Alex Robinson

This commit **updates the documentation** for the **Durable Objects API**, specifically clarifying that the `locationHint` parameter is supported by the `getByName` method. It addresses a discrepancy where the existing documentation in `src/content/docs/durable-objects/api/namespace.mdx` did not reflect this functionality, despite it being present in the source code. This **documentation update** ensures developers have accurate information about an **optional parameter**, improving the clarity and completeness of the API reference for Durable Objects.

This commit **reverts** a previous change, effectively removing a **logging mechanism** from the `workerd/io` subsystem. Specifically, it undoes the addition of periodic warning logs that would trigger when the `deleteAll` method was called on **`ActorCache`** or **`ActorSqlite`** instances while an alarm was still set. This **maintenance** revert aims to reduce log noise by eliminating these specific warnings, impacting the observability of alarm management within these actor implementations.

Improves diagnostic tracing for actor SQLite commits, making delays between commit spans and callbacks more obvious. This is an isolated change to observability, with no functional impact.

This commit introduces **tracing support for SQLite commit callbacks** within the `workerd` environment, enabling better observability of database operations. It **modifies the `ActorSqlite` component** to accept and pass a `SpanParent` object to its `commitCallback` function, ensuring that commit operations are properly integrated into the distributed tracing system. This **new capability** enhances the debugging and monitoring of SQLite interactions by linking commit events to their parent spans, providing a clearer picture of request flows. The changes primarily affect the `workerd/io` subsystem, specifically `ActorSqlite` and its instantiation within `WorkerService`, to propagate tracing context.

This change adds internal diagnostic logging to `workerd`'s actor I/O layer, specifically for cache and SQLite actors. It helps gather data on a specific operational scenario without affecting user-facing functionality or introducing new capabilities.

This change affects the core `deleteAll` operation within actor state management, introducing a compatibility flag that alters its default behavior to also delete alarms. This is a significant behavioral change for users of `deleteAll`, as it will now remove alarms in addition to data, potentially impacting existing applications that relied on alarms persisting after a `deleteAll` call.

This commit **improves documentation** for **Durable Objects (DO) alarm handlers** by adding crucial guidance on exception handling. It clarifies that **uncaught exceptions** within an `alarm` handler can lead to **retry exhaustion**, causing alarms to stop functioning. The update includes a new warning in `src/content/docs/durable-objects/api/alarms.mdx` and modifies the example in `src/content/docs/durable-objects/best-practices/rules-of-durable-objects.mdx` to demonstrate proper `try-catch` usage and alarm rescheduling. This **documentation enhancement** helps developers build more robust and reliable Durable Object applications by preventing silent alarm failures.

This change enhances diagnostic logging within the `workerd` runtime's sockets API, specifically adding a warning when Hyperdrive connection attempts fail due to an untriggered override. This improves observability for a specific class of connection issues without altering core connection functionality.

This is an isolated enhancement to `IO gate` tracing spans, adding new tags to improve observability and debugging for operations within the IO gate. There is no direct impact on functionality or downstream services.

Fixes a memory safety bug (use-after-move) within `workerd`'s `runAlarm` exception logging, preventing potential crashes or incorrect error reporting in the runtime. This is an isolated fix with no broader system impact.

This commit fixes an accuracy issue in `workerd`'s SQL API tracing, ensuring `rows_read` metrics are correctly reported in user traces. It also introduces system-level tracing for SQL queries, significantly improving observability and debugging capabilities for SQL operations.

This commit **improves the documentation** for **Durable Objects** by adding a clearer note on error handling best practices. It specifically highlights the critical need to **recreate `DurableObjectStub` instances after exceptions** to ensure proper recovery and prevent stale connections. Additionally, the documentation now provides a more explicit recommendation for implementing **exponential backoff** strategies when retrying operations. This **documentation update** in `src/content/docs/durable-objects/best-practices/error-handling.mdx` helps developers build more resilient applications by guiding them on correct stub management and retry mechanisms.

This commit **updates documentation examples** by **removing explicit `sql.end()` calls** from various **Hyperdrive** and **Workers PostgreSQL tutorials**. This **maintenance fix** addresses the fact that the Workers runtime automatically handles closing outgoing database connections, rendering `sql.end()` redundant in these contexts. Furthermore, removing these calls mitigates a potential issue where `sql.end()` is suspected to cause **hangs due to a bug in `postgres.js`**, thereby improving the reliability and clarity of database connection examples for users.

This commit **improves code readability** by adding detailed comments to the `QueueCustomEventImpl::run` method within `src/workerd/api/queue.c++`. These comments explicitly explain the intended flow and delineate the three main logical sections of the method. This **documentation enhancement** clarifies the complex event processing logic within the **`workerd` queue API**, making it significantly easier for developers to understand and maintain this critical component. The change is purely for **maintenance** and does not alter any functional behavior.

This commit **fixes a logic error** in the **Workerd Queue API**'s `waitUntil` mechanism, specifically within the `QueueCustomEventImpl::run` method. It corrects the deferral of the `drain()` operation, ensuring it is only called when `finishScheduled()` (which already handles `waitUntilTasks`) is not invoked, preventing redundant waiting. Furthermore, this **bug fix** properly attaches necessary state references for `waitUntilTasks` to be accessible from JavaScript. This improves the robustness of queue event processing and ensures correct `waitUntil` promise resolution.

This commit **fixes an error in the Durable Objects SQL storage documentation** by correcting an example for using `sql.exec()` with bound parameters. Specifically, it **removes an unnecessary array wrapping** around the parameters in the `src/content/docs/durable-objects/api/sql-storage.mdx` file. This **documentation fix** ensures that users following the example will now have correct code, preventing potential runtime issues when interacting with the **Durable Objects SQL storage API**. The change improves the accuracy and reliability of the **Durable Objects** documentation.

This commit introduces a **new capability** by implementing the `QueueConsumerNoWaitForWaitUntil` compatibility flag. This flag modifies the **queue event processing** logic in `src/workerd/api/queue.c++`, specifically within `startQueueEvent` and `QueueCustomEventImpl::run`. When enabled, **queue consumers will no longer block on `waitUntil`'ed work**, allowing for more flexible and potentially faster event handling by decoupling queue processing from asynchronous tasks. The flag's definition and documentation are also added to the `src/workerd/io/compatibility-date.capnp` schema.

This commit introduces a **new compatibility flag**, `queueConsumerNoWaitForWaitUntil`, within the `workerd` runtime to modify the behavior of **queue consumers**. This flag, added to the `CompatibilityFlags` structure in `src/workerd/io/compatibility-date.capnp`, is intended to allow queue handlers to resolve and acknowledge messages without waiting for all `waitUntil`'ed tasks to complete. The `QueueCustomEventImpl::run` method in `src/workerd/api/queue.c++` is updated to read this flag and conditionally adjust the `waitUntil` behavior for queue events. This is **preparatory work** for a future **new capability** that will prevent consumer invocations from being slowed down or failing due to slow or hung `waitUntil` tasks, thereby improving the **resilience and performance of queue processing**. The actual implementation of the non-blocking behavior will follow in a subsequent commit.

This commit **fixes broken links** within the **Hyperdrive documentation** by updating incorrect Cloudflare dashboard URLs. Specifically, it addresses the configuration links found in the `src/content/docs/hyperdrive/configuration/connect-to-postgres.mdx` and `src/content/docs/hyperdrive/configuration/rotate-credentials.mdx` guides. This **documentation maintenance** ensures that users following the **Hyperdrive configuration guides** are now directed to the correct Cloudflare dashboard pages. The **bug fix** significantly improves the accuracy and usability of the **Hyperdrive documentation**, preventing user frustration from encountering dead ends.

This commit **enhances error logging** within the **queue consumer event handler** in `src/workerd/api/queue.c++`. It extends the existing error reporting to specifically capture and log messages when a queue consumer `run` operation is **aborted**, in addition to existing timeout logging. This **observability improvement** is crucial for debugging unexpected consumer isolate timeouts and provides more specific diagnostic information. The change helps diagnose why queue processing might fail by logging the specific abort error message, making it easier to understand and resolve issues with queue consumer stability.