Practice the vocabulary of preventing one skewed partition key from overwhelming a single shard.
0 / 5 completed
1 / 5
At standup, a dev mentions one partition key value receiving a wildly disproportionate share of traffic compared to every other key, overwhelming the single shard responsible for it while the rest of the cluster sits idle. What is this problem called?
A hot partition, or hot key, is a single partition key value receiving a wildly disproportionate share of traffic compared to every other key, overwhelming the one shard responsible for it while the rest of the cluster sits comparatively idle. Assuming traffic always distributes perfectly evenly ignores that real-world access patterns, like a viral post or a popular customer, are rarely uniform. This imbalance is what makes partition key design a critical decision rather than an afterthought in a sharded system.
2 / 5
During a design review, the team wants to append a random or hashed suffix to a naturally skewed key so its traffic spreads across several sub-partitions instead of landing on just one. Which capability supports this?
Key salting appends a random or hashed suffix to a naturally skewed key so its traffic spreads across several sub-partitions instead of concentrating entirely on one. Using the key exactly as-is, with no salting, leaves all of that key's traffic landing on a single shard regardless of how much load it receives. This salting technique is a standard way to relieve a hot partition without redesigning the entire sharding scheme.
3 / 5
In a code review, a dev notices per-partition request-rate metrics are monitored continuously so an emerging hot key can be identified before it actually causes a latency spike. What does this represent?
Per-partition load monitoring tracks the request rate at the level of an individual partition, letting the team spot an emerging hot key before it actually causes a latency spike elsewhere in the cluster. Monitoring only an aggregate, cluster-wide rate hides exactly this kind of imbalance, since the cluster's overall average can look perfectly healthy while one shard is overwhelmed. This per-partition visibility is essential for catching a developing hot-key problem early enough to act on it.
4 / 5
An incident report shows one popular customer's account key received such disproportionate traffic that its single shard became overloaded and started timing out, while every other shard in the cluster remained well under capacity. What practice would prevent this?
Salting the hot key spreads its traffic across multiple sub-partitions, while per-partition load monitoring catches the imbalance early enough to act before it causes an outage. Leaving the key on a single shard with no salting or monitoring is exactly what let one shard become overloaded while the rest of the cluster stayed comfortably under capacity, as this incident describes. This combination of salting and monitoring is a standard defense against a hot partition in any sharded system with skewed access patterns.
5 / 5
During a PR review, a teammate asks why the team salts a hot partition key instead of just leaving the sharding scheme as-is and trusting that traffic will average out over time. What is the reasoning?
A real-world access pattern is often skewed toward a small number of keys, like a popular customer or a trending item, and that skew doesn't simply average out over time on its own. Salting spreads that skewed traffic across multiple sub-partitions rather than letting it overwhelm a single shard. The tradeoff is the added complexity of reassembling or aggregating results across a salted key's several sub-partitions when a query needs the full picture.