perf(detail): cache + parallelize + skip redundant work in the detail page handler

[priosshrsth]

Ref: OUT-3701 · Stacked on #1215.

Changes

• getOneTask — parallelize assignee Copilot fetch with task.update / task.count (was sequential after). Skip task.update entirely when the body has no signed-URL changes. Include workflowState in the initial findFirst so no second query is needed when update is skipped.
• tasks.controller.getTask — drop now-redundant getTaskAssignee call.
• page.tsx — replace 5 HTTP-loopback SSR fetches with direct service-method calls. New loaders.ts sibling for clarity. OneTaskDataFetcher gets useFallback so the post-hydration client refire of /api/tasks/{id} is eliminated.

Why

The previous SSR pass fanned out to 5 separate Vercel function invocations, each with its own authenticate() and react.cache scope — so Copilot calls couldn't dedup across loaders. Direct service calls collapse this to 1 invocation, 1 auth, 1 cache scope; the react.cache wraps from #1215 then make dedup actually work end-to-end.

Measured impact (staging, time-to-task-body p50)

	Baseline	With #1215 only	With this PR
getInternalUser:self actual fetches	4	1	1
Time-to-task-body	~7.5 s	~6.6 s	~4.4 s

Numbers are from local-dev measurements against staging Supabase. Real prod (Vercel us-east-1, colocated with DB) will land lower since these are bottlenecked by laptop→us-east-1 RTT.

Test plan

• IU on task with IU assignee — title/body render, edits work
• IU on task with client assignee — assignee picker populates
• Client on a task — read-only view works
• Image-heavy task — signed URLs still refresh (exercises the bodyChanged path, not the skip-update path)
• Preview mode — header renders correctly
• Vercel preview measurement to validate prod numbers

Out of scope

• Apply OUT-3702 pattern to WorkflowStateFetcher + TemplatesFetcher — separate ticket
• Suspense-wrap secondary detail sections — separate ticket
• Path-traversal query merge — blocked on Prisma Unsupported(\"ltree\")

🤖 Generated with Claude Code

[linear-code]

OUT-3701

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
tasks-app	Ready	Preview, Comment	May 13, 2026 10:25am

[vercel]

Deployment failed with the following error:

Deploying Serverless Functions to multiple regions is restricted to the Pro and Enterprise plans.

Learn More: https://vercel.link/multiple-function-regions

[greptile-apps]

Greptile Summary

This PR optimises the task detail page by collapsing 5 HTTP loopback SSR fetches into direct service calls (single auth, single react.cache scope), parallelising subtask count + Copilot assignee fetch + conditional body update inside getOneTask, and surfacing SSR task data immediately to the client via SWR fallbackData.

• getOneTask refactor — includes workflowState in the initial findFirst, skips task.update when no image-src changes are detected, and runs the DB update, subtask count, and assignee fetch concurrently in a single Promise.all.
• loaders.ts — new SSR loader module replaces HTTP loopback helpers with direct service calls; per-loader try/catch blocks restore the redirect behaviour for deleted tasks.
• OneTaskDataFetcher + Redux cleanup — fallbackData eliminates the SSR→client loading flash; deprecated accesibleTaskIds slice state and setAccesibleTaskIds action are removed.

Confidence Score: 5/5

Safe to merge — the refactoring is well-scoped, the previous blocking concerns around deleted-task redirects and signed-URL refresh have been addressed, and no correctness regressions were found.

The core changes (direct service calls, parallel Promise.all, fallbackData) are logically sound. The per-loader error handling correctly restores the redirect path for missing tasks. The two remaining findings are minor: an unused selector binding and an overly-broad catch clause in the subtask loader that returns a safe default rather than crashing.

src/app/detail/[task_id]/[user_type]/loaders.ts — the loadSubtaskStatus catch block is broader than needed.

Important Files Changed

Filename	Overview
src/app/api/tasks/tasks.service.ts	Parallelises subtask count, assignee fetch, and optional body update; includes workflowState in the initial findFirst to avoid a second query. Logic is sound; the newBody !== task.body string-value comparison works correctly in JS.
src/app/detail/[task_id]/[user_type]/loaders.ts	New SSR loader module replacing HTTP loopback fetches with direct service calls. Error handling for loadTask and loadTaskPath is appropriately scoped to NOT_FOUND; loadSubtaskStatus over-catches all APIError types (see comment).
src/app/detail/[task_id]/[user_type]/page.tsx	Removes 5 HTTP loopback fetches in favour of direct loader calls; auth is consolidated to a single authenticateWithToken call. The null guard on task before rendering OneTaskDataFetcher ensures the non-nullable initialTask prop is always satisfied.
src/app/_fetchers/OneTaskDataFetcher.tsx	Adds fallbackData so the SSR task renders immediately without a loading flash; SWR still revalidates on mount by default, preserving the signed-URL refresh path.
src/hoc/ClientSideStateUpdate.tsx	Removes the deprecated accesibleTaskIds prop and setAccesibleTaskIds dispatch; activeTaskInStore is newly destructured from the selector but is never used in the component (see comment).
src/app/api/tasks/tasks.controller.ts	Removes the now-redundant getTaskAssignee call; the assignee is included in the getOneTask return value and the JSON response shape is unchanged.
src/app/api/tasks/subtasks.service.ts	Adds getSubtaskStatus as a thin wrapper around getSubtaskCounts so callers get the combined { count, canCreateSubtask } shape directly from the service.
src/redux/features/taskBoardSlice.tsx	Removes the deprecated setAccesibleTaskIds reducer and its export; no remaining callers after ClientSideStateUpdate.tsx cleanup.

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    A[Browser Request] --> B[page.tsx SSR]
    B --> C[authenticateWithToken]
    C --> D[user object]
    D --> E[Promise.all loaders]
    E --> F[loadTask]
    E --> G[loadSubtaskStatus]
    E --> H[loadTaskPath]
    E --> I[loadViewSettings]
    F --> J[TasksService.getOneTask]
    J --> K[DB findFirst with workflowState]
    K --> L[replaceImageSrc]
    L --> M[inner Promise.all]
    M --> N[DB count subtasks]
    M --> O[getTaskAssignee]
    M --> P{bodyChanged?}
    P -- yes --> Q[DB update body]
    P -- no --> R[skip update]
    N & O & Q --> S[task plus assignee plus subtaskCount]
    N & O & R --> S
    F & G & H & I --> T[Render HTML with SSR data]
    T --> U[SWR shows fallbackData immediately]
    U --> V[SWR revalidates on mount]
    V --> W[API refreshes signed URLs]

Loading

_{Reviews (3): Last reviewed commit: "fetch task detail on mount" | Re-trigger Greptile}

[greptile-apps]

Promise.all rejects when task is hard-deleted, hiding the redirect page

loadTaskPath calls getTraversalPath, which queries SELECT "path" FROM "Tasks" WHERE id = ${id}::uuid with no workspaceId or deletedAt guard. For a task that no longer exists in the DB (hard-deleted or bad task_id), it throws APIError(NOT_FOUND). Similarly, loadSubtaskStatus → getSubtaskCounts throws INTERNAL_SERVER_ERROR when the row is missing. Either error causes the whole Promise.all to reject before the if (!task) check at line 81 is reached, so the user sees an unhandled error page instead of DeletedRedirectPage. The old HTTP-loopback code used raw fetch without res.ok checks, so those helpers silently returned malformed JSON and task === null was still reachable. A guard like wrapping loadTaskPath and loadSubtaskStatus in per-call try/catch (returning a sensible default on NOT_FOUND) would preserve the redirect behavior.

[priosshrsth]

@greptileai Addressed this.

[greptile-apps]

Client-side signed-URL refresh is permanently disabled

With revalidateOnFocus: false, refreshInterval: 0, and now revalidateOnMount: false whenever initialTask is present (which is always), SWR will never call /api/tasks/{id} again after the initial render. The server re-signs image URLs on every SSR pass, so they're fresh at page load. But if the user keeps the tab open past the signed-URL TTL, images will silently break with no client-side mechanism left to refresh them. The previous mount revalidation was the only safety net for long-lived sessions on image-heavy tasks.

[greptile-apps]

The as unknown as TaskResponse double-cast silences type errors that may indicate a real mismatch between the Prisma return shape (Task & { workflowState: WorkflowState | null; assignee?: ... }) and TaskResponse. If TaskResponse expects a non-nullable workflowState, callers will see a runtime null where the type promises otherwise. Consider aligning the return type or using a proper serializer instead of a double cast.

Suggested change

	try {
	return (await new TasksService(user).getOneTask(taskId)) as unknown as TaskResponse
	} catch (err) {
	try {
	const task = await new TasksService(user).getOneTask(taskId)
	// TODO: replace with a typed serializer once TaskResponse is aligned with
	// the Prisma shape (workflowState nullability, assignee union, etc.)
	return task as unknown as TaskResponse
	} catch (err) {

[priosshrsth]

@greptileai Addressed this.

[priosshrsth]

@greptileai review again. If your comments are ersolved then resolve them.

[priosshrsth]

@greptileai review this PR again.