CSE Roadmap

Specialist tier · about 2 hours · 5 parts

Part 5 of 5 · 4 min read

Exercises

Easy

Easy 1: Define rebalancing (15 minutes)

State, in one sentence, what rebalancing is and what typically triggers it.

Done when: moving data between partitions, typically triggered by a change in the number of nodes, is correctly described.

Hint

Moving data between partitions, typically triggered by a change in the number of nodes in the cluster.

Easy 2: Name the two rebalancing approaches (15 minutes)

Name the two approaches to triggering rebalancing covered on this page.

Done when: automatic rebalancing and manual rebalancing are both correctly named.

Hint

Automatic rebalancing and manual rebalancing.

Easy 3: State the key requirement for zero-downtime rebalancing (15 minutes)

State what rebalancing must do to avoid any downtime.

Done when: moving data incrementally while keeping routing information continuously updated is correctly described.

Hint

Moving data incrementally, a portion at a time, while keeping routing information continuously updated.

Medium

Medium 1: Identify the risk of skipping rebalancing after adding a node (25 minutes)

A cluster adds a new node but never rebalances. Identify the practical consequence.

Done when: the new node sitting idle while existing nodes keep their original, full load is correctly identified, defeating the purpose of adding the node.

Hint

The new node sits idle while the existing nodes keep their original, full load, defeating the purpose of adding capacity.

Medium 2: Choose a rebalancing approach for an unpredictable, fast-growing cluster (25 minutes)

A cluster's size changes frequently and unpredictably, and an operator cannot always be available to react quickly. Recommend a rebalancing approach.

Done when: automatic rebalancing is recommended, correctly reasoning that frequent, unpredictable changes need a response faster than an operator can reliably provide.

Hint

Automatic rebalancing, since frequent, unpredictable size changes need a faster response than an operator can reliably provide.

Medium 3: Choose a rebalancing approach for a cost-sensitive, carefully scheduled cluster (30 minutes)

A cluster changes size rarely, and the operator wants to carefully control exactly when data movement happens, to avoid competing with a known daily traffic peak. Recommend a rebalancing approach.

Done when: manual rebalancing is recommended, correctly reasoning that the operator's need for precise timing control outweighs the convenience automatic rebalancing would offer here.

Hint

Manual rebalancing, since the operator's need for precise timing control outweighs automatic rebalancing's convenience here.

Hard

Hard 1: Explain why routing information must stay updated during a move (30 minutes)

Explain what would go wrong if a query were routed using only the pre-move partition assignment while a rebalancing operation was already underway.

Done when: the explanation correctly notes that a query could be sent to a partition that no longer holds the data, since it has already moved, resulting in a missing or stale read unless routing information is kept continuously updated throughout the move.

Hint

A query routed to the old partition after data has already moved would miss the data or read a stale copy, unless routing information is kept continuously updated throughout the move.

Hard 2: Explain why moving all data at once would be worse than moving it incrementally (35 minutes)

Explain why rebalancing a large fraction of a cluster's data in one single large transfer would likely be worse for the system than moving the same amount incrementally over time.

Done when: the explanation correctly notes that a single large transfer would consume a large share of available network and disk bandwidth all at once, competing heavily with ordinary read and write traffic during that period, whereas spreading the same total movement out incrementally keeps the competing load smaller and more manageable at any given moment.

Hint

A single large transfer consumes a large share of bandwidth all at once, competing heavily with ordinary traffic, while the same total movement spread out incrementally keeps that competing load smaller at any given moment.

Hard 3: Evaluate a flawed automatic rebalancing policy (40 minutes)

A cluster is configured to automatically rebalance the instant any node's load crosses a threshold, even briefly. Identify the practical problem with this policy and propose a specific improvement.

Done when: the practical problem, triggering a costly rebalancing operation in response to a brief, temporary spike that would have settled on its own, is correctly identified, and a reasonable improvement is proposed, such as requiring the load threshold to be sustained for some minimum duration before triggering a rebalance.

Hint

Triggering a rebalance on every brief, temporary load spike wastes effort on moves that were not actually needed; requiring the threshold to be sustained for some minimum duration before triggering avoids reacting to noise.