Understanding query summaries

Both backends expose summarize(raw). Import it from the backend that produced the response. The parser operates offline and returns the shared QuerySummary class.

from bondi.mongodb import summarize

# profile is an ExecutionResult from executor.profile(query).
summary = summarize(profile.raw)
print(summary.plan_stages, summary.index_names)
print(summary.documents_examined, summary.keys_examined)

Field	Meaning
execution_time_ms	Server-reported execution time, when available
rows_returned	Server-reported output count
documents_examined	Whole-operation document examinations, not unique documents
keys_examined	Whole-operation index entries examined
plan_stages	Native structured stage names, retaining traversal order and repetitions
index_names	Unique index names in encounter order

None means unavailable, not zero. An empty index tuple means no named index was found within a supported plan; it does not alone prove a collection scan. The plan-stage tuple is a convenient description, not a complete plan tree.

MongoDB structured plan stages such as IXSCAN and GROUP differ from aggregation pipeline operators such as $group. The parser does not conflate them. Couchbase operator output counts are not automatically document/index examination counts; unsupported totals remain None.

Use result.elapsed_ms for client-observed duration and the summary for server measurements. Check workload result errors before parsing. summarize(None) returns an empty summary; malformed or failed responses are not silently accepted.

See shared fields, MongoDB parsing and Couchbase parsing for supported layouts and boundaries.