Add Agent Host hung turn lifecycle diagnostics

[roblourens]

Summary

Adds bounded lifecycle attribution to the existing agentHost.turnHung event so provider-stage hangs can be assigned to an owning boundary without adding telemetry for healthy turns.

• records provider diagnostic availability and Copilot send/fleet call state
• records whether the Copilot SDK turn started and whether its session is active, disconnecting, disconnected, or shut down
• records whether the initiating AHP client remains connected
• derives process-wide client liveness from authoritative ProtocolServerHandler records rather than mirrored transport state
• keeps missing chats/turns and diagnostic failures explicit and bounded

Motivation

On the complete 2026-08-20 UTC day, Copilot noProgress hangs affected 83 of 3,271 Linux Agents Window SSH messages (2.5375%) and 63 of 2,648 WSL messages (2.3792%), versus 311 of 104,282 local Windows Agents Window messages (0.2982%). In the preceding complete-day cohort, only 6.897% of SSH and 3.333% of WSL provider-stage noProgress turns completed within 24 hours, compared with 75.362% for local Windows Agents Window turns.

Current telemetry cannot distinguish a pending provider call, a resolved/rejected call with no provider turn start, a missing provider chat/turn, a dead SDK session, or initiating-client loss. This change adds that attribution while preserving existing timeout, cancellation, and recovery behavior.

Validation

• 15,120 full node unit tests passed in final review
• 1,023 directly affected Agent Host tests passed after the final changes
• npm run typecheck-client
• pre-commit hygiene hook

(Written by Copilot)

[Copilot]

Pull request overview

Adds bounded provider and client-liveness diagnostics to agentHost.turnHung.

Changes:

• Tracks Copilot call, turn, and session lifecycle state.
• Replaces mirrored connection tracking with a process-wide service.
• Expands lifecycle and telemetry tests.

Show a summary per file

File	Description
src/vs/platform/agentHost/common/agent.ts	Defines diagnostic contracts.
src/vs/platform/agentHost/node/agentHostBootstrap.ts	Registers the connection service.
src/vs/platform/agentHost/node/agentHostClientConnectionService.ts	Aggregates authoritative client connections.
src/vs/platform/agentHost/node/agentHostClientConnectionTelemetry.ts	Removes the superseded tracker.
src/vs/platform/agentHost/node/agentHostMain.ts	Uses service-based tracking.
src/vs/platform/agentHost/node/agentHostServerMain.ts	Removes manual tracker wiring.
src/vs/platform/agentHost/node/agentHostTelemetryReporter.ts	Emits new hang attributes.
src/vs/platform/agentHost/node/agentHostTurnTracker.ts	Collects lifecycle and liveness snapshots.
src/vs/platform/agentHost/node/agentSideEffects.ts	Preserves agent and initiator context.
src/vs/platform/agentHost/node/copilot/copilotAgent.ts	Exposes turn diagnostics.
src/vs/platform/agentHost/node/copilot/copilotAgentSession.ts	Tracks provider calls and turn starts.
src/vs/platform/agentHost/node/copilot/copilotSessionWrapper.ts	Tracks session lifecycle.
src/vs/platform/agentHost/node/protocolServerHandler.ts	Supplies authoritative connection records.
src/vs/platform/agentHost/test/node/agentHostTelemetryReporter.test.ts	Verifies telemetry fields.
src/vs/platform/agentHost/test/node/agentHostToolCallTelemetry.test.ts	Updates test DI.
src/vs/platform/agentHost/test/node/agentHostTurnHangTelemetry.test.ts	Tests hang diagnostics.
src/vs/platform/agentHost/test/node/agentHostTurnTelemetry.test.ts	Updates test DI.
src/vs/platform/agentHost/test/node/agentServiceTestUtils.ts	Wires the service in tests.
src/vs/platform/agentHost/test/node/agentSideEffects.test.ts	Updates side-effect test setup.
src/vs/platform/agentHost/test/node/copilotAgentSession.test.ts	Tests provider lifecycle states.
src/vs/platform/agentHost/test/node/mockAgent.ts	Supports mock diagnostics.
src/vs/platform/agentHost/test/node/protocolServerHandler.test.ts	Tests connection aggregation and rollback.

Review details

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

• Files reviewed: 22/22 changed files
• Comments generated: 3
• Review effort level: Balanced