Which statement about partitioning time-series data is NOT part of a good strategy?

• A. Partition by time with pruning.
• B. Balance partition count to avoid small files.
• C. Partition by a user ID, which is not used in queries.
• D. Manage partition metadata for efficient queries.

Answer: (C)

Focusing partitioning on how you actually query the data is essential for good performance. For time-series data, the goal is to enable partition pruning so queries only read the relevant time slices, which dramatically reduces I/O and speeds up results. Time-based partitioning naturally supports this because most queries filter by a time range, allowing the system to skip entire partitions outside that range. It’s also important to keep the number of partitions balanced so you don’t end up with lots of tiny files or oversized partitions, which can slow down reads and complicate maintenance. Managing partition metadata helps the query engine locate and prune partitions efficiently, avoiding full scans.

Partitioning by a user ID when that field isn’t used in queries doesn’t help performance. If queries don’t filter on or aggregate by user IDs, data would be spread across many user-based partitions, forcing the engine to touch more partitions than necessary and potentially causing data skew and higher overhead. This approach misses the practical query patterns and undermines pruning and efficiency.