Files
PaperClipAI/ui/src/context/LiveUpdatesProvider.mention-pool.test.ts
T
DottaandClaude Opus 5.5 2a2d3f6e23 fix(ui): stop refetching the @task mention pool on every task event (#19931)
## Thinking Path

> - Paperclip is the open source app people use to manage AI agents for
work
> - The board UI keeps its caches fresh from live events in
`LiveUpdatesProvider`
> - Each task activity event invalidates the whole `["issues",
companyId]` query prefix
> - That prefix also holds the `@task` mention pool: the 100 most
recently updated tasks, about 315 KB per response
> - A chat turn emits many comment, status and run events, so an open
task page downloads that list again for almost every event
> - This pull request keeps the pool fresh from push events without a
refetch per event: a closed server flag marks real membership or title
changes, progress on pooled tasks reorders the cache in place, and other
progress refetches at most once per 30 seconds
> - The benefit is less bandwidth and CPU for users with an open task or
chat, and smaller eval traces

## Linked Issues or Issue Description

**What happened**
While a task chat page is open, the browser repeats `GET
/api/companies/:id/issues?limit=100&sortField=updated&sortDir=desc`
(about 315 KB each). One 10-minute chat-eval attempt recorded about 194
of these requests. The trigger is `invalidateActivityQueries` in
`ui/src/context/LiveUpdatesProvider.tsx`. For every `entityType:
"issue"` activity event, it invalidates
`queryKeys.issues.list(companyId)`. That prefix matches
`queryKeys.issues.mentionPool(companyId)`, which `IssueDetail` (and
`NewIssueDialog`) keep active for the `@task` picker. Invalidation
refetches active queries and ignores `staleTime`.

**Expected behavior**
Comments, run progress and status changes on any task do not download
the full mention pool again. The pool still updates when a task that the
picker can show is created, deleted or renamed.

**Steps to reproduce**
1. Run `pnpm dev` and create a company with about 140 tasks.
2. Open a task page, for example `/REF/issues/REF-1`.
3. Every 4 seconds for 15 rounds, post comments on the open task, on a
task in the top 100, and on a task outside the top 100, and change the
open task's priority (60 events).
4. Count requests to `issues?limit=100&sortField=updated`. Before this
change there are 15 (about 3.3 MB). After this change there are 2 (about
480 KB).

## What Changed

- New shared helper `issueActivityChangesTaskIdentity`
(`packages/shared/src/issue-activity-identity.ts`). It is true for a
created, deleted, hidden, unhidden or renamed task.
- `server/src/realtime/live-events-ws.ts`: task `activity.logged` events
now carry the closed boolean `taskIdentityChanged`. Authenticated
sockets still get no `details`; the flag carries no task content.
- `ui/src/context/LiveUpdatesProvider.tsx` refreshes the mention pool
this way:
- The task identity flag is true (or full details show a create, delete,
hide, unhide or rename): the pool refetches at once.
- Other events on a task already in the pool: the task moves to the
front of the cached list. There is no request. The freshness clock does
not change, and a pending invalidation is never overwritten.
- Other events on a task outside the pool: one trailing refetch at most
every 30 seconds, because the task can now be in the top 100.
  - The picker is closed: the pool is only marked stale.
- `ui/src/lib/query-invalidation-batcher.ts`: new `excludingQueryKeys()`
keyed exclusion predicate. The batcher coalesces it and uses it for
prefix coverage. The broad `["issues", companyId]` invalidation uses it
to skip the pool, so a burst is again one broad refresh per batch.
- Tests: `ui/src/context/LiveUpdatesProvider.mention-pool.test.ts`
(authenticated event shape, hide and unhide, the 30-second window,
reorder safety, a burst through the batcher), new batcher tests,
`packages/shared/src/issue-activity-identity.test.ts`, and socket
assertions in `server/src/__tests__/live-events-toast.test.ts`.

## Verification

- `cd ui && npx vitest run src/context/LiveUpdatesProvider
src/lib/query-invalidation-batcher`: 231 tests pass.
- `cd server && npx vitest run src/__tests__/live-events-toast.test.ts`:
11 tests pass. In authenticated mode the socket keeps `details` off and
sends `taskIdentityChanged` (false for routine updates, true for a
create, rename or hide).
- `cd packages/shared && npx vitest run
src/issue-activity-identity.test.ts`: 14 tests pass.
- The new UI tests fail against the first commit of this PR: the
authenticated burst, hide, unhide, the 30-second window and the batcher
burst.
- `tsc --noEmit` for ui, server and shared: no new errors.
- Browser measurement: a local `pnpm dev` instance (local trusted),
headless Chromium on a task page, 141 tasks, 60 events over about 63
seconds (the reproduction steps above):
  - Before (master UI): **15** mention-pool requests, about 3.3 MB.
  - After: **2** requests, about 480 KB (one per 30-second window).
- Freshness after the fix: one rename, hide, unhide or create each
caused **1** refetch. A priority change caused **0**.
- I did not run a browser session in authenticated mode. The
authenticated event shape is covered by the socket test and the UI unit
tests.

## Risks

- A task that moves into the top 100 can be missing from the `@task`
picker for up to 30 seconds.
- The cached order is approximate between refetches: a moved task goes
to the front, but its `updatedAt` value is not changed.
- Other queries under `["issues", companyId]` (descendants,
created-from, summaries, onboarding first task) still refresh for each
event batch. They are small, but they can get the same treatment later.

## Model Used

- Anthropic Claude Opus 5.5 (`claude-opus-5-5`), with tool use and code
execution in Claude Code.

## 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>
2026-11-07 20:55:39 -06:00

175 lines
7.2 KiB
TypeScript

// @vitest-environment node
import { afterEach, beforeEach, describe, expect, it, vi } from "vitest";
import { QueryClient, QueryObserver } from "@tanstack/react-query";
import type { Issue } from "@paperclipai/shared";
import { __liveUpdatesTestUtils } from "./LiveUpdatesProvider";
import { createCoalescingQueryClient, createInvalidationBatcher } from "../lib/query-invalidation-batcher";
import { queryKeys } from "../lib/queryKeys";
const COMPANY = "company-1";
const ACTOR = { userId: "user-1", agentId: null };
const POOL_KEY = queryKeys.issues.mentionPool(COMPANY);
const WINDOW_MS = __liveUpdatesTestUtils.MENTION_POOL_REFRESH_WINDOW_MS;
function task(id: string): Issue {
return { id, identifier: id.toUpperCase(), title: `Task ${id}` } as Issue;
}
function mountQuery(client: QueryClient, queryKey: readonly unknown[], data: unknown[] = []) {
const queryFn = vi.fn(async () => data);
const observer = new QueryObserver(client, { queryKey, queryFn, staleTime: 60_000 });
const unsubscribe = observer.subscribe(() => {});
return { queryFn, unsubscribe };
}
async function settle(ms = 0) {
await vi.advanceTimersByTimeAsync(ms);
}
/** Authenticated sockets carry no `details`, only the closed identity flag. */
function authenticatedEvent(entityId: string, action: string, taskIdentityChanged: boolean) {
return { entityType: "issue", entityId, action, actorType: "agent", actorId: "agent-1", taskIdentityChanged };
}
function emit(client: QueryClient, payload: Record<string, unknown>) {
__liveUpdatesTestUtils.invalidateActivityQueries(client, COMPANY, { entityType: "issue", ...payload }, ACTOR);
}
const POOL = [task("t1"), task("t2"), task("t3")];
describe("LiveUpdatesProvider @task mention pool", () => {
let client: QueryClient;
beforeEach(() => {
vi.useFakeTimers();
client = new QueryClient();
});
afterEach(() => {
client.clear();
vi.useRealTimers();
});
it("does not refetch for a burst of authenticated progress events on pooled tasks, and reorders in place", async () => {
const pool = mountQuery(client, POOL_KEY, POOL);
const otherList = mountQuery(client, [...queryKeys.issues.list(COMPANY), "descendants", "t1"]);
await settle();
expect(pool.queryFn).toHaveBeenCalledTimes(1);
for (let i = 0; i < 25; i += 1) {
emit(client, authenticatedEvent("t3", "issue.comment_added", false));
emit(client, authenticatedEvent("t3", "issue.updated", false));
emit(client, authenticatedEvent("t2", "issue.checked_out", false));
await settle();
}
await settle(WINDOW_MS * 2);
expect(pool.queryFn).toHaveBeenCalledTimes(1);
expect(client.getQueryData<Issue[]>(POOL_KEY)?.map((issue) => issue.id)).toEqual(["t2", "t3", "t1"]);
// Other company-scoped issue queries keep their live refresh.
expect(otherList.queryFn.mock.calls.length).toBeGreaterThan(1);
pool.unsubscribe();
otherList.unsubscribe();
});
it.each([
["an authenticated identity change", authenticatedEvent("t9", "issue.updated", true)],
["a created task", { entityId: "t9", action: "issue.created", details: { identifier: "T9" } }],
["a deleted task", { entityId: "t1", action: "issue.deleted", details: {} }],
["a renamed task", { entityId: "t1", action: "issue.updated", details: { title: "Renamed" } }],
["a hidden task", { entityId: "t1", action: "issue.updated", details: { hiddenAt: "2026-11-07T00:00:00.000Z" } }],
["an unhidden task", { entityId: "t1", action: "issue.updated", details: { hiddenAt: null } }],
])("refetches immediately for %s", async (_label, payload) => {
const pool = mountQuery(client, POOL_KEY, POOL);
await settle();
emit(client, payload);
await settle();
expect(pool.queryFn).toHaveBeenCalledTimes(2);
pool.unsubscribe();
});
it("refetches at most once per window when events touch tasks outside the pool", async () => {
const pool = mountQuery(client, POOL_KEY, POOL);
await settle();
for (let i = 0; i < 20; i += 1) {
emit(client, authenticatedEvent(`outside-${i}`, "issue.comment_added", false));
await settle(100);
}
expect(pool.queryFn).toHaveBeenCalledTimes(1);
// The trailing refresh lands one window after the last fetch.
await settle(WINDOW_MS);
expect(pool.queryFn).toHaveBeenCalledTimes(2);
// Quiet afterwards: no further refetch without new events.
await settle(WINDOW_MS * 2);
expect(pool.queryFn).toHaveBeenCalledTimes(2);
pool.unsubscribe();
});
it("only marks an unmounted pool stale", async () => {
const pool = mountQuery(client, POOL_KEY, POOL);
await settle();
pool.unsubscribe();
emit(client, authenticatedEvent("outside", "issue.updated", false));
await settle(WINDOW_MS * 2);
expect(pool.queryFn).toHaveBeenCalledTimes(1);
expect(client.getQueryState(POOL_KEY)?.isInvalidated).toBe(true);
});
it("keeps a pending refresh when a later event reorders a closed picker's pool", async () => {
const pool = mountQuery(client, POOL_KEY, POOL);
await settle();
pool.unsubscribe();
const fetchedAt = client.getQueryState(POOL_KEY)?.dataUpdatedAt;
// A rename while the picker is closed marks the pool stale without a fetch.
emit(client, { entityId: "t1", action: "issue.updated", details: { title: "Renamed" } });
emit(client, authenticatedEvent("t3", "issue.comment_added", false));
await settle();
const state = client.getQueryState(POOL_KEY);
expect(state?.isInvalidated).toBe(true);
expect(state?.dataUpdatedAt).toBe(fetchedAt);
// Reopening the picker within staleTime still fetches the renamed list.
const reopened = mountQuery(client, POOL_KEY, POOL);
await settle();
expect(reopened.queryFn).toHaveBeenCalledTimes(1);
reopened.unsubscribe();
});
it("does not reset the freshness clock when it reorders in place", async () => {
const pool = mountQuery(client, POOL_KEY, POOL);
await settle();
const fetchedAt = client.getQueryState(POOL_KEY)?.dataUpdatedAt;
await settle(5_000);
emit(client, authenticatedEvent("t3", "issue.comment_added", false));
await settle();
expect(client.getQueryData<Issue[]>(POOL_KEY)?.[0]?.id).toBe("t3");
expect(client.getQueryState(POOL_KEY)?.dataUpdatedAt).toBe(fetchedAt);
pool.unsubscribe();
});
it("coalesces a burst through the live-event batcher into one broad refresh", async () => {
const batcher = createInvalidationBatcher(client);
const coalescing = createCoalescingQueryClient(client, batcher);
const invalidate = vi.spyOn(client, "invalidateQueries");
const pool = mountQuery(client, POOL_KEY, POOL);
const otherList = mountQuery(client, [...queryKeys.issues.list(COMPANY), "descendants", "t1"]);
await settle();
for (let i = 0; i < 30; i += 1) {
__liveUpdatesTestUtils.invalidateActivityQueries(coalescing, COMPANY, authenticatedEvent("t1", "issue.comment_added", false), ACTOR);
}
await settle(1_000);
const broad = invalidate.mock.calls.filter(([filters]) =>
JSON.stringify(filters?.queryKey) === JSON.stringify(queryKeys.issues.list(COMPANY)));
expect(broad).toHaveLength(1);
expect(otherList.queryFn).toHaveBeenCalledTimes(2);
expect(pool.queryFn).toHaveBeenCalledTimes(1);
batcher.dispose();
pool.unsubscribe();
otherList.unsubscribe();
});
});