Phase 5: Caching Plan

Goal

Reduce Prisma operations, SQL statements, and user-visible latency by caching safe repeated reads.

Prisma Accelerate provides global query caching for reads using TTL and stale-while-revalidate patterns. Prisma's pricing page includes Prisma Accelerate and cache invalidation details by plan. Sources: Prisma pricing, Prisma query optimization.

Cache candidates

Data	Suggested TTL	Invalidation
Workspace shell metadata (non-auth only)	5-15 min	Workspace display/settings update
Brand profile / brand voice	5-30 min	Brand update
Social account list	1-5 min	Connect/disconnect/sync
Calendar summary counts	30-120 sec	Create/update/delete/schedule
Content list first page	30-120 sec	Create/update/delete/archive/schedule
Campaign list	1-5 min	Campaign create/update/archive
Static platform metadata	1 hour+	Deploy/config change
AI prompt templates	5-30 min	Template edit

Cache layers

Layer A: Prisma Accelerate query cache

Use for repeated read queries where short staleness is acceptable.

Good targets:

workspace shell metadata that excludes role, membership, owner, billing, OAuth, and permission decisions
brand settings
calendar summaries
content list summaries
platform metadata
campaign lists

Avoid for:

auth-critical permission checks
OAuth token reads/writes
billing-critical state
publishing queue state
write-after-read consistency flows

Layer B: Request-scope cache

Use a per-request cache to prevent the same workspace/user context from being loaded multiple times inside one API action.

Example:

const requestCache = new Map<string, Promise<unknown>>();

export function oncePerRequest<T>(key: string, loader: () => Promise<T>): Promise<T> {
  if (!requestCache.has(key)) {
    requestCache.set(key, loader());
  }
  return requestCache.get(key) as Promise<T>;
}

Layer C: Short in-process cache

Useful for Cloud Run warm instances, but not durable.

type CacheEntry<T> = {
  value: T;
  expiresAt: number;
};

const cache = new Map<string, CacheEntry<unknown>>();

Use only for small, safe, non-critical values.

Layer D: Private HTTP cache

For authenticated read-heavy routes:

Cache-Control: private, max-age=30, stale-while-revalidate=60

Use carefully because TendSocial data is workspace-private.

Layer E: Public edge cache

Use Vercel/Cloudflare edge caching only for public content:

marketing pages
docs pages
public landing data
public blog/SEO pages if any

Do not cache authenticated workspace API responses at the public edge.

Cache invalidation strategy

Create a single invalidation utility:

export async function invalidateWorkspaceCaches(workspaceId: string, keys: string[]) {
  // Prisma Accelerate cache tags if used
  // in-process cache delete
  // optional Redis/KV delete later
}

Example tags:

workspace:{workspaceId}:shell
workspace:{workspaceId}:brands
workspace:{workspaceId}:social-accounts
workspace:{workspaceId}:content-list
workspace:{workspaceId}:calendar-summary
workspace:{workspaceId}:campaigns

Mutation handlers must invalidate relevant tags:

Mutation	Invalidate
Update workspace display/settings	workspace shell metadata
Update member/role/owner	invalidate affected non-auth shell caches; auth decisions stay uncached
Update brand voice	brand profile
Connect social account	social account list
Create/update/archive content	content list, calendar summary
Schedule/unschedule content	calendar summary, content list
Update campaign	campaign list/detail

Measurement requirements

For each cached route/action, compare:

cold run
warm run
after invalidation

Report:

Field	Meaning
coldOrmOps	DB operations without cache
warmOrmOps	DB operations after cache
reductionPct	Operation reduction
staleRisk	Low/medium/high
invalidationPath	Mutation path that clears cache

Acceptance criteria

Item	Target
Workspace shell repeated load operation reduction	50%+
Dashboard repeated load operation reduction	30%+
Calendar repeated same-range load reduction	30%+
Brand profile repeated load reduction	50%+
Auth/permission stale bugs	0
Cache invalidation paths documented	Yes
Cached routes covered in operation tests	Yes

Sources

• Prisma pricing
• Prisma query optimization