Atlas Query Profiler and slow-ops logs for shape discovery
Parent: Operation rejection filters · Published reference · snapshot 2026-10-31
↓ Facts as markdownall context files
Source-anchored research for Atlas Query Profiler and slow-ops logs for shape discovery. 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.
Structure and components
- Query Profiler data and diagnostic logs can contain query contents and literal values. Treat captured operation details as potentially sensitive when exporting examples for investigation. (medium confidence) [source]
How it works
- Atlas can temporarily stop collecting logs for its analysis tools when log volume spikes. Downloaded log files are not subject to these UI analysis rate limits, so use downloaded logs to check whether apparently missing operations were logged. (medium confidence) (verified as of 2026-10-14) [source]
Problems, failure modes and limitations
- Atlas Query Profiler identifies slow operations from mongod logs, whereas the Database Profiler writes operation details to system.profile. Atlas log-based discovery does not require enabling writes to system.profile. (medium confidence) [source]
- In Query Insights > Query Profiler, selecting a scatterplot point and opening View more details exposes the full query shape, execution statistics, and query log. Use these details to turn a slow-operation observation into a concrete query to investigate. (low confidence) (verified as of 2026-10-14) [source]
- Slow-operation diagnostic log entries carry command and execution information suitable for shape discovery. Where available, the queryShapeHash field can be extracted alongside namespace and execution metrics; its availability is version-dependent. (medium confidence) [source]
- Downloaded logs remove the Atlas display-sampling layer, but do not reconstruct operations that server logging never recorded. When slow-operation thresholds select the input, frequently executed fast operations can be absent even if their cumulative workload cost is large. (medium confidence) [source]
- Atlas manages the slow-operation threshold per mongod host using average operation execution time. Changing the threshold through db.setProfilingLevel affects the logging input and may affect performance; documented Query Profiler settings reset after a node restart. (low confidence) (verified as of 2026-10-14) [source]
- Atlas Query Profiler excludes bulk operations, including bulkWrite and updateMany with multiple stmtIds, from its displayed sampling and analysis. Separately, a shard-local lookup can increase foreign-collection latency metrics without producing a separate foreign-collection slow log entry. (low confidence) (verified as of 2026-10-14) [source]
- The mdbkit project documents offline grouping of structured slow-query logs with literal values stripped. Its queries command ranks recorded groups by cumulative duration, count, mean, maximum, documents examined, or scan ratio; these are measurements of the supplied logs. (low confidence) (verified as of 2026-10-30) [source]
- Use execution time together with documents examined, documents returned, and plan information when triaging discovered operations. Ranking cumulative recorded duration helps distinguish a frequently repeated expensive pattern from a single slow outlier. (medium confidence) [source]
- Atlas Query Profiler discovers inefficient operations from mongod logs; changing the Database Profiler's profiling level does not change which slow queries it displays. (high confidence) [source]
- A slow-log line is not necessarily one client operation: multiple slow writes in one bulk operation can produce separate entries. Account for this before interpreting grouped log-line counts as operation frequencies. (low confidence) [source]
- Slow-operation diagnostic logs can provide historical evidence even when Database Profiler collection is disabled. A missing log entry does not establish that a shape never ran. (medium confidence) [source]
- MongoDB's Query Insights agent demonstrates automated discovery: it reads Performance Advisor slow logs and index suggestions, then reports high-impact recorded query shapes. (low confidence) (verified as of 2026-10-30) [source]
- Queryable Encryption omits CRUD operations on encrypted collections from the slow query log; log redaction also reduces diagnostic detail. These settings can restrict shape discovery. (low confidence) (verified as of 2026-10-14) [source]
Comparisons and alternatives
- Use recorded command examples and execution metrics to identify candidate shapes for investigation. Slow-log retrieval supplies actual query examples rather than an exhaustive shape inventory. (high confidence) [source]
Facts and statements
- Atlas Query Profiler bins operations using operation type, namespace, matching hasSort and usedIndex values, and nearby timestamps. These documented binning criteria do not establish that every plotted bin represents one distinct query shape. (low confidence) (verified as of 2026-10-14) [source]
- The documented Query Profiler view samples roughly 100,000 logs and displays at most 100,000 chart points. Time-range bucketing can make a wider range show fewer sampled operations than a narrower range; zooming can refetch more detailed data. (low confidence) (verified as of 2026-10-14) [source]
- The documented default Query Profiler view includes operations across the cluster, including primary and secondary hosts, with host filters available. Check host and time filters before concluding that an operation was absent. (low confidence) (verified as of 2026-10-14) [source]
- Inspect execution time, documents and keys examined, examined-to-returned ratios, and planSummary to prioritize inefficient recorded operations; then inspect the full query and execution details. (high confidence) [source]
- Query Profiler is available on M10+ clusters and displays cluster-wide operations by default, with host, time-range, and operation-type filters. (medium confidence) (verified as of 2026-10-30) [source]
- The UI samples roughly 100,000 logs, may lag five minutes, and changes sampling with the selected time range. Download mongod logs for exact counts of logged operations. (low confidence) (verified as of 2026-10-30) [source]
- Query Profiler excludes bulk operations and may pause collection during extreme log volume; downloadable logs avoid those UI collection limits. (low confidence) (verified as of 2026-10-30) [source]
Children
- No children recorded.