Skip to content

Driver

Drive one node through a sandbox end to end: stage its inputs, run it, collect its outputs, and map the raw exit into a typed outcome. drive_node is the high-level entry; drive_node_raw exposes the un-mapped result; run_node_in_sandbox is the low-level single-shot runner.

resoluto.sandbox.drive_node async 

drive_node(
    runtime,
    store,
    spec,
    *,
    admit=None,
    on_event=None,
    poll_interval_s=2.0,
    dead_after_s=120.0,
    clock=monotonic,
)

Drive one node and read its work product from result.json as a NodeResult.

Source code in src/resoluto/sandbox/driver.py 
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
async def drive_node(
    runtime: SandboxRuntime,
    store: Conduit,
    spec: SandboxLaunchSpec,
    *,
    admit: Admission | None = None,
    on_event: OnEvent | None = None,
    poll_interval_s: float = 2.0,
    dead_after_s: float = 120.0,
    clock: Callable[[], float] = time.monotonic,
) -> NodeResult:
    """Drive one node and read its work product from result.json as a `NodeResult`."""
    node_id = spec.labels.get("resoluto.node_id", "")
    outcome = await drive_node_raw(
        runtime,
        store,
        spec,
        admit=admit,
        on_event=on_event,
        poll_interval_s=poll_interval_s,
        dead_after_s=dead_after_s,
        clock=clock,
    )
    if outcome.disposition != "completed":
        return NodeResult(
            node_id=node_id,
            status="failure",
            observed_phase=outcome.observed_phase,
            reason=outcome.reason,
            substrate_logs=outcome.substrate_logs,
        )
    try:
        raw = await store.get(result_key(spec.store_prefix))
    except (ConduitError, OSError):
        return NodeResult(
            node_id=node_id,
            status="failure",
            observed_phase=outcome.observed_phase,
            reason="no result.json in store",
        )
    try:
        result = NodeResult.model_validate_json(raw)
    except ValidationError as e:
        return NodeResult(
            node_id=node_id,
            status="failure",
            observed_phase=outcome.observed_phase,
            reason=f"result.json failed to parse: {e.error_count()} validation error(s)",
        )
    result.observed_phase = outcome.observed_phase
    return result

resoluto.sandbox.drive_node_raw async 

drive_node_raw(
    runtime,
    store,
    spec,
    *,
    admit=None,
    on_event=None,
    result_ready=None,
    poll_interval_s=2.0,
    dead_after_s=120.0,
    unstartable_polls=15,
    external_gone_polls=15,
    clock=monotonic,
)

Launch, tail telemetry, and reap; returns a NodeOutcome. Optional result_ready completes as soon as the work product appears.

Source code in src/resoluto/sandbox/driver.py 
 79
 80
 81
 82
 83
 84
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
async def drive_node_raw(
    runtime: SandboxRuntime,
    store: Conduit,
    spec: SandboxLaunchSpec,
    *,
    admit: Admission | None = None,
    on_event: OnEvent | None = None,
    result_ready: Callable[[], Awaitable[bool]] | None = None,
    poll_interval_s: float = 2.0,
    dead_after_s: float = 120.0,
    unstartable_polls: int = 15,
    external_gone_polls: int = 15,
    clock: Callable[[], float] = time.monotonic,
) -> NodeOutcome:
    """Launch, tail telemetry, and reap; returns a `NodeOutcome`. Optional `result_ready` completes as soon as the work product appears."""
    lease_cm = (await admit.acquire(spec)) if admit is not None else _direct_lease(runtime, spec)
    async with lease_cm as leased:
        handle = leased.handle
        reader = ChunkReader(store, spec.store_prefix, dead_after_s=dead_after_s, clock=clock)
        phase = "unknown"
        unknown_streak = 0
        unstartable_streak = 0
        while True:
            for ev in await reader.poll():
                await _fire(on_event, ev)
            if result_ready is not None and await result_ready():
                return NodeOutcome(disposition="completed", observed_phase=phase)
            st = await runtime.status(handle)
            phase = st.phase
            if phase == "running":
                reader.arm()
            if st.terminal:
                for ev in await reader.poll():
                    await _fire(on_event, ev)
                return NodeOutcome(disposition="completed", observed_phase=phase)
            unstartable_streak = (
                unstartable_streak + 1
                if (phase != "running" and st.reason in _FATAL_WAITING)
                else 0
            )
            if unstartable_streak >= unstartable_polls:
                return NodeOutcome(
                    disposition="unstartable",
                    observed_phase=phase,
                    reason=f"{st.reason} (sustained {unstartable_streak} polls)",
                )
            unknown_streak = unknown_streak + 1 if phase == "unknown" else 0
            if unknown_streak >= external_gone_polls and reader.is_dead():
                return NodeOutcome(
                    disposition="external",
                    observed_phase=phase,
                    reason="pod terminated externally (sustained 'unknown' + telemetry silence)",
                )
            if reader.is_dead():
                try:
                    logs = await runtime.logs(handle)
                except Exception:  # noqa: BLE001
                    logs = "(unavailable)"
                return NodeOutcome(
                    disposition="silent",
                    observed_phase=phase,
                    reason="substrate dead — no telemetry within death window",
                    substrate_logs=logs[-4000:],
                )
            await asyncio.sleep(poll_interval_s)

resoluto.sandbox.NodeOutcome dataclass 

NodeOutcome(
    disposition,
    observed_phase,
    reason="",
    substrate_logs="",
)

Substrate-level disposition of one driven node: 'completed', 'unstartable', 'external', or 'silent'.

resoluto.sandbox.NodeResult

Bases: BaseModel

Typed work product the in-sandbox runner writes to <prefix>/result.json.

resoluto.sandbox.run_node_in_sandbox async 

run_node_in_sandbox(
    *,
    store,
    prefix,
    run_id,
    node_id,
    workload_argv,
    workspace_dir=None,
    output_paths=None,
    setup_argv=None,
    cleanup_argv=None,
    heartbeat_interval_s=5.0,
    clock=time,
    canary_probe_host="1.1.1.1",
    canary_probe_port=80,
    run_canary=None,
)

Run one node's workload (with optional setup/cleanup hooks and input/output staging), self-report telemetry to the store, and return the NodeResult (also written to <prefix>/result.json). run_canary overrides the egress-isolation canary (tests inject a stub); the canary always runs.

Source code in src/resoluto/sandbox/runner.py 
 62
 63
 64
 65
 66
 67
 68
 69
 70
 71
 72
 73
 74
 75
 76
 77
 78
 79
 80
 81
 82
 83
 84
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
async def run_node_in_sandbox(
    *,
    store: Conduit,
    prefix: str,
    run_id: str,
    node_id: str,
    workload_argv: list[str],
    workspace_dir: str | None = None,
    output_paths: list[str] | None = None,
    setup_argv: list[str] | None = None,
    cleanup_argv: list[str] | None = None,
    heartbeat_interval_s: float = 5.0,
    clock: Callable[[], float] = time.time,
    canary_probe_host: str = "1.1.1.1",
    canary_probe_port: int = 80,
    run_canary: CanaryRunner | None = None,
) -> NodeResult:
    """Run one node's workload (with optional setup/cleanup hooks and input/output staging), self-report telemetry to the store, and return the NodeResult (also written to `<prefix>/result.json`). `run_canary` overrides the egress-isolation canary (tests inject a stub); the canary always runs."""
    shipper = ChunkShipper(store, prefix, clock=clock)
    em = SpanEmitter(shipper, run_id, clock=clock)
    hb = asyncio.ensure_future(_heartbeat(shipper, heartbeat_interval_s))
    result = NodeResult(node_id=node_id)
    canary = run_canary or _default_canary(canary_probe_host, canary_probe_port)
    try:
        async with em.span("", "node", node_id, inputs={"argv": workload_argv}) as node_sid:
            async with em.span(node_sid, "egress_canary", "egress_canary") as canary_sid:
                verdict = await canary(store, prefix)
                for r in verdict.results:
                    await em.log(
                        canary_sid,
                        f"probe {r.target}: passed={r.passed} "
                        f"(expected_reachable={r.expected_reachable}, actual={r.actual_reachable})",
                    )
                canary_ok = verdict.passed
                if not verdict.passed:
                    result.status = "failure"
                    result.reason = verdict.reason

            if canary_ok:
                if workspace_dir is not None:
                    Path(workspace_dir).mkdir(parents=True, exist_ok=True)
                    staged = await stage_inputs(store, prefix, workspace_dir)
                    await em.log(
                        node_sid, f"staged {len(staged)} input archive(s) → {workspace_dir}"
                    )
                setup_ok = True
                if setup_argv:
                    src = await _exec_logged(
                        em, node_sid, "setup", "setup", setup_argv, workspace_dir
                    )
                    if src != 0:
                        result.exit_code, result.status, setup_ok = src, "failure", False
                        await em.log(
                            node_sid, f"setup hook failed (exit {src}) — skipping workload"
                        )
                if setup_ok:
                    rc = await _exec_logged(
                        em, node_sid, "workload", node_id, workload_argv, workspace_dir
                    )
                    result.exit_code = rc
                    result.status = "success" if rc == 0 else "failure"
                    if rc == 0 and workspace_dir is not None and output_paths:
                        result.output_archive = await collect_outputs(
                            store, prefix, workspace_dir, output_paths
                        )
                        await em.log(node_sid, f"collected outputs → {result.output_archive}")
    finally:
        if cleanup_argv:
            try:
                await _exec_logged(em, "", "cleanup", "cleanup", cleanup_argv, workspace_dir)
            except Exception:  # noqa: BLE001
                pass
        hb.cancel()
        await store.put(result_key(prefix), result.model_dump_json().encode("utf-8"))
        await shipper.close()
    return result