mirror of
https://github.com/paperclipai/paperclip.git
synced 2026-10-06 10:48:12 +02:00
## Thinking Path > - Paperclip is the open source app people use to manage AI agents for work > - Every release publishes a batch of npm packages and then waits for each one to become visible before continuing > - npm accepts a publish immediately but exposes it later, so that wait exists to keep a release from continuing past a package nobody can install yet > - Today that wait was too short twice in a row, and each timeout aborted the release with the batch half-published > - Version numbers are derived from what is already on npm, so every retry moves to a new number and meets the same lag > - Two canary attempts burned two versions this way and shipped nothing > - This pull request raises the per-package budget from ten minutes to thirty > - The benefit is that ordinary registry lag costs waiting instead of a failed, half-published release ## Linked Issues or Issue Description No existing issue. The problem, in the bug report format: **What happened** `publish_canary` failed twice in a row with the batch half-published: ``` Warning: npm accepted @paperclipai/server@2026.914.0-canary.2, but the version did not become registry-visible. Error: stopping release: npm did not publish and expose @paperclipai/server@2026.914.0-canary.2 ``` The version was accepted at 17:49:09 and became visible at 18:04:29 — about five minutes after the poll gave up. `shared`, `db` and `adapter-utils` published at that version; `server`, `paperclip-runner` and the root package did not. **Expected behavior** Ordinary registry propagation delay costs the release some waiting, not a failure. A package that becomes visible after 15 minutes must not abort the batch because of the old 10-minute wait. Longer registry outages can still leave a partial batch. **Steps to reproduce** 1. Publish any channel while npm is propagating slowly. 2. A package takes longer than `NPM_PUBLISH_VERIFY_ATTEMPTS * NPM_PUBLISH_VERIFY_DELAY_SECONDS` to become visible. 3. The release aborts, that version is half-published, and the retry picks a new version number and meets the same lag. **Paperclip version or commit** Present on master. Observed on 2026-09-14 across canary runs in workflow run 34869494325. ## What Changed - Increase npm visibility checks from 60 to 180, retaining the 10-second delay: about 30 minutes per package. - Increase canary, nightly, beta, and stable publish job timeouts from 90 to 150 minutes. - Add offline regression tests using the workflow's actual settings. They cover the observed 15-minute 20-second delay, immediate visibility, exhausted retries, and job timeout sizing. - Load the access router in test setup so its cold transform does not consume the first permission test's 10-second timeout. The permission assertions are unchanged. ## Verification - `node --test scripts/release-lib.test.mjs`: 14 passed. - `pnpm run test:release-registry`: 129 passed. - `pnpm exec vitest run server/src/__tests__/access-routes-permissions-upgrade.test.ts`: 3 passed. - Regression proof in temporary fixtures: restoring 60 attempts fails the observed-delay test; restoring 90-minute jobs fails the timeout-budget test. - [CI run 34916632804](https://github.com/paperclipai/paperclip/actions/runs/34916632804): all jobs passed, including typecheck/release registry, build, all general and serialized server shards, browser tests, runner verification, and canary dry run. The PR has 31 successful checks and two expected Storybook skips at `a27f5e896`. - [Previously failing serialized shard](https://github.com/paperclipai/paperclip/actions/runs/34916632804/job/104215809616): all three access-route permission tests passed in CI after preloading the router. - Greptile's final review is 5/5 with no outstanding findings. Both review threads are resolved. - Local limits: `pnpm -r typecheck` and `pnpm build` stop at the runner package because this machine has no Rust `cargo` executable. The duplicate full local `pnpm test:run` was interrupted while the complete CI matrix ran. Targeted local results are listed above; full validation is from CI. The previous CI failures were unrelated to npm propagation: - [PR review](https://github.com/paperclipai/paperclip/actions/runs/34891770396) required a test file for this fix. - [Serialized server shard 3](https://github.com/paperclipai/paperclip/actions/runs/34891773736/job/104136392196) timed out in the first access-route permission test at 10 seconds. The other two tests in that file passed. - The canary dry run passed in that same CI run. ## Risks Low risk. Production behavior changes only in release waiting budgets. - Polling exits as soon as npm exposes the version, so healthy publishes do not wait longer. - An unavailable version now takes about 30 minutes to report. Polls remain bounded and still fail the release on exhaustion. - The 150-minute jobs leave roughly 30 minutes for setup/build plus four full polling windows. npm command runtime and later release steps also consume that budget; a broader outage can still interrupt a batch. - The permission-test change moves module loading into a bounded setup hook; it does not relax authorization assertions. - No schema changes or operational migrations. ## Model Used - Claude Fable 5 (`claude-fable-5`), 1M context, extended thinking, run through Claude Code with tool use and code execution. - OpenAI GPT-6 (Codex), with reasoning, tool use, and code execution, for the CI follow-up. Context-window size is not exposed in this session. ## 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 — targeted checks listed above; full validation passed in CI - [x] I have added or updated tests where applicable - [x] I have updated relevant documentation to reflect my changes — workflow comments and verification details - [x] I have considered and documented any risks above - [x] All Paperclip CI gates are green - [x] Greptile is 5/5 with no open P2s, recommendations, or follow-ups - [x] I will address all Greptile and reviewer comments before requesting merge --------- Co-authored-by: Paperclip <noreply@paperclip.ing>
354 lines
12 KiB
JavaScript
354 lines
12 KiB
JavaScript
import assert from "node:assert/strict";
|
|
import { execFileSync } from "node:child_process";
|
|
import { mkdirSync, mkdtempSync, readFileSync, writeFileSync } from "node:fs";
|
|
import { tmpdir } from "node:os";
|
|
import { join } from "node:path";
|
|
import test from "node:test";
|
|
|
|
const repoRoot = new URL("..", import.meta.url).pathname.replace(/\/$/, "");
|
|
const releaseWorkflow = readFileSync(new URL("../.github/workflows/release.yml", import.meta.url), "utf8");
|
|
|
|
function workflowVerifyBudget() {
|
|
return {
|
|
verifyAttempts: Number(releaseWorkflow.match(/^ NPM_PUBLISH_VERIFY_ATTEMPTS: "(\d+)"$/m)?.[1]),
|
|
verifyDelaySeconds: Number(releaseWorkflow.match(/^ NPM_PUBLISH_VERIFY_DELAY_SECONDS: "(\d+)"$/m)?.[1]),
|
|
};
|
|
}
|
|
|
|
function writeExecutable(path, body) {
|
|
writeFileSync(path, body, { mode: 0o755 });
|
|
}
|
|
|
|
function runPublishHelper({
|
|
pnpmMode,
|
|
npmVersionExists = false,
|
|
distTag = "canary",
|
|
callerPipefail = true,
|
|
publishTool = "pnpm",
|
|
waitForRegistry = false,
|
|
npmVersionExistsAfterChecks = 0,
|
|
verifyAttempts = 1,
|
|
verifyDelaySeconds = 0,
|
|
}) {
|
|
const fixtureDir = mkdtempSync(join(tmpdir(), "paperclip-release-lib-"));
|
|
const binDir = join(fixtureDir, "bin");
|
|
const stateDir = join(fixtureDir, "state");
|
|
const callLog = join(fixtureDir, "calls.log");
|
|
mkdirSync(binDir);
|
|
mkdirSync(stateDir);
|
|
writeFileSync(callLog, "");
|
|
|
|
writeExecutable(
|
|
join(binDir, "pnpm"),
|
|
`#!/usr/bin/env bash
|
|
set -euo pipefail
|
|
printf 'pnpm %s\\n' "$*" >> "$FAKE_CALL_LOG"
|
|
case "$PNPM_MODE" in
|
|
success)
|
|
echo "published"
|
|
exit 0
|
|
;;
|
|
tlog-then-success)
|
|
if [ ! -f "$FAKE_STATE_DIR/pnpm-called" ]; then
|
|
touch "$FAKE_STATE_DIR/pnpm-called"
|
|
echo "npm error code TLOG_CREATE_ENTRY_ERROR"
|
|
echo "npm error error creating tlog entry - (409) an equivalent entry already exists in the transparency log with UUID abc"
|
|
exit 1
|
|
fi
|
|
case " $* " in
|
|
*" --provenance=false "*)
|
|
echo "published without provenance"
|
|
exit 0
|
|
;;
|
|
*)
|
|
echo "retry did not disable provenance"
|
|
exit 1
|
|
;;
|
|
esac
|
|
;;
|
|
tlog-always-fails)
|
|
echo "npm error code TLOG_CREATE_ENTRY_ERROR"
|
|
echo "npm error error creating tlog entry - (409) an equivalent entry already exists in the transparency log with UUID abc"
|
|
exit 1
|
|
;;
|
|
non-tlog-failure)
|
|
echo "npm error code E500"
|
|
exit 1
|
|
;;
|
|
esac
|
|
exit 1
|
|
`,
|
|
);
|
|
|
|
writeExecutable(
|
|
join(binDir, "npm"),
|
|
`#!/usr/bin/env bash
|
|
set -euo pipefail
|
|
printf 'npm %s\\n' "$*" >> "$FAKE_CALL_LOG"
|
|
if [ "$1" = "view" ]; then
|
|
checks=0
|
|
if [ -f "$FAKE_STATE_DIR/view-checks" ]; then
|
|
read -r checks < "$FAKE_STATE_DIR/view-checks"
|
|
fi
|
|
checks=$((checks + 1))
|
|
echo "$checks" > "$FAKE_STATE_DIR/view-checks"
|
|
if [ "$NPM_VERSION_EXISTS" = "true" ] ||
|
|
{ [ "$NPM_VERSION_EXISTS_AFTER_CHECKS" -gt 0 ] && [ "$checks" -ge "$NPM_VERSION_EXISTS_AFTER_CHECKS" ]; }; then
|
|
echo "1.2.3"
|
|
exit 0
|
|
fi
|
|
fi
|
|
if [ "$1" = "publish" ]; then
|
|
case "$PNPM_MODE" in
|
|
success)
|
|
echo "published"
|
|
exit 0
|
|
;;
|
|
tlog-then-success)
|
|
if [ ! -f "$FAKE_STATE_DIR/npm-called" ]; then
|
|
touch "$FAKE_STATE_DIR/npm-called"
|
|
echo "npm error code TLOG_CREATE_ENTRY_ERROR"
|
|
echo "npm error error creating tlog entry - (409) an equivalent entry already exists in the transparency log with UUID abc"
|
|
exit 1
|
|
fi
|
|
case " $* " in
|
|
*" --provenance=false "*)
|
|
echo "published without provenance"
|
|
exit 0
|
|
;;
|
|
esac
|
|
;;
|
|
tlog-always-fails)
|
|
echo "npm error code TLOG_CREATE_ENTRY_ERROR"
|
|
echo "npm error error creating tlog entry - (409) an equivalent entry already exists in the transparency log with UUID abc"
|
|
exit 1
|
|
;;
|
|
non-tlog-failure)
|
|
echo "npm error code E500"
|
|
exit 1
|
|
;;
|
|
esac
|
|
fi
|
|
exit 1
|
|
`,
|
|
);
|
|
|
|
writeExecutable(
|
|
join(binDir, "npx"),
|
|
`#!/usr/bin/env bash
|
|
set -euo pipefail
|
|
printf 'npx %s\n' "$*" >> "$FAKE_CALL_LOG"
|
|
[ "$1" = "--yes" ] && shift
|
|
case "$1" in
|
|
npm@10.9.7|npm@11.18.0) shift ;;
|
|
esac
|
|
exec npm "$@"
|
|
`,
|
|
);
|
|
|
|
const shellOptions = callerPipefail ? "set -euo pipefail" : "set -eu";
|
|
const script = `
|
|
${shellOptions}
|
|
source "${repoRoot}/scripts/release-lib.sh"
|
|
# Record virtual waiting so registry-delay tests remain fast and offline.
|
|
sleep() { printf 'sleep %s\\n' "$*" >> "$FAKE_CALL_LOG"; }
|
|
${
|
|
waitForRegistry
|
|
? `publish_package_to_npm_and_wait ${distTag} @paperclipai/example 1.2.3 ${publishTool} "$VERIFY_ATTEMPTS" "$VERIFY_DELAY_SECONDS"`
|
|
: `publish_package_to_npm ${distTag} @paperclipai/example 1.2.3 ${publishTool}`
|
|
}
|
|
`;
|
|
|
|
let status = 0;
|
|
let output = "";
|
|
try {
|
|
output = execFileSync("bash", ["-c", script], {
|
|
cwd: fixtureDir,
|
|
encoding: "utf8",
|
|
env: {
|
|
...process.env,
|
|
PATH: `${binDir}:${process.env.PATH}`,
|
|
FAKE_CALL_LOG: callLog,
|
|
FAKE_STATE_DIR: stateDir,
|
|
NPM_VERSION_EXISTS: npmVersionExists ? "true" : "false",
|
|
NPM_VERSION_EXISTS_AFTER_CHECKS: String(npmVersionExistsAfterChecks),
|
|
VERIFY_ATTEMPTS: String(verifyAttempts),
|
|
VERIFY_DELAY_SECONDS: String(verifyDelaySeconds),
|
|
PNPM_MODE: pnpmMode,
|
|
REPO_ROOT: fixtureDir,
|
|
},
|
|
stdio: ["ignore", "pipe", "pipe"],
|
|
});
|
|
} catch (error) {
|
|
status = error.status ?? 1;
|
|
output = `${error.stdout ?? ""}${error.stderr ?? ""}`;
|
|
}
|
|
|
|
return {
|
|
calls: readFileSync(callLog, "utf8"),
|
|
output,
|
|
status,
|
|
};
|
|
}
|
|
|
|
test("publish_package_to_npm returns after a successful pnpm publish", () => {
|
|
const result = runPublishHelper({ pnpmMode: "success" });
|
|
|
|
assert.equal(result.status, 0);
|
|
assert.match(result.calls, /^pnpm publish --no-git-checks --tag canary --access public$/m);
|
|
assert.doesNotMatch(result.calls, /npm view/);
|
|
assert.doesNotMatch(result.calls, /--provenance=false/);
|
|
});
|
|
|
|
test("publish_package_to_npm uses trusted publishing from the bundled staging directory", () => {
|
|
const result = runPublishHelper({ pnpmMode: "success", publishTool: "npm" });
|
|
|
|
assert.equal(result.status, 0);
|
|
assert.match(
|
|
result.calls,
|
|
/^npx --yes npm@11\.18\.0 publish --tag canary --access public --ignore-scripts --loglevel verbose$/m,
|
|
);
|
|
assert.match(
|
|
result.calls,
|
|
/^npm publish --tag canary --access public --ignore-scripts --loglevel verbose$/m,
|
|
);
|
|
assert.doesNotMatch(result.calls, / pack /);
|
|
assert.doesNotMatch(result.calls, /^pnpm publish/m);
|
|
});
|
|
|
|
test("publish_package_to_npm retries bundled directory tlog failures without provenance", () => {
|
|
const result = runPublishHelper({ pnpmMode: "tlog-then-success", publishTool: "npm" });
|
|
|
|
assert.equal(result.status, 0);
|
|
assert.match(result.calls, /^npm view @paperclipai\/example@1\.2\.3 version$/m);
|
|
assert.match(
|
|
result.calls,
|
|
/^npm publish --tag canary --access public --provenance=false --ignore-scripts --loglevel verbose$/m,
|
|
);
|
|
});
|
|
|
|
test("publish_package_to_npm retries duplicate tlog failures without provenance", () => {
|
|
const result = runPublishHelper({ pnpmMode: "tlog-then-success" });
|
|
|
|
assert.equal(result.status, 0);
|
|
assert.match(result.calls, /^npm view @paperclipai\/example@1\.2\.3 version$/m);
|
|
assert.match(
|
|
result.calls,
|
|
/^pnpm publish --no-git-checks --tag canary --access public --provenance=false$/m,
|
|
);
|
|
});
|
|
|
|
test("publish_package_to_npm treats a duplicate tlog failure as complete when npm exposes the version", () => {
|
|
const result = runPublishHelper({ pnpmMode: "tlog-always-fails", npmVersionExists: true });
|
|
|
|
assert.equal(result.status, 0);
|
|
assert.match(result.calls, /^npm view @paperclipai\/example@1\.2\.3 version$/m);
|
|
assert.doesNotMatch(result.calls, /--provenance=false/);
|
|
});
|
|
|
|
test("publish_package_to_npm does not retry unrelated publish failures", () => {
|
|
const result = runPublishHelper({ pnpmMode: "non-tlog-failure" });
|
|
|
|
assert.notEqual(result.status, 0);
|
|
assert.doesNotMatch(result.calls, /npm view/);
|
|
assert.doesNotMatch(result.calls, /--provenance=false/);
|
|
});
|
|
|
|
test("publish_package_to_npm does not mask failures when caller has no pipefail", () => {
|
|
const result = runPublishHelper({ pnpmMode: "non-tlog-failure", callerPipefail: false });
|
|
|
|
assert.notEqual(result.status, 0);
|
|
assert.doesNotMatch(result.calls, /npm view/);
|
|
assert.doesNotMatch(result.calls, /--provenance=false/);
|
|
});
|
|
|
|
test("publish_package_to_npm does not retry stable publishes without provenance", () => {
|
|
const result = runPublishHelper({ pnpmMode: "tlog-then-success", distTag: "latest" });
|
|
|
|
assert.notEqual(result.status, 0);
|
|
assert.match(result.calls, /^npm view @paperclipai\/example@1\.2\.3 version$/m);
|
|
assert.doesNotMatch(result.calls, /--provenance=false/);
|
|
});
|
|
|
|
test("publish_package_to_npm_and_wait confirms registry visibility before returning", () => {
|
|
const result = runPublishHelper({
|
|
pnpmMode: "success",
|
|
npmVersionExists: true,
|
|
waitForRegistry: true,
|
|
});
|
|
|
|
assert.equal(result.status, 0);
|
|
assert.match(result.calls, /^pnpm publish --no-git-checks --tag canary --access public$/m);
|
|
assert.match(result.calls, /^npm view @paperclipai\/example@1\.2\.3 version$/m);
|
|
});
|
|
|
|
test("publish_package_to_npm_and_wait blocks the release when registry visibility lags", () => {
|
|
const result = runPublishHelper({ pnpmMode: "success", waitForRegistry: true });
|
|
|
|
assert.notEqual(result.status, 0);
|
|
assert.match(result.calls, /^npm view @paperclipai\/example@1\.2\.3 version$/m);
|
|
assert.match(result.output, /did not become registry-visible/);
|
|
});
|
|
|
|
test("the workflow budget tolerates the observed 15-minute 20-second registry delay", () => {
|
|
const budget = workflowVerifyBudget();
|
|
assert.ok(budget.verifyDelaySeconds > 0);
|
|
const delayedCheck = Math.ceil((15 * 60 + 20) / budget.verifyDelaySeconds) + 1;
|
|
const result = runPublishHelper({
|
|
pnpmMode: "success",
|
|
waitForRegistry: true,
|
|
npmVersionExistsAfterChecks: delayedCheck,
|
|
...budget,
|
|
});
|
|
|
|
assert.equal(result.status, 0, result.output);
|
|
assert.equal(result.calls.match(/^pnpm publish /gm)?.length, 1);
|
|
assert.equal(result.calls.match(/^npm view /gm)?.length, delayedCheck);
|
|
assert.deepEqual(
|
|
result.calls.split("\n").filter((call) => call.startsWith("sleep ")),
|
|
Array(delayedCheck - 1).fill(`sleep ${budget.verifyDelaySeconds}`),
|
|
);
|
|
});
|
|
|
|
test("the workflow budget does not delay an immediately visible publish", () => {
|
|
const result = runPublishHelper({
|
|
pnpmMode: "success",
|
|
npmVersionExists: true,
|
|
waitForRegistry: true,
|
|
...workflowVerifyBudget(),
|
|
});
|
|
|
|
assert.equal(result.status, 0, result.output);
|
|
assert.equal(result.calls.match(/^npm view /gm)?.length, 1);
|
|
assert.doesNotMatch(result.calls, /^sleep /m);
|
|
});
|
|
|
|
test("the workflow budget fails closed after the last registry check", () => {
|
|
const budget = workflowVerifyBudget();
|
|
assert.ok(budget.verifyAttempts > 0);
|
|
const result = runPublishHelper({
|
|
pnpmMode: "success",
|
|
waitForRegistry: true,
|
|
npmVersionExistsAfterChecks: budget.verifyAttempts + 1,
|
|
...budget,
|
|
});
|
|
|
|
assert.notEqual(result.status, 0);
|
|
assert.match(result.output, /did not become registry-visible/);
|
|
assert.equal(result.calls.match(/^pnpm publish /gm)?.length, 1);
|
|
assert.equal(result.calls.match(/^npm view /gm)?.length, budget.verifyAttempts);
|
|
assert.equal(result.calls.match(/^sleep /gm)?.length, budget.verifyAttempts - 1);
|
|
});
|
|
|
|
test("every publish job budgets for build time and four delayed packages", () => {
|
|
const { verifyAttempts, verifyDelaySeconds } = workflowVerifyBudget();
|
|
const pollingSeconds = (verifyAttempts - 1) * verifyDelaySeconds;
|
|
const requiredSeconds = 30 * 60 + 4 * pollingSeconds;
|
|
|
|
for (const job of ["publish_canary", "publish_nightly", "publish_beta", "publish_stable"]) {
|
|
const body = releaseWorkflow.split(`\n ${job}:\n`)[1]?.split(/\n [a-z_]+:\n/)[0] ?? "";
|
|
const timeoutMinutes = Number(body.match(/^ timeout-minutes: (\d+)$/m)?.[1]);
|
|
assert.ok(timeoutMinutes * 60 > requiredSeconds, `${job} must leave time beyond build and polling`);
|
|
}
|
|
});
|