KVStore

Simple key-value storage backed by DynamoDB.

When to use: Fast, single-key lookups with get/put/delete semantics. Good for caches, session stores, feature flags, and config values.

When NOT to use: If you need to query by multiple fields or secondary indexes, use DistributedTable. If you need full SQL, use Database.

Design & mock parity details: DESIGN.md

API

const store = new KVStore(scope, id, options?)

Runtime only. Data methods (get, put, delete, scan) run at request time — call them inside an ApiNamespace method, RawRoute handler, job handler, or a runtime script, not at the top level of your aws-blocks/index.ts. Top-level code runs during CDK synth, where the block resolves to its infrastructure construct, whose data methods are guards: a top-level call throws an Error reading KVStore.<method>() cannot be called during CDK synth. Data methods run at request time — …. To seed data, do it from inside a handler or a separate runtime script. Constructing the block at module scope is fine; only method calls must move into handlers.

Options

Expiring Items (TTL)

TTL is opt-in in two steps — turn it on for the table, then set an expiry per write:

const cache = new KVStore(scope, 'cache', { ttl: true });

// Relative expiry — delete 5 minutes from now
await cache.put('otp:alice', code, { ttlSeconds: 300 });

// Absolute expiry — a Date, or Unix epoch time in SECONDS
await cache.put('session:1', record, { expiresAt: new Date('2027-01-01') });

// No expiry — stored indefinitely (the default)
await cache.put('config:theme', 'dark');

• ttl defaults to false. Enabling it on a table that already exists is a CloudFormation update to the live table, so it never happens implicitly.
• The attribute is named ttl. It is written only when ttlSeconds or expiresAt is supplied, and it holds Unix epoch seconds.
• ttlSeconds and expiresAt are mutually exclusive. Passing both — or a value that looks like epoch milliseconds — throws ValidationFailedException rather than silently storing a year-5138 expiry.
• Reads never return expired items. DynamoDB's reaper is asynchronous (typically within 48 hours), so get returns null and scan skips an item as soon as its expiry passes, in every runtime.
• Expiry is per write. Re-putting a key without expiry options clears any previous expiry; re-putting with them slides it forward.
• TTL composes with conditional writes: put(key, value, { ifNotExists: true, ttlSeconds: 60 }).

TTL is a retention control, not an authorization one. Anything security-sensitive (session validity, token revocation) must still be checked on read.

Conditional Operations

Both put and delete accept an optional options object:

// Only write if key doesn't exist (idempotent create)
await store.put('user:alice', data, { ifNotExists: true });

// Only write if current value matches (optimistic locking / compare-and-swap)
await store.put('counter', newVal, { ifValueEquals: oldVal });

// Compose both (OR): create it, or update it only if unchanged — write succeeds
// if the key is absent OR its current value matches; fails only if it exists and differs.
await store.put('config', next, { ifNotExists: true, ifValueEquals: prev });

// Only delete if key exists
await store.delete('temp', { ifExists: true });

// Only delete if value matches
await store.delete('lock', { ifValueEquals: expectedVal });

All condition failures throw with error.name === KVStoreErrors.ConditionalCheckFailed. They serialize to JSON-RPC 409 (Conflict) over the wire (not 500), so error.status === 409 on the client and isBlocksError still matches by name. A conflict is flagged retriable when a value check participated (a stale-value optimistic-lock conflict — re-read and retry may succeed). For put, ifNotExists and ifValueEquals compose with OR, so a failure whenever ifValueEquals was set — including the combined ifNotExists+ifValueEquals case — is the stale-value case and is retriable; a pure ifNotExists put conflict is not retriable. For delete (independent checks, not OR), a pure ifExists conflict — or a combined ifExists+ifValueEquals conflict — is not retriable; only a value-only ifValueEquals delete conflict is. A blind retry of a non-retriable conflict fails identically.

Error Handling

import { isBlocksError } from '@aws-blocks/core';
import { KVStoreErrors } from '@aws-blocks/bb-kv-store';

try {
  await store.put('key', value, { ifNotExists: true });
} catch (e: unknown) {
  if (isBlocksError(e, KVStoreErrors.ConditionalCheckFailed)) {
    // key already exists
  }
  if (isBlocksError(e, KVStoreErrors.ItemTooLarge)) {
    // value is too large — consider compressing or splitting
  }
  throw e;
}

Examples

Basic Usage (string values)

const store = new KVStore(scope, 'cache');

export const api = new ApiNamespace(scope, 'api', (context) => ({
  async setGreeting(value: string) {
    await store.put('greeting', value);
  },
  async getGreeting() {
    return { value: await store.get('greeting') };
  },
}));

Without a schema, the value type defaults to string. To store structured data without runtime validation, pass a type argument — new KVStore<MyType>(scope, id); values are JSON-serialized on write and parsed on read automatically (no manual JSON.stringify needed).

Typed Values with Schema

Use a schema to get type-safe values with runtime validation:

The examples use Zod, but schema accepts any StandardSchemaV1 implementation (Zod, Valibot, ArkType). Install your chosen library, e.g. npm install zod.

import { z } from 'zod';

const sessionSchema = z.object({
  userId: z.string(),
  expiresAt: z.number(),
});

const sessions = new KVStore(scope, 'sessions', { schema: sessionSchema });

// Type is inferred from the schema — no generic arg needed
await sessions.put(token, { userId: 'alice', expiresAt: Date.now() + 3600000 });
const session = await sessions.get(token); // { userId: string; expiresAt: number } | null

// Invalid data is rejected at runtime
await sessions.put(token, { userId: 123 }); // throws ValidationFailedException

Wrapping an Existing Table

const legacy = new KVStore(scope, 'legacy', {
  table: KVStore.fromExisting('my-existing-table'),
});

Best Practices

• Keep keys short and descriptive (e.g., user:{id}, session:{token})
• Store one logical entity per KVStore instance
• Use a schema for structured values — it ensures runtime data matches your types
• Use { ifNotExists: true } for idempotent creates
• Use { ifValueEquals } for compare-and-swap when multiple writers are possible
• scan() returns an AsyncIterable — collect with await Array.fromAsync(store.scan()) or for await. Prefer get(key) over scan() (scans read every item).

Scaling & Cost (AWS)

• Billing: PAY_PER_REQUEST — no provisioned capacity to manage
• Latency: Single-digit ms reads and writes
• Throughput: Scales automatically, no upper limit on table size
• Item size limit: 400 KB per item
• Cost: ~$1.25 per million writes, ~$0.25 per million reads
• Durability: 99.999999999% (11 nines) across 3 AZs

Local Development

Mock data persists to disk at .bb-data/{fullId}/ across dev server restarts. Wipe with rm -rf .bb-data. The mock validates the 400 KB item size limit, schema validation, and conditional check failures, matching AWS behavior.