Other AWS Databases

Purpose-built databases: DocumentDB, ElastiCache and MemoryDB, Neptune, Timestream, Keyspaces and OpenSearch, with examples and a guide to choosing.

1. Purpose-Built Databases on AWS

AWS offers a database for each data model. Data engineers often extract from these, feed them as serving layers, or choose between them in design discussions.

2. DocumentDB

from pymongo import MongoClient

client = MongoClient("mongodb://app:<password>@docdb-cluster.cluster-xyz.ap-south-1.docdb.amazonaws.com:27017/"
                     "?tls=true&tlsCAFile=global-bundle.pem&replicaSet=rs0&readPreference=secondaryPreferred"
                     "&retryWrites=false")
orders = client["shop"]["orders"]
orders.create_index([("customer_id", 1), ("order_ts", -1)])
recent = list(orders.find({"customer_id": "C1"}).sort("order_ts", -1).limit(10))

• Implements much of the MongoDB API, but not every feature or operator: test compatibility before migrating.
• Storage architecture similar to Aurora: shared storage across three AZs, up to 15 replicas.
• Change streams provide CDC; DMS can migrate from MongoDB or extract to S3.

3. ElastiCache & MemoryDB

import redis, json

r = redis.Redis(host="risk-cache.xxxxxx.aps1.cache.amazonaws.com", port=6379, ssl=True)

def get_risk_score(account_id: str) -> dict:
    cached = r.get(f"risk:{account_id}")
    if cached:
        return json.loads(cached)  # cache hit: microseconds
    score = load_from_feature_store(account_id)  # cache miss: slower source
    r.setex(f"risk:{account_id}", 300, json.dumps(score))  # expire after 5 minutes
    return score

• Cache-aside (above) is the most common pattern; set TTLs so stale values expire.
• ElastiCache is a cache (data can be lost on failure); MemoryDB keeps a durable transaction log so it can be the primary database.
• Valkey is the open-source fork of Redis supported by both services.

4. Neptune for Graph Problems

// openCypher: accounts sharing a device with a known fraudulent account (fraud ring detection)
MATCH (bad:Account {flagged: true})-[:USED]->(d:Device)<-[:USED]-(other:Account)
WHERE other <> bad
RETURN d.device_id, collect(DISTINCT other.account_id) AS linked_accounts
LIMIT 50;

Graph databases shine when the question is about relationships several hops deep (rings, networks, lineage). Load data in bulk from S3 with the Neptune bulk loader.

5. Timestream, Keyspaces & OpenSearch

6. Choosing a Database

7. Common Interview Questions