Atlas Performance Advisor drop-index recommendations (unused, redundant, hidden)
Parent: Hidden indexes and safe index removal · Published reference · snapshot 2026-10-30
Also known as: Atlas Performance Advisor drop-index recommendations
↓ Facts as markdownall context files
Source-anchored research for Atlas Performance Advisor drop-index recommendations (unused, redundant, hidden). 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.
How it works
- Atlas flags an index as unused when it has supported no query for at least seven days after creation or server restart; it inspects only the 20 most active collections for unused indexes. (low confidence) (verified as of 2026-10-30) [source]
- Atlas always recommends dropping hidden indexes, but tells users to drop only after deciding they are unnecessary. Its hidden-index display covers the top 20 most active namespaces when the page loads, so it may omit others. (low confidence) (verified as of 2026-10-30) [source]
- Review usage over a representative application cycle before removing an apparently unused index. Percona warns specifically about monthly reporting indexes; OneUptime likewise qualifies zero-access candidates by a representative observation period. (medium confidence) [source]
- Hide a candidate before dropping it to test query-planning impact; if performance regresses, unhide it. A hidden index remains maintained and is immediately available when unhidden, avoiding a rebuild. (medium confidence) [source]
- Server restarts reset index-usage counters, and changing an index between hidden and visible also resets its $indexStats. Account for these observation-window resets when reviewing removal evidence. (medium confidence) [source]
Comparisons and alternatives
- Hiding does not remove index maintenance or storage costs: hidden indexes are updated on writes and continue consuming disk space and memory. A hidden trial therefore tests query dependence rather than the resource savings from actual removal. (medium confidence) [source]
Facts and statements
- Atlas Performance Advisor recommends removing unused, redundant, and hidden indexes to reduce write overhead and recover storage. (low confidence) (verified as of 2026-10-30) [source]
- Atlas labels a same-collection prefix index redundant: with {a:1,b:-1}, {a:1} is a candidate but {b:-1} is not. The related covering indexes shown beneath a candidate are not themselves removal recommendations. (low confidence) (verified as of 2026-10-30) [source]
- Hiding a candidate removes it from query planning but retains write maintenance and storage costs. Unhiding restores availability without rebuilding the index. (medium confidence) [source]
- Before dropping based on usage counters, inspect every read-serving replica-set member: reads on another member can make one node appear to have an unused index. (medium confidence) [source]
- A zero access counter does not prove an index has no purpose: uniqueness checks and TTL expiration do not increment these read-usage counters. Hidden unique and TTL indexes still enforce uniqueness and expire documents, so a successful hide trial does not prove that losing those behaviors is safe. (medium confidence) [source]
- Prefix coverage alone is insufficient grounds for automatic removal: MongoDB's compound-index removal guidance assumes neither index has a sparse or unique constraint, qualifications omitted from OneUptime's simple redundancy example. (medium confidence) [source]
- Atlas's Drop Index dialog supplies shell commands for dropping or hiding and a link to the index-management UI; the user then executes the selected action through the UI or shell. (low confidence) (verified as of 2026-10-30) [source]
- A June 2017 incident report for MongoDB 3.2.11 describes query latency increasing after dropping zero-use indexes. SERVER-29419 was closed as a duplicate of WT-3207, a checkpoint-wait issue; it is historical evidence, not proof of a current Atlas recommendation defect. (low confidence) [source]
- A practical validation sequence is to inspect usage and index definitions, hide the candidate, monitor representative queries for regressions, then drop only after validation; unhide if query performance regresses. (medium confidence) [source]
Children
- No children recorded.