Connect timeout propagated through pool

[mdaigle]

This pull request introduces improvements to connection timeout handling in the SQL client by switching from simple timeouts to a more robust TimeoutTimer system. It also adds the Microsoft.Bcl.TimeProvider dependency, and updates several method signatures to consistently propagate the timeout budget across connection creation and pooling logic. These changes help ensure that the total timeout budget is respected throughout the connection process, improving reliability and observability for connection attempts.

Dependency updates:

• Added Microsoft.Bcl.TimeProvider as a dependency in both the solution's package version list and the main project file to support advanced timeout handling. [1] [2] [3]

Timeout handling improvements:

• Replaced usages of raw timeouts with the TimeoutTimer class throughout the connection and pooling code, ensuring that the overall timeout budget is consistently tracked and enforced during connection attempts and retries. [1] [2] [3] [4] [5]
• Updated connection pool methods such as TryGetConnection and GetInternalConnection to accept and use a TimeoutTimer instead of a TimeSpan, and to derive cancellation tokens from the timer for accurate timeout enforcement. [1] [2]

API and signature changes:

• Updated method signatures for connection opening and replacement (e.g., TryOpenConnection, TryReplaceConnection, and related overrides) to include the TimeoutTimer parameter, ensuring the timeout is propagated through all relevant layers. [1] [2] [3] [4] [5]

Connection creation and pooling:

• Modified the connection pool's internal logic to pass the TimeoutTimer through to the physical connection factory, so that time spent waiting in the pool is deducted from the overall connection timeout budget. [1] [2] [3]

Code cleanup and comments:

• Added comments and TODOs to clarify timeout propagation and highlight areas for future improvement (e.g., passing cancellation tokens deeper into the connection factory). [1] [2]

These changes collectively improve the accuracy and reliability of connection timeouts, making the client's behavior more predictable under high load or slow network conditions.

[Copilot]

Pull request overview

Copilot reviewed 17 out of 17 changed files in this pull request and generated 2 comments.

Comments suppressed due to low confidence (1)

src/Microsoft.Data.SqlClient/src/Microsoft/Data/SqlClient/ConnectionPool/WaitHandleDbConnectionPool.cs:918

• The PendingGetConnection DueTime is still computed from CreationTimeout/ADP.TimerCurrent rather than the passed-in TimeoutTimer. For async opens this can schedule pending work beyond the caller’s remaining budget. Consider setting DueTime based on the TimeoutTimer (e.g., its ExpirationTicks or remaining duration) so WaitForPendingOpen computes delays that respect the overall timeout.

            var pendingGetConnection =
                new PendingGetConnection(
                    CreationTimeout == 0 ? Timeout.Infinite : ADP.TimerCurrent() + ADP.TimerFromSeconds(CreationTimeout / 1000),
                    owningObject,
                    taskCompletionSource,

[apoorvdeshmukh]

LGTM! Just wondering if we should create a separate issue to add any additional event tracing for scenarios where we throw exceptions to identify what caused it in the first place, but this is beyond the scope of this PR.

[Copilot]

Copilot encountered an error and was unable to review this pull request. You can try again by re-requesting a review.