`) to explicitly detect and flag instances where file system events are missed by the underlying `notify` crate. When such an event loss is identified during a `sync` operation, the entire **DICE graph** is proactively dropped, forcing a complete re-evaluation of dependencies. This critical change prevents stale caches and resolves issues where the build system could become out of sync with the actual file system, ensuring reliable builds. A minor debug log was also added to `unstable_drop_everything` within the `dice` module for improved observability.

This commit introduces a **new capability** by adding **network statistics tracking** to the **OSS Remote Execution (RE) client**. It implements this by wrapping the transport layer of the `re_grpc` client channels with a `CountingConnector` to monitor and count uploaded and downloaded bytes. This enhancement provides crucial visibility into data transfer, enabling these **network statistics to be displayed in the CLI interface** for remote execution operations. The change involves integrating the counting mechanism into the `REClientBuilder` and exposing global statistics via functions like `get_network_stats`.

This commit performs **maintenance** by **removing the execute bit** from several **C++ tooling Python scripts** located in `prelude/cxx/tools/`, including `linker_wrapper.py` and `hmap_wrapper.py`. This change addresses an inconsistency in build inputs that arose when compiling for Linux from Windows, where the execute bit was not preserved, leading to different file metadata compared to native Linux builds. By standardizing these permissions, the commit significantly improves **build cacheability** and ensures **consistent build inputs** regardless of the host operating system, thereby enhancing the reliability and efficiency of the build system.

This commit **upgrades the Google protocol buffer definitions** to their latest versions, a **maintenance update** that primarily affects the **remote execution gRPC client** and the **protocol definition generation**. This update enables support for new methods and compression types, such as Brotli, within the remote execution system. To manage upstream deprecations while preserving current functionality, `#[allow(deprecated)]` annotations were added in `buck2_action_impl`, `buck2_execute`, and `buck2_execute_impl` for fields that remain in use. Concurrently, gRPC client methods in `remote_execution/oss/re_grpc/src/client.rs` were **refactored** to align with the new proto structures, ensuring this definition upgrade introduces **no behavioral changes**.

This commit **adds support for bytestream compression** to the **remote execution gRPC client**, enabling more efficient data transfer for uploads and downloads when server capabilities allow. It also **improves the robustness and specification compliance** of blob handling by allowing zero-sized blob uploads to succeed silently and treating a `committed_size` of -1 (indicating an already-uploaded blob) as a success rather than an error. This **new capability and set of bug fixes** primarily affect the `remote_execution/oss/re_grpc` module, specifically its `upload` and `download` operations. The changes will lead to **faster and more reliable remote execution operations** by reducing network overhead and correctly handling specific API edge cases.

This commit introduces a **performance optimization** to the **remote execution GRPC client** by enhancing how `FindMissingBlobs` calls are handled in `client.rs`. It implements a cache to track in-flight requests for individual blobs, preventing redundant queries, and improves the chunking logic to send full requests containing up to 100 blobs. This **feature enhancement** significantly **reduces the number of HTTP requests** (up to 90% in tests) and improves overall efficiency when querying missing blobs, particularly for large projects.

This commit introduces a **new configuration option**, `cas_ttl_secs`, enabling users to customize the Content Addressable Storage (CAS) Time-To-Live (TTL) for **Bazel's remote execution API**. This value, previously hardcoded, can now be specified via `buckconfig` within the `Buck2OssReConfiguration` module. The **remote execution gRPC client** will now utilize this configurable TTL when setting the expiration for uploaded digests, replacing a fixed value. This **feature enhancement** provides greater control over remote cache behavior and resource management within Buck2's **remote execution settings**.

This commit **fixes certificate errors** that emerged in the **Remote Execution (RE) client** following an upgrade of the `tonic` library from version 0.10 to 0.12. The updated `tonic` version requires an additional step to properly include `webpki` roots in the TLS configuration, which was previously missing. This **bug fix** modifies the `create_tls_config` function within `remote_execution/oss/re_grpc/src/client.rs` to correctly enable these roots, thereby restoring the **RE client's ability to establish secure gRPC connections**.