Atlas Query Profiler and Performance Advisor
Parent: MongoDB Explain Plans, Query Planner and Plan Cache · Published reference · snapshot 2026-10-21 · skill ranked-8f7c06b1
↓ Facts as markdownall context files
Source-anchored research for Atlas Query Profiler and Performance Advisor. Run status: COMPLETED.
These notes link each claim to its source. Some citations identify archived research reports that are no longer hosted on this site. A published reference means the content is available; it does not certify independent review or accuracy.Read the editorial policy and follow the sources before relying on a claim.
Definitions
- Atlas Query Profiler is not an exact operation census: it displays approximately 100,000 sampled logs at a time, caps charts at 100,000 points, and uses time-range-dependent bucketing that can show fewer operations for a wider range. It excludes bulk operations and log processing can lag by up to five minutes. During extreme log volume Atlas may stop collecting logs temporarily; UI log-analysis limits do not apply to downloadable log files. (high confidence) (verified as of 2026-10-19) [source]
Problems, failure modes and limitations
- MongoDB's current Atlas documentation lists both Query Profiler and Performance Advisor as available on M10+ clusters. Query Profiler presents mongod query-log data, while Database Profiler stores command details in system.profile according to the configured profiling level; changing that level does not change slow queries displayed in Atlas Query Profiler. (high confidence) (verified as of 2026-10-14) [source]
- Performance Advisor monitors operations MongoDB considers slow, groups supporting sample queries by query shape, and suggests indexes. Its displayed metrics include execution count, average time, query targeting, documents scanned and returned, and object size; it ranks recommendations by estimated impact based on total wasted bytes read. (high confidence) (verified as of 2026-10-14) [source]
- Performance Advisor monitors slow queries and suggests indexes. It groups sample queries by query shape and reports workload measures such as execution count, average execution time, average query targeting, documents scanned, and documents returned. Its slow-query threshold varies with average operation time on the cluster. (high confidence) (verified as of 2026-10-20) [source]
- Use Atlas Query Profiler to inspect observed slow-operation log data. Use explain() when you need the query planner's plan tree and execution statistics. MongoDB documents that explain() ignores the plan cache, so it does not measure the cached plan used by an ordinary execution. (high confidence) [source]
Comparisons and alternatives
- Atlas adjusts the slow-operation threshold used by Query Profiler per mongod host based on that host's average execution time. Performance Advisor's managed threshold varies with average operation time across the cluster, so the log-derived population depends on the current threshold rather than covering every operation. (high confidence) (verified as of 2026-10-14) [source]
- Use the Profiler and Advisor to find candidate patterns, then run explain with executionStats on a representative query and compare the plan and examined-versus-returned work before changing indexes. MongoDB's troubleshooting guidance recommends explain executionStats or allPlansExecution for deeper analysis; independent guides from OneUptime, TideWave, and ADHDecode also recommend validating slow-query findings with explain output. (high confidence) [source]
- Treat suggested indexes as candidates rather than automatic fixes. MongoDB says to consider how often the query runs and the read/write balance because every added index adds write-maintenance work; a MongoDB staff reply in the Community Hub likewise notes that the Advisor considers query frequency and that excess indexes can hurt writes. (high confidence) [source]
- A July 2023 Community Hub thread records a user whose poorly targeted query completed in under 30 ms and did not appear in the then-current Profiler; a MongoDB employee said the Profiler then relied on slowms and described planned support for operations with query-targeting ratios above 10,000. Current docs expose an examined-to-returned display metric but do not say whether a high ratio alone causes a fast operation to be logged, so treat the thread as a historical coverage warning rather than a statement of current behavior. (medium confidence) (verified as of 2026-10-14) [source]
Facts and statements
- The Query Profiler supports filtering by host, time range, and operation type. Its selectable metrics include execution time, response length, documents and keys examined, documents returned, yields, and examined-to-returned ratios; a selected operation exposes its query shape, execution statistics, log entry, and improvement suggestions. (high confidence) (verified as of 2026-10-14) [source]
- Performance Advisor ranks index suggestions by Impact using aggregated workload measures, including total wasted bytes read by associated operations. Treat a suggestion as a candidate: check whether the query occurs often enough, whether an existing index already serves it, and whether index maintenance cost is justified by the read benefit. (high confidence) [source]
- Datadog's Atlas integration exposes query-shape metrics under mongodb.atlas.querystats and attaches samplingstrategy and samplingrate tags. For MongoDB 9.0+ with sample-based collection, these metric values represent sampled read and write operations, so check the tags before interpreting exported query-shape counts as complete workload totals. (low confidence) (verified as of 2026-10-14) [source]
Children
- No children recorded.