Episode 5. DynamoDB. Day 5. Key value at planetary scale, and the interview loves its failure modes.

Key facts. DynamoDB partitions data by partition key hash. On demand versus provisioned capacity with auto scaling. Global Tables give multi Region active active replication. DAX adds microsecond caching. Streams capture item level changes. Say this sentence: A hot partition is a partition key design bug, not a capacity bug. Adding throughput rarely fixes it.

Scenario 1. Our DynamoDB table is throttling even though average utilization is only forty percent. Help.

Understand. Ask: which keys are hot? Is traffic skewed to a few partition keys? Any scans or hot sort keys? Onset sudden or gradual? Any new access pattern or deploy?

Prepare. Say the data in order. CloudWatch ThrottledRequests split by operation. Top partition keys by consumed capacity. Access pattern review: queries versus scans. Global secondary index throttling separately. DAX hit rates if present.

Say your translation line: Same instinct as index hot block contention: the hash key is the index, and skew is the bug.

Identify. Rank the hypotheses. One: hot partition key with uneven traffic. Two: a scan or large page consuming a partition's share. Three: GSI with a different hot key throttling independently. Four: burst traffic exceeding partition level token buckets. Fix: redesign the key with a composite or write sharding suffix, replace scans with queries, add DAX for read heavy keys, use exponential backoff with jitter.

Support. Success: throttles at zero for seven days, consumption evenly spread across partitions, and alarms on throttles plus hot key share.

Scenario 2: single table design question. Say: I'd model access patterns first, then entities, with overloaded GSIs and sparse indexes for each query. Then commit. Say: my recommendation is. Know when to say no: complex joins and ad hoc analytics belong on Aurora or Athena, not DynamoDB.

Your Day 5 lab: create a table with a deliberately skewed key, drive load, watch throttles, then redesign the key with a sharded suffix and compare. Delete the table the same day.

Tomorrow, Episode 6: caching, document, and graph. ElastiCache, MemoryDB, DocumentDB, and Neptune. Understand. Prepare. Identify. Support.