DynamoDB Production Patterns

TTL cleanup, on-demand scaling, and a social-stories-style access pattern walkthrough.

Expert11 min read

Read these first

DynamoDB Streams, DAX, and Transactions

Why this matters

VaultCommerce purges expired session tokens and idempotency keys with TTL — no scheduled Scan jobs deleting rows one page at a time.
Throttled requests (ProvisionedThroughputExceededException) mean keys or capacity misconfigured — not "DynamoDB is down."
The Instagram-stories pattern — high write volume, time-ordered reads, automatic expiry — maps directly to VaultCommerce flash-sale countdown banners.

Partition key: USER#42Routes to shard

Items in partition

ORDER#2026-001Sort key

ORDER#2026-002Sort key

PROFILESort key

Partition key routes to a shard. Sort key orders items within that partition.

TTL automatic cleanup

Java

aws dynamodb update-time-to-live \
  --table-name VaultCommerce \
  --time-to-live-specification "Enabled=true, AttributeName=expiresAt"

Java

import time

table.put_item(Item={
    "PK": f"IDEM#{token}",
    "SK": "TOKEN",
    "expiresAt": int(time.time()) + 86400,  # epoch seconds
})

DynamoDB deletes expired items within ~48 hours at no WCU charge — do not rely on instant deletion for security-critical expiry; enforce TTL plus application checks.

Production knobs

On-demand billing — default for unknown traffic; watch for cost spikes on scans.
Point-in-time recovery — continuous backups for accidental deletes.
CloudWatch — ConsumedReadCapacityUnits, UserErrors, throttled requests.
PartiQL / BatchGet — batch reads up to 100 items per call for BFF layers.

Stories-style time-ordered pattern

VaultCommerce flash banners mirror social-stories access: many writes, read recent items per user, auto-expire.

Java

{ "PK": "USER#42", "SK": "STORY#1710000000", "mediaUrl": "s3://...", "expiresAt": 1710086400 }
{ "PK": "USER#42", "SK": "STORY#1710003600", "mediaUrl": "s3://...", "expiresAt": 1710086400 }

Java

table.query(
    KeyConditionExpression=Key("PK").eq("USER#42") & Key("SK").begins_with("STORY#"),
    ScanIndexForward=False,
    Limit=10,
)

High-cardinality PK=USER#id spreads writes across partitions. TTL on expiresAt removes stale stories without batch jobs.

Monitoring and alarms

Java

aws cloudwatch put-metric-alarm \
  --alarm-name vaultcommerce-dynamo-throttles \
  --metric-name UserErrors \
  --namespace AWS/DynamoDB \
  --dimensions Name=TableName,Value=VaultCommerce \
  --statistic Sum --period 60 --threshold 10 \
  --comparison-operator GreaterThanThreshold

Investigate throttles: hot partition keys, insufficient GSI capacity, or sudden scan in a deploy.

Quick recall

Everything you need if you only revisit this box.

TTL deletes expired items asynchronously — ideal for tokens, stories, session data.
On-demand scales automatically; still avoid scans and hot partition keys.
Stories pattern: PK=USER#, time-ordered SK, TTL on expiresAt.
CloudWatch throttling alarms catch capacity and key-design issues early.
Enable point-in-time recovery on production tables.
VaultCommerce stores media in S3; DynamoDB holds keys, metadata, and expiry.

Test yourself

Answer these before moving on — recall is what makes it stick.