FTDC Storage Overhead & Retention Budgeting
Parent: FTDC and Log Tooling · Published reference · snapshot 2026-10-23 · viability 3.50 / 5 (research)
↓ Facts as markdownall context files
Source-anchored research for FTDC Storage Overhead & Retention Budgeting. 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
- Percona's captured Atlas bundle contained recent samples in metrics.interim after the newest numbered file had ended. Preserve the whole directory, including that file, when collecting recent incident evidence. (low confidence) (verified as of 2026-08-06) [source]
- The retention control is diagnosticDataCollectionDirectorySizeMB: MongoDB deletes the oldest diagnostic files when the directory exceeds its configured size. A fixed byte budget does not promise a fixed time window. (high confidence) [source]
- Include metrics.interim when preserving an incident capture: OneUptime identifies it as recent unflushed data, and Percona observed it extending five minutes beyond the newest numbered file in its downloaded bundle. (medium confidence) [source]
- Disabling FTDC to recover its small disk allocation removes diagnostic history. MongoDB discourages disabling it in production because support diagnosis can then require more time or resources. (low confidence) [source]
Measurements and reference values
- The MongoDB 8.0 FTDC page documents 250 MB by default, or 500 MB for sharded clusters, with the increase from 200/400 MB occurring in 8.0.10. Inspect the exact server version and effective parameter before budgeting. (low confidence) (verified as of 2026-10-23) [source]
- Percona's August 2026 test found diagnosticDataCollectionDirectorySizeMB=400 on an Atlas MongoDB 8.0.29 deployment. Its idle-node measurement was approximately 0.93 MB per 32 minutes, roughly ten days at that cap. This is a specific observation, not a universal Atlas default or a production retention guarantee. (low confidence) (verified as of 2026-08-06) [source]
- A historical Red Hat Satellite 6.7 report attributed about 5% of mongod CPU time on a busy installation to FTDC. Treat this as an environment-specific overhead observation from 2020, not a modern MongoDB baseline or a general recommendation to disable diagnostics. (low confidence) (verified as of 2020-05-11) [source]
- Self-managed mongot has a separate FTDC budget: its documentation lists advancedConfigs.ftdc.directorySizeMb=100, fileSizeMb=10 and collectionPeriodMillis=1000, and describes approximately 1 GB/day per instance depending on load. Do not substitute mongod's retention assumptions for mongot's; measure each process separately. (low confidence) (verified as of 2026-10-14) [source]
- MongoDB 8.3 documents a 500 MB directory default in its parameter reference and a 500 MiB default in its release notes. The FTDC overview also lists 500 MB. Inspect the effective setting when budgeting an installed deployment. (high confidence) (verified as of 2026-10-23) [source]
- diagnosticDataCollectionDirectorySizeMB and diagnosticDataCollectionPeriodMillis can be changed at startup or runtime. The latter defaults to 1000 ms in v8.3. MongoDB advises changing FTDC defaults only for specific diagnostic purposes requested by its engineers. (high confidence) (verified as of 2026-10-23) [source]
- OneUptime reports typical FTDC costs below 1% CPU and around 1 MB/hour. These are single-source estimates without a disclosed benchmark in the article; they should not be treated as guaranteed overhead or used alone to promise a retention window. (low confidence) (verified as of 2026-03-31) [source]
Problems, failure modes and limitations
- The current manual and MongoDB 8.3 parameter reference specify a 500 MB diagnostic.data directory limit and a one-second sampling default. These are version-scoped defaults, not proof of a deployment's effective configuration. (high confidence) (verified as of 2026-10-23) [source]
- MongoDB 8.3 exposes diagnosticDataCollectionDirectorySizeMB, diagnosticDataCollectionFileSizeMB and diagnosticDataCollectionPeriodMillis. The directory minimum is 10 MB and must exceed the file limit; file size defaults to 10 MB with a 1 MB minimum; the sample interval defaults to 1000 ms with a 100 ms minimum. These parameters support startup and runtime changes. (low confidence) (verified as of 2026-10-23) [source]
- A larger storage budget cannot compensate for stopped sampling. Percona's April 2025 incident on MongoDB 6/7 traced frozen FTDC to statfs() blocking on an unrelated autofs mount; both diagnostic files and getDiagnosticData stopped advancing while the cluster remained operational. Check sample freshness as well as retained bytes. (low confidence) (verified as of 2025-04-22) [source]
- The v8.3 parameter reference documents a 10 MB per-file default, a directory minimum of 10 MB, and requires the directory limit to exceed the per-file limit. File rotation and directory retention are separate controls. (low confidence) (verified as of 2026-10-23) [source]
Facts and statements
- FTDC stores compressed per-second diagnostic samples. A size-limited directory rotates away older files, so its configured capacity does not guarantee a fixed number of retained days. (high confidence) [source]
- FTDC's space efficiency does not establish weeks of guaranteed retention. MongoDB's encoding-library README describes weeks in a few hundred megabytes, while Percona reports only two to five days on its production-traffic Atlas example. (medium confidence) (verified as of 2026-10-16) [source]
- Budgeting inference: measure compressed bytes generated per unit time under representative workloads; estimate a node's retention as its usable FTDC capacity divided by that rate. Sum local caps across processes and separately budget durable archives by generated bytes, desired history and retained copies. Allow margin for rotation and workload variation; the estimate is not a retention SLA. (medium confidence) [source]
- Capture the diagnostic.data directory during an incident before rotation deletes older evidence. On ephemeral hosts, preserve it on persistent storage or copy it periodically to durable storage. Budget that external retention in addition to local FTDC capacity. (high confidence) [source]
- FTDC is enabled by default and normally samples once per second. It writes compressed binary metrics; mongod places them under storage.dbPath/diagnostic.data. This directory therefore consumes local database-volume capacity. (high confidence) [source]
Children
- No children recorded.