Public dry run dryrun-a43e907efd9d

Source commit: a43e907efd9d1561c23fe73499478e881f868355

Public tree identity: sha256:453dea30195485dd8939f575a69b39f4bb7acd84c4df23f8aa29b55d044f2673

scripts/node-lifecycle-contract-smoke.js

@@ -0,0 +1,155 @@
+#!/usr/bin/env node
+
+const assert = require("assert");
+const fs = require("fs");
+const path = require("path");
+
+const repo = path.resolve(__dirname, "..");
+
+function read(relativePath) {
+  return fs.readFileSync(path.join(repo, relativePath), "utf8");
+}
+
+function expect(source, name, pattern) {
+  assert.match(source, pattern, `missing node lifecycle evidence: ${name}`);
+}
+
+const nodeMain = read("crates/clusterflux-node/src/daemon.rs");
+const nodeIdentity = read("crates/clusterflux-node/src/node_identity.rs");
+const cliNode = read("crates/clusterflux-cli/src/node.rs");
+const nodeTaskReports = read("crates/clusterflux-node/src/task_reports.rs");
+const nodeDebugAgent = read("crates/clusterflux-node/src/debug_agent.rs");
+const nodeLib = read("crates/clusterflux-node/src/lib.rs");
+const sharedWasmtimeRuntime = `${read("crates/clusterflux-wasm-runtime/src/lib.rs")}\n${read("crates/clusterflux-wasm-runtime/src/task_host_linker.rs")}`;
+const nodeRuntimeSurface = `${nodeLib}\n${sharedWasmtimeRuntime}`;
+const nodeLifecycleSurface = `${nodeMain}\n${nodeIdentity}\n${nodeTaskReports}\n${nodeDebugAgent}`;
+const nodeAssignmentRunner = `${read("crates/clusterflux-node/src/assignment_runner.rs")}\n${read("crates/clusterflux-node/src/assignment_runner/control_watcher.rs")}\n${read("crates/clusterflux-node/src/assignment_runner/process_runner.rs")}\n${read("crates/clusterflux-node/src/assignment_runner/validation.rs")}`;
+const coordinatorCore = read("crates/clusterflux-coordinator/src/lib.rs");
+const coordinatorService = `${read("crates/clusterflux-coordinator/src/service.rs")}\n${read("crates/clusterflux-coordinator/src/service/routing.rs")}`;
+const coordinatorServiceTests = read("crates/clusterflux-coordinator/src/service/tests.rs");
+const coordinatorServiceSurface = `${coordinatorService}\n${coordinatorServiceTests}`;
+const wasmtimeAssignmentSmoke = read("scripts/wasmtime-assignment-smoke.js");
+const wasmtimeSmoke = read("scripts/wasmtime-node-smoke.js");
+const debugCore = read("crates/clusterflux-core/src/debug.rs");
+
+assert.strictEqual(
+  (nodeMain.match(/CoordinatorSession::connect/g) || []).length,
+  1,
+  "node runtime should open one coordinator session in the local process-boundary runtime"
+);
+
+for (const [name, pattern] of [
+  ["enrollment exchange over session", /"type": "exchange_node_enrollment_grant"/],
+  ["persisted node identity is reused locally", /"type": "node_identity_reused"/],
+  ["heartbeat over session", /"type": "node_heartbeat"/],
+  ["node-originated requests use signed envelope", /"type": "signed_node"/],
+  ["capability report over session", /"type": "report_node_capabilities"/],
+  ["task assignment polling over session", /"type": "poll_task_assignment"/],
+  ["process start over session", /"type": "start_process"/],
+  ["reconnect over session", /"type": "reconnect_node"/],
+  ["debug command polling over session", /"type": "poll_debug_command"/],
+  ["log event over session", /"type": "report_task_log"/],
+  ["VFS metadata over session", /"type": "report_vfs_metadata"/],
+  ["task control polling over session", /"type": "poll_task_control"/],
+  ["completion over session", /"type": "task_completed"/],
+  ["cancellation uses same session", /poll_task_cancellation\(session, args, &task, node_private_key\)/],
+  ["request count is reported", /session\.requests\(\)/],
+]) {
+  expect(nodeLifecycleSurface, name, pattern);
+}
+
+expect(cliNode, "user-authorized node attach over Client session", /"type": "attach_node"/);
+assert.doesNotMatch(
+  nodeIdentity,
+  /"type": "attach_node"/,
+  "a persisted node must authenticate with its signed identity instead of replaying Client attach"
+);
+
+for (const [name, pattern] of [
+  ["worker uses enrollment exchange", /workerReady\.node_status, "ready"/],
+  ["worker receives the task instance selected by the SDK runtime", /nodeRun\.task_assignment_response\.task,[\s\S]*sdkRun\.task_spec\.task_instance/],
+  ["worker verifies task ABI", /task_assignment_response\.task_spec\.dispatch\.abi/],
+  ["worker records task completion", /nodeRun\.coordinator_response\.type, "task_recorded"/],
+  ["worker exposes task result", /events\.events\[0\]\.result, \{ SmallJson: 42 \}/],
+  ["Wasm task observes cooperative cancellation", /task_definition: "cooperative_cancellation_probe"[\s\S]*type: "cancel_process"/],
+  ["cooperative cancellation returns under task control", /cancellationNodeRun\.terminal_state, "completed"/],
+  ["running command abort is exercised", /task_definition: "abort_probe"[\s\S]*type: "abort_process"/],
+  ["running command abort reaches terminal state", /abortedNodeRun\.terminal_state, "cancelled"/],
+  ["running native command receives a real Debug Epoch freeze", /waitForNativeSleep[\s\S]*type: "create_debug_epoch"[\s\S]*fully_frozen/],
+  ["native Linux process is observed stopped and resumed", /assert\.match\(procState\(nativeSleepPid\)[\s\S]*stopped[\s\S]*resume_debug_epoch[\s\S]*fully_resumed[\s\S]*assert\.doesNotMatch\(procState\(nativeSleepPid\)/],
+  ["parent and child Wasm tasks are frozen as one process", /task_definition: "debug_parent_probe"[\s\S]*debugChildInstance = "debug_parent_probe-1:child:1"[\s\S]*all-stop across active parent and child Wasm participants[\s\S]*affected_tasks\.length, 2[\s\S]*acknowledgements\.length, 2[\s\S]*\[debugChildInstance, "debug_parent_probe-1"\]/],
+  ["parent and child Wasm tasks both acknowledge resume", /multiParticipantResume[\s\S]*fully_resumed[\s\S]*acknowledgements\.length, 2/],
+  ["abort releases process slot", /afterAbort\.processes, \[\]/],
+]) {
+  expect(wasmtimeAssignmentSmoke, name, pattern);
+}
+
+for (const [name, pattern] of [
+  ["cooperative cancellation and abort are distinct", /cancel_requested: false,[\s\S]*abort_requested: true/],
+  ["controlled runner polls abort while command runs", /fn abort_requested[\s\S]*poll_task_control/],
+  ["controlled runner creates a process group", /process\.process_group\(0\)/],
+  ["controlled runner freezes the native process group", /libc::kill\(process_group, libc::SIGSTOP\)/],
+  ["controlled runner resumes the native process group", /libc::kill\(process_group, libc::SIGCONT\)/],
+  ["controlled runner kills the process group", /libc::kill\(process_group, libc::SIGKILL\)/],
+  ["Wasm code can poll cooperative cancellation", /task_control_v1/],
+  ["matched Wasm probes remain at a quiescent boundary", /TaskHostOperation::DebugProbe[\s\S]*enter_quiescent_host_boundary[\s\S]*leave_quiescent_host_boundary/],
+  ["debug snapshots use the live Wasm task handle registry", /debug_handle_snapshot[\s\S]*task_handle_\{handle_id\}[\s\S]*state=active/],
+  ["native command status comes from the controlled runner", /set_command_status[\s\S]*frozen native command pid[\s\S]*native command exited with status/],
+]) {
+  expect(`${coordinatorServiceSurface}\n${nodeLifecycleSurface}\n${sharedWasmtimeRuntime}\n${nodeAssignmentRunner}`, name, pattern);
+}
+
+for (const [name, pattern] of [
+  ["coordinator rejects stale process ownership", /fn node_reconnect_rejects_stale_process_epoch_after_restart\(\)/],
+  ["reconnect preserves enrolled node identity", /reconnect_node\(&NodeId::from\("node"\), None\)/],
+  ["stale process epoch is rejected", /CoordinatorError::StaleProcessEpoch/],
+]) {
+  expect(coordinatorCore, name, pattern);
+}
+
+for (const [name, pattern] of [
+  ["coordinator delivers cancellation to connected node", /fn service_delivers_cancellation_to_connected_node_and_records_terminal_state\(\)/],
+  ["node polls task control", /CoordinatorRequest::PollTaskControl/],
+  ["cancelled terminal state is recorded", /TaskTerminalState::Cancelled/],
+]) {
+  expect(coordinatorServiceSurface, name, pattern);
+}
+
+for (const [name, pattern] of [
+  ["native lifecycle test exists", /fn linux_task_lifecycle_supports_cancel_and_all_stop_freeze_resume\(\)/],
+  ["native freeze succeeds when supported", /lifecycle\.freeze_for_debug_epoch\(\)\.unwrap\(\)/],
+  ["native resume succeeds", /lifecycle\.resume_after_debug_epoch\(\)/],
+  ["native cancel reaches lifecycle", /lifecycle\.cancel\(\)/],
+  ["unsupported freeze errors", /BackendError::DebugFreezeUnsupported/],
+  ["wasmtime runtime exposes freeze resume probe", /pub fn freeze_resume_i32_export_probe/],
+  ["wasmtime runtime captures Wasm frame locals", /debug_i32_export_snapshot[\s\S]*local_values/],
+  ["wasmtime runtime creates Wasm debug participant", /kind: DebugParticipantKind::WasmTask/],
+  ["wasmtime debug participant carries local values", /local_values: snapshot\.local_values\.clone\(\)/],
+  ["wasmtime runtime resumes after freeze", /epoch\.continue_all\(\)/],
+]) {
+  expect(nodeRuntimeSurface, name, pattern);
+}
+
+for (const [name, pattern] of [
+  ["wasmtime smoke runs debug freeze resume mode", /--debug-freeze-resume/],
+  ["wasmtime smoke verifies frozen state", /debugReport\.frozen_state, "Frozen"/],
+  ["wasmtime smoke verifies resumed state", /debugReport\.resumed_state, "Running"/],
+  ["wasmtime smoke verifies frame local values", /debugReport\.local_values[\s\S]*wasm_local_0/],
+  ["wasmtime smoke proves node runtime reached wasm task", /node_runtime_reached_wasm_task/],
+  ["wasmtime smoke proves node captured locals", /node_runtime_captured_wasm_locals/],
+]) {
+  expect(wasmtimeSmoke, name, pattern);
+}
+
+for (const [name, pattern] of [
+  ["debug model freezes wasm and command participants", /fn breakpoint_creates_all_stop_debug_epoch_for_wasm_and_command_tasks\(\)/],
+  ["debug model rejects unsupported freeze", /fn debug_epoch_reports_failure_when_no_participant_can_freeze\(\)/],
+  ["debug model resumes frozen participants", /fn continue_resumes_every_frozen_participant\(\)/],
+  ["debug model includes captured locals", /local_values/],
+  ["wasm participants are modeled", /DebugParticipantKind::WasmTask/],
+  ["controlled native command participants are modeled", /DebugParticipantKind::ControlledNativeCommand/],
+]) {
+  expect(debugCore, name, pattern);
+}
+
+console.log("Node lifecycle contract smoke passed");