fix(dgw): retry transient KDC connection drops

[Benoît Cortier (CBenoit)]

The KDC proxy path sometimes sees a connection accepted and then dropped before a reply is returned (suspected DC-side load: connection throttling or port exhaustion), surfacing to clients as a 502 Bad Gateway. Direct KDC exchanges are now retried up to 3 times on such transient failures, spaced with equal-jitter exponential backoff (max 1.2s / 2.4s / 4.8s) so a loaded KDC gets breathing room and concurrent retries do not synchronize into a storm. Permanent failures are still surfaced immediately, and the KDC target host is now shown at info level for easier diagnosis.

Issue: DGW-423

[Copilot]

Pull request overview

Adds bounded retries for transient direct KDC failures with exponential backoff and improved diagnostics.

Changes:

• Classifies transient and permanent KDC I/O failures.
• Retries direct exchanges with equal-jitter backoff.
• Adds supporting dependencies and helper tests.

Reviewed changes

Copilot reviewed 2 out of 3 changed files in this pull request and generated 3 comments.

File	Description
devolutions-gateway/src/kdc_connector.rs	Implements retry behavior, logging, classification, and tests.
devolutions-gateway/Cargo.toml	Adds retry and randomness dependencies.
Cargo.lock	Locks the new dependency graph.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

Pull request overview

Copilot reviewed 2 out of 3 changed files in this pull request and generated 2 comments.

Comments suppressed due to low confidence (1)

devolutions-gateway/src/kdc_connector.rs:147

• The retry tests verify attempt counts but not the configured 1.2/2.4/4.8-second delay maxima, so an accidental change to this base or factor would leave all tests passing while changing the main operational behavior. Please assert the elapsed paused Tokio time stays within the equal-jitter bounds, or expose the strategy and test each generated delay range.

        let backoff = ExponentialBackoff::from_millis(2)
            .factor(600)
            .max_delay(Duration::from_secs(10))
            .map(equal_jitter)
            .take(KDC_MAX_RETRIES);