fix: backoff retry and timeouts in HNS restart

[rbtr]

#3529 refactored the HNS restart such that the stop/start could be retried - this adds a deadline to fail and cause that retry, and changes the retry strategy to exponential backoff instead of constant

[rbtr]

/azp run Azure Container Networking PR

[azure-pipelines]

Azure Pipelines successfully started running 1 pipeline(s).

[Copilot]

Pull Request Overview

This PR addresses backoff and retry improvements to enhance the stability of HNS restart operations on Windows.

• Introduces exponential backoff delay in retry operations for stopping and starting the HNS service.
• Refactors timeout creation in the tryStartServiceFn and tryStopServiceFn functions by using a deadline function.

Comments suppressed due to low confidence (4)

platform/os_windows.go:335

• [nitpick] The variables 'n' and 'limit' are ambiguous; consider renaming them to 'attemptCount' and 'maxAttempts' to improve clarity.

var n, limit time.Duration = 0, 3

platform/os_windows.go:355

• [nitpick] The variable 'deadline' shadows the 'deadline' function; consider renaming it (e.g., 'ctxWithTimeout') to avoid confusion.

deadline, cancel := deadline(ctx)

platform/os_windows.go:382

• [nitpick] Consider renaming 'n' and 'limit' to 'attemptCount' and 'maxAttempts' respectively for consistency and clarity in this context.

var n, limit time.Duration = 0, 3

platform/os_windows.go:402

• [nitpick] The reuse of the name 'deadline' causes shadowing of the deadline function; renaming it to something like 'ctxWithTimeout' could reduce ambiguity.

deadline, cancel := deadline(ctx)

[rbtr]

/azp run Azure Container Networking PR

[Copilot]

Pull Request Overview

This PR refactors the HNS restart process by introducing deadlines for service stop/start operations and using an exponential backoff retry strategy. Key changes include:

• Adding a delay type for exponential backoff in the retry calls.
• Implementing deadlines with a 90-second timeout for both stopping and starting the service.
• Modifying the error handling to use the deadline's signal instead of the original context cancellation.

Comments suppressed due to low confidence (1)

platform/os_windows.go:361

• The error wrap message 'context cancelled' may be misleading for a timeout scenario. Consider changing it to 'deadline exceeded' to better reflect a timeout condition.

case <-deadline.Done():

[azure-pipelines]

Azure Pipelines successfully started running 1 pipeline(s).

[QxBytes]

/azp run Azure Container Networking PR

[azure-pipelines]

Azure Pipelines successfully started running 1 pipeline(s).