Overview

The current InsertCmdShape (introduced in SERVER-122050) hashes only on the target namespace; the documents payload is serialized as a fixed ?array<?object> placeholder and HashValue is a no-op. As a result, all inserts into the same namespace – regardless of batch size – accumulate into a single queryStats store entry.

Concern raised

PR review on SERVER-122054 raised that this makes per-shape metrics like totalExecMicros hard to interpret: a single-document insert and a 10k-document insert share one bucket despite very different execution costs.

Options to evaluate

• Bucket by document count (e.g. log-scale buckets 1, 2-10, 11-100, 101-1000, 1000+) so workload sizes do not blur together.
• Per-statement shaping parallel to how update is keyed per-statement.
• Add a non-shape metric (e.g. totalDocsInserted) – gives workload context without splitting shapes.
• Status quo – match find convention (shape captures intent, not workload size); document the trade-off.

Action

Discuss in the weekly Query Stats sync. Decide whether to take action now or defer until customer feedback indicates need.