mirror of
https://github.com/paperclipai/paperclip.git
synced 2026-11-10 01:50:13 +01:00
## Thinking Path > - Paperclip is the open source app people use to manage AI agents for work > - The restart chaos harness (`tests/restart-chaos/`) restarts a local control plane on purpose while agent tasks run, and grades every task on rules R1 to R6 > - The plan targets agent runs in Daytona sandboxes, but the harness only supported `--env local` > - A local controller needs two inputs that match the sandbox image exactly: the image's Linux `paperclip-runnerd` and its provider pack. The documented way to get them uses Docker, and Docker is not available on the operator machine > - The Daytona org that chaos uses also runs the live Cloud fleet, so the harness must never change a sandbox it did not create through its own Paperclip environment > - This pull request adds `--env daytona`. It extracts both inputs from the image without Docker, and refuses a controller whose runner source differs from the image. It seeds a Daytona environment from the runner-e2e fixtures and adds a read-only Daytona R6 sweep > - The benefit is that chaos waves can now cross real Daytona sandboxes with a local, killable control plane ## Linked Issues or Issue Description **Feature description** Run restart chaos waves with a local control plane and agent runs in Daytona sandboxes (plan `doc/plans/2026-11-06-control-plane-restart-chaos.md`, Phase 1b). **Problem it solves** `--env daytona` threw "not wired yet". The controller-side runnerd and provider pack had to come from `docker cp`, which needs a Docker daemon. **Proposed solution** Use `skopeo copy` (no daemon) to fetch the linux/amd64 image into a directory, walk the OCI layers from top to bottom, and extract only the two paths. Before the wave starts, compare the image's content-ID label with this checkout. Refs #19477 ## What Changed - `tests/restart-chaos/daytona-image.ts`: a streaming tar reader and an OCI layer walk (top to bottom, `.wh.<name>` and `.wh..wh..opq` whiteouts, hard links, symlinks written last, no `..` paths). It extracts `/usr/local/bin/paperclip-runnerd` and `/opt/paperclip-runner/provider-pack/`. It checks the layer blob digests, the ELF header (x86-64), the provider-pack manifest schema, and its `runnerSourceRevision` against the image revision label. It records digests in `receipt.json`, caches the result by image digest under `results/daytona-images/`, and deletes the layer blobs. It also runs standalone. - Controller/image check: `assertControllerMatchesImage` compares the image's `io.paperclip.runner.content-id` label with `computeDaytonaImageContentId` for this checkout. On a mismatch it fails and prints both IDs and the image's source revision. - `seed.ts`: for `--env daytona`, it installs the bundled Daytona plugin (`installFixtureDaytonaProvider`) and waits for provider readiness. It creates the `daytona` environment fixture from `catalog.ts` with a `DAYTONA_API_KEY` company secret and runs the native Codex platform-qualification probe. Then it assigns the agents and the project to that environment. `--daytona-cpu` sets the sandbox vCPU (default 4, 4 GiB, 10 GiB). - `daytona-sweep.ts`: a read-only R6 for Daytona. The Daytona client is a frozen wrapper that exposes only a label-filtered `list`, and no stop, archive or delete exists. The sweep keeps only rows whose `paperclip-company-id` and `paperclip-environment-id` both equal the wave's own UUIDs, and it re-checks every row. It reconciles sandboxes with the environment's leases: - A sandbox archived after its lease was released is expected. - An orphaned sandbox or an `active`/`pending_cleanup` lease fails R6 for the owning task, or the wave. - `assertHarnessMayMutate` refuses `paperclip-create-*` names and `fleet-sbx-*` snapshots. - Cleanup only deletes the wave's environment through Paperclip and reports what is still running. - `config.ts`: `--daytona-image` (or `PAPERCLIP_E2E_DAYTONA_IMAGE`) needs an immutable digest. `--concurrency × --daytona-cpu` must stay within `RESTART_CHAOS_DAYTONA_VCPU_BUDGET` (default 40). B5 is local-only, because its oracle reads the live workspace. - `main.ts`: Daytona preflight (credentials, image extraction, content-ID check, Daytona plugin build as CI does). It passes `PAPERCLIP_RUNNER_REMOTE_BINARY_PATH` and `PAPERCLIP_RUNNER_REMOTE_PROVIDER_PACK_PATH` to every server start and writes `sandbox-sweep.json`. Cleanup errors never skip stopping the controller. - README: a Daytona section (image matching, image-only CI dispatch, vCPU budget, fleet safety, R6). - Unit tests: synthetic tar layers for the tar reader, whiteouts, opaque directories, overrides, hard links, symlinks and parent replacement. Further tests cover the content-ID check, sweep reconciliation, exact wave selection (unlabelled rows and other labels are never selected), the read-only client, the fleet guard, and the CLI options. ## Verification - `pnpm exec vitest run --config tests/restart-chaos/vitest.config.ts`: 62/62 pass, credential-free. - `node scripts/typecheck-test-projects.mjs tests/restart-chaos`: clean. - Real extraction without Docker, image `…@sha256:b0c71f4d…`: 42 layers, about 8 minutes, a 3 GB pack. - The content-ID check refused the current checkout. The image is `b8b16e15…` from `7e6d882558`; the checkout is `aec45589…`. - A checkout of `7e6d882558` computes `b8b16e15…`, so the check compares like with like. - Current image: `ghcr.io/paperclipai/paperclip-daytona-runner@sha256:a84579522192de0393ae24d09e67c4e752df4c701bd315ff54ad9bbca42e2dc0` (source `5d25117ffa`, content `4e9a221f…`). It matches this branch, and the receipt shows the content ID match. - Paid waves, `--env daytona --adapters codex,claude --tasks 3 --concurrency 2 --faults K1 --seed 11`: - Wave 1 (10 vCPU org, 4 vCPU sandboxes): 0/3. Every create returned HTTP 400 "Total CPU limit exceeded. Maximum allowed: 10", reproduced directly with the SDK. This led to the vCPU budget, and the archived-sandbox fix in R6. - Wave 2 (same org, 2 vCPU sandboxes): **3/3 PASS**, all restart-crossing. Codex completed through one working Retry; both Claude tasks continued automatically. R6: 4 sandboxes archived after release, 0 orphans, 0 held leases. - Wave 3 (company org, 4 vCPU sandboxes): 0/3. Two product failures were found and reported to the campaign ledger: an ACPX Claude recovery blocked on `ambiguous_remote_recovery_evidence` with no Retry, and a false budget hard stop at 10.75¢ of 3000¢ that paused the company. The harness changed no sandbox. R6 reported 2 held leases, and the environment delete returned 409 `pending_sandbox_cleanup` because of them. - Fleet-org key check: one sandbox with a unique non-`paperclip-create` name and label was created, then deleted by the exact ID that create returned. ## Risks - This change touches the test harness only, with no product code. - The extracted provider pack is about 3 GB per image in the gitignored results directory. Delete old `results/daytona-images/<digest>` directories by hand. - The harness never changes Daytona sandboxes. If Paperclip's environment cleanup is blocked, sandboxes stay until their 60-minute ttl or a person acts. The harness reports them in `sandbox-sweep.json` and fails the wave. > For core feature work, check [`ROADMAP.md`](ROADMAP.md) first and discuss it in `#dev` before opening the PR. Feature PRs that overlap with planned core work may need to be redirected — check the roadmap first. See `CONTRIBUTING.md`. ## Model Used - Claude Opus 5.5 (`claude-opus-5-5`), Anthropic, in Claude Code with tool use (shell, file edits, GitHub CLI). Default context window, extended thinking. ## Checklist - [x] I have included a thinking path that traces from project context to this change - [x] I have specified the model used (with version and capability details) - [x] I have checked ROADMAP.md and confirmed this PR does not duplicate planned core work - [x] I have searched GitHub for duplicate or related PRs and linked them above - [x] I have either (a) linked existing issues with `Fixes: #` / `Closes #` / `Refs #` OR (b) described the issue in-PR following the relevant issue template - [x] I have not referenced internal/instance-local Paperclip issues or links (only public GitHub `#NNN` / `github.com/paperclipai/paperclip` URLs) - [x] My branch name describes the change (e.g. `docs/...`, `fix/...`) and contains no internal Paperclip ticket id or instance-derived details - [x] I have run tests locally and they pass - [x] I have added or updated tests where applicable - [x] I have updated relevant documentation to reflect my changes - [x] I have considered and documented any risks above - [ ] All Paperclip CI gates are green - [ ] Greptile is 5/5 with no open P2s, recommendations, or follow-ups - [x] I will address all Greptile and reviewer comments before requesting merge 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
204 lines
11 KiB
TypeScript
204 lines
11 KiB
TypeScript
import type { FaultReceipt, OracleResult, RuleId, RuleResult, TaskGrade, TaskRecord, TaskSample } from "./types.js";
|
||
|
||
/** Pure R1–R6 grading over observer samples (plan §2). No I/O. */
|
||
|
||
export const ACTIVE_RUN_STATUSES = new Set(["queued", "running", "scheduled_retry"]);
|
||
export const STALL_LIMIT_MS = 3 * 60_000;
|
||
const REST_STATUSES_WITH_WORK = new Set(["todo", "in_progress"]);
|
||
|
||
const ms = (iso: string | null | undefined) => (iso ? Date.parse(iso) : NaN);
|
||
|
||
function faultWindow(fault: FaultReceipt): [number, number] {
|
||
return [fault.startedAt, fault.readyAt ?? fault.healthyAt ?? Number.POSITIVE_INFINITY];
|
||
}
|
||
|
||
/**
|
||
* Restart-crossing: some run of the task (queued wake through finalization)
|
||
* overlaps the fault window, or the task was not yet at rest when it started.
|
||
*/
|
||
export function crossingFaults(record: TaskRecord, faults: readonly FaultReceipt[]): string[] {
|
||
if (record.createdAt === null) return [];
|
||
const latest = new Map<string, { run: NonNullable<TaskSample["runs"]>[number]; at: number }>();
|
||
for (const sample of record.samples) for (const run of sample.runs ?? []) latest.set(run.id, { run, at: sample.at });
|
||
const runs = new Map<string, { start: number; end: number }>();
|
||
for (const [id, { run, at }] of latest) {
|
||
const start = ms(run.createdAt) || ms(run.startedAt) || at;
|
||
const end = ACTIVE_RUN_STATUSES.has(run.status) ? Number.POSITIVE_INFINITY : (ms(run.finishedAt) || at);
|
||
runs.set(id, { start, end });
|
||
}
|
||
const restAt = record.restAt ?? Number.POSITIVE_INFINITY;
|
||
return faults.filter(fault => {
|
||
const [start, end] = faultWindow(fault);
|
||
if (record.createdAt! <= start && restAt > start) return true;
|
||
return [...runs.values()].some(run => run.start <= end && run.end >= start);
|
||
}).map(fault => fault.faultId);
|
||
}
|
||
|
||
function reachable(samples: readonly TaskSample[]) { return samples.filter(sample => sample.reachable); }
|
||
|
||
function lastReachable(record: TaskRecord): TaskSample | undefined {
|
||
const values = reachable(record.samples);
|
||
return values[values.length - 1];
|
||
}
|
||
|
||
export function outageOverlapMs(record: TaskRecord, faults: readonly FaultReceipt[]): number {
|
||
if (record.createdAt === null) return 0;
|
||
const end = record.restAt ?? Number.POSITIVE_INFINITY;
|
||
return faults.reduce((total, fault) => {
|
||
const [start, stop] = faultWindow(fault);
|
||
const overlap = Math.min(stop, end) - Math.max(start, record.createdAt!);
|
||
return total + (Number.isFinite(overlap) && overlap > 0 ? overlap : 0);
|
||
}, 0);
|
||
}
|
||
|
||
export function gradeR1(record: TaskRecord, budgetMs: number, faults: readonly FaultReceipt[]): RuleResult {
|
||
if (record.launchError) return { verdict: "fail", detail: `launch_failed: ${record.launchError}` };
|
||
if (record.retry) return { verdict: "n/a", detail: "task rested needing Retry; graded under R3" };
|
||
if (record.finalStatus !== "done") {
|
||
return { verdict: "fail", detail: record.timedOut ? `no_rest_within_budget (last status ${record.finalStatus ?? "unknown"})` : `rested_not_done (status ${record.finalStatus ?? "unknown"}, no Retry offered)` };
|
||
}
|
||
const allowed = budgetMs + outageOverlapMs(record, faults);
|
||
const took = (record.restAt ?? Number.POSITIVE_INFINITY) - record.createdAt!;
|
||
return took <= allowed
|
||
? { verdict: "pass", detail: `done after ${Math.round(took / 1000)}s (allowed ${Math.round(allowed / 1000)}s)` }
|
||
: { verdict: "fail", detail: `done_too_late ${Math.round(took / 1000)}s > ${Math.round(allowed / 1000)}s` };
|
||
}
|
||
|
||
export function gradeR2(record: TaskRecord, oracle: OracleResult): RuleResult {
|
||
const last = lastReachable(record);
|
||
if (!last) return { verdict: "fail", detail: "no_reachable_sample" };
|
||
if (last.interruption?.workspaceFailure) return { verdict: "fail", detail: `workspace_failure_visible_at_rest: ${last.interruption.workspaceFailure}` };
|
||
if (record.preFaultSnapshotError) {
|
||
return { verdict: "fail", detail: `pre_fault_snapshot_failed: durability not checked (${record.preFaultSnapshotError})` };
|
||
}
|
||
if (record.restSnapshotError) {
|
||
return { verdict: "fail", detail: `rest_snapshot_failed: durability not checked (${record.restSnapshotError})` };
|
||
}
|
||
if (record.lostFiles?.length) {
|
||
return { verdict: "fail", detail: `pre_fault_files_lost: ${record.lostFiles.slice(0, 10).join(",")}` };
|
||
}
|
||
const fileChecks = oracle.checks.filter(check => /file|workspace_located/.test(check.name) && !check.ok);
|
||
if (fileChecks.length && record.finalStatus === "done") {
|
||
return { verdict: "fail", detail: `workspace_files_missing_after_done: ${fileChecks.map(check => check.name).join(",")}` };
|
||
}
|
||
const transient = record.samples.some(sample => sample.interruption?.workspaceFailure);
|
||
return { verdict: "pass", detail: transient ? "workspace failure appeared during recovery and cleared" : "no workspace failure at rest" };
|
||
}
|
||
|
||
export function gradeR3(record: TaskRecord): RuleResult {
|
||
const last = lastReachable(record);
|
||
if (!record.retry) {
|
||
if (record.finalStatus !== "done" && !record.timedOut && !record.launchError) {
|
||
return { verdict: "fail", detail: `missing_retry: rested ${record.finalStatus ?? "unknown"} with canRetry=${String(last?.interruption?.canRetry ?? false)}` };
|
||
}
|
||
return { verdict: "n/a", detail: "no Retry needed" };
|
||
}
|
||
if (record.retry.error || record.retry.accepted !== true) {
|
||
return { verdict: "fail", detail: `retry_rejected: http=${record.retry.httpStatus ?? "none"} accepted=${String(record.retry.accepted)} ${record.retry.error ?? ""}`.trim() };
|
||
}
|
||
const retryAt = record.retry.at;
|
||
const newRun = record.samples.some(sample => (sample.runs ?? []).some(run => (ms(run.createdAt) || 0) >= retryAt - 1_000));
|
||
if (!newRun) return { verdict: "fail", detail: "retry_no_new_run" };
|
||
if (record.finalStatus !== "done") return { verdict: "fail", detail: `retry_did_not_complete (status ${record.finalStatus ?? "unknown"}${record.timedOut ? ", timed out" : ""})` };
|
||
return { verdict: "pass", detail: "one Retry started a run that completed the task" };
|
||
}
|
||
|
||
/**
|
||
* R4: at most one active run (queued, running or scheduled_retry) per issue in
|
||
* any sample, and no comment consumed twice. A wake comment may be reused only
|
||
* by a run that follows a run which did not succeed (an explicit retry or
|
||
* continuation after a failure); reuse after a successful run is a replay.
|
||
*/
|
||
export function gradeR4(record: TaskRecord): RuleResult {
|
||
for (const sample of reachable(record.samples)) {
|
||
const active = (sample.runs ?? []).filter(run => ACTIVE_RUN_STATUSES.has(run.status));
|
||
if (active.length > 1) return { verdict: "fail", detail: `concurrent_runs: ${active.map(run => `${run.id}(${run.status})`).join(",")}` };
|
||
}
|
||
const finalRuns = new Map<string, NonNullable<TaskSample["runs"]>[number]>();
|
||
for (const sample of reachable(record.samples)) for (const run of sample.runs ?? []) finalRuns.set(run.id, run);
|
||
const begun = [...finalRuns.values()].filter(run => run.startedAt)
|
||
.sort((a, b) => (ms(a.createdAt) || 0) - (ms(b.createdAt) || 0));
|
||
const consumers = new Map<string, typeof begun>();
|
||
for (const run of begun) for (const commentId of run.wakeCommentIds) consumers.set(commentId, [...(consumers.get(commentId) ?? []), run]);
|
||
for (const [commentId, runs] of consumers) {
|
||
const replayed = runs.findIndex((_run, index) => index > 0 && runs.slice(0, index).some(prior => prior.status === "succeeded"));
|
||
if (replayed > 0) return { verdict: "fail", detail: `comment_consumed_twice: ${commentId} by ${runs.slice(0, replayed + 1).map(run => run.id).join(",")}` };
|
||
}
|
||
return { verdict: "pass", detail: "at most one active run per sample; no comment consumed after a successful run" };
|
||
}
|
||
|
||
function namedWait(sample: TaskSample): boolean {
|
||
const interruption = sample.interruption;
|
||
return Boolean(sample.executionBlocker || sample.scheduledRetry || sample.activeRecoveryAction || interruption?.retryWait
|
||
|| interruption?.state === "retry_scheduled" || interruption?.canRetry
|
||
|| (interruption && ["failed", "waiting_for_user"].includes(interruption.state)));
|
||
}
|
||
|
||
export function isStalledSample(sample: TaskSample): boolean {
|
||
if (!sample.reachable || !sample.serverReady) return false;
|
||
if (!REST_STATUSES_WITH_WORK.has(sample.issueStatus ?? "")) return false;
|
||
if ((sample.runs ?? []).some(run => ACTIVE_RUN_STATUSES.has(run.status))) return false;
|
||
return !namedWait(sample);
|
||
}
|
||
|
||
export function deadRunningRun(sample: TaskSample, readySince: number): { id: string; idleMs: number } | null {
|
||
if (!sample.reachable || !sample.serverReady || sample.at - readySince <= STALL_LIMIT_MS) return null;
|
||
for (const run of sample.runs ?? []) {
|
||
if (run.status !== "running") continue;
|
||
const evidence = Math.max(ms(run.lastEvidenceAt) || 0, ms(run.startedAt) || 0, readySince);
|
||
if (sample.at - evidence > STALL_LIMIT_MS) return { id: run.id, idleMs: sample.at - Math.max(ms(run.lastEvidenceAt) || 0, ms(run.startedAt) || 0) };
|
||
}
|
||
return null;
|
||
}
|
||
|
||
export function gradeR5(record: TaskRecord): RuleResult {
|
||
let stallStart: number | null = null;
|
||
let worst = 0;
|
||
let readySince: number | null = null;
|
||
for (const sample of record.samples) {
|
||
if (!sample.reachable || !sample.serverReady) { stallStart = null; readySince = null; continue; }
|
||
readySince ??= sample.at;
|
||
const dead = deadRunningRun(sample, readySince);
|
||
if (dead) return { verdict: "fail", detail: `silent_dead_run ${dead.id} running with no event, activity or start evidence for ${Math.round(dead.idleMs / 1000)}s` };
|
||
if (isStalledSample(sample)) {
|
||
stallStart ??= sample.at;
|
||
worst = Math.max(worst, sample.at - stallStart);
|
||
if (sample.at - stallStart > STALL_LIMIT_MS) {
|
||
return { verdict: "fail", detail: `silent_stall ${Math.round((sample.at - stallStart) / 1000)}s in ${sample.issueStatus} with no live run, wake or wait reason` };
|
||
}
|
||
} else stallStart = null;
|
||
}
|
||
return { verdict: "pass", detail: worst ? `longest unexplained idle ${Math.round(worst / 1000)}s` : "no unexplained idle" };
|
||
}
|
||
|
||
export function gradeR6(record: TaskRecord): RuleResult {
|
||
const sweep = record.sweep;
|
||
if (!sweep) return { verdict: "fail", detail: "rest_sweep_missing" };
|
||
if (sweep.error) return { verdict: "fail", detail: `rest_sweep_error: ${sweep.error}` };
|
||
const leaks = [
|
||
...(sweep.activeRuns.length ? [`active_runs=${sweep.activeRuns.join(",")}`] : []),
|
||
...(sweep.runningWorkspaceOperations.length ? [`running_workspace_operations=${sweep.runningWorkspaceOperations.join(",")}`] : []),
|
||
...(sweep.activeRecoveryAction ? ["active_recovery_action"] : []),
|
||
...(sweep.daytonaLeaks ?? []),
|
||
];
|
||
return leaks.length ? { verdict: "fail", detail: `leak: ${leaks.join("; ")}` } : { verdict: "pass", detail: "clean rest" };
|
||
}
|
||
|
||
export function gradeTask(record: TaskRecord, faults: readonly FaultReceipt[], budgetMs: number, oracle: OracleResult): TaskGrade {
|
||
const crossing = crossingFaults(record, faults);
|
||
const rules: Record<RuleId, RuleResult> = {
|
||
R1: gradeR1(record, budgetMs, faults),
|
||
R2: gradeR2(record, oracle),
|
||
R3: gradeR3(record),
|
||
R4: gradeR4(record),
|
||
R5: gradeR5(record),
|
||
R6: gradeR6(record),
|
||
};
|
||
const failed = (Object.entries(rules) as Array<[RuleId, RuleResult]>).find(([, rule]) => rule.verdict === "fail");
|
||
const pass = !failed && oracle.verdict === "pass";
|
||
let signature: string | null = null;
|
||
if (failed) signature = `${failed[0]}:${failed[1].detail.split(/[\s:(]/)[0]}:${record.plan.workload}:${record.plan.fault}`;
|
||
else if (oracle.verdict === "fail") signature = `T:${oracle.checks.find(check => !check.ok)?.name ?? "unknown"}:${record.plan.workload}:${record.plan.fault}`;
|
||
return { crossed: crossing.length > 0, crossingFaults: crossing, rules, T: oracle, pass, signature };
|
||
}
|