MongoDB Atlas Online Archive

Parent: MongoDB Atlas · researched 2026-05-28T18:33:37.810Z· 13 sources · 10 concepts · skill mongodb-atlas-online-archive

Online Archive automatically moves documents matching an archival rule out of the live Atlas cluster into Atlas-managed object storage, while keeping those documents queryable through the cluster's FD

Overview

Online Archive automatically moves documents matching an archival rule out of the live Atlas cluster into Atlas-managed object storage, while keeping those documents queryable through the cluster's FDI endpoint (transparent to the application). [source]

Requirements

Dedicated cluster M10+ (not available on M0/Flex/Serverless) [source]
MongoDB 5.0+ on the source cluster [source]

DATE (most common)

dateFormat options: "ISODATE" (default ISODate), "EPOCH_MILLISECONDS", "EPOCH_SECONDS" [source]

Partition Fields Strategy

Partition fields organize archived data into S3 prefix paths for efficient filtering. Critical for query performance. Choose fields commonly used in query filters: [source]
Maximum 2 partition fields per archive rule [source]
First field MUST match the dateField (for DATE criteria) or be the most selective filter field (for CUSTOM criteria) [source]
Order matters: highest-selectivity filter field should be first [source]
Supported partition pruning operators: $eq, $gt, $lt, $gte, $lte, $ne, $in [source]
Fields NOT in partitionFields trigger full archive scan when queried [source]

Archive Job Timing & Processing

Runs every 5 minutes [source]
Max throughput: 2 GB per 5-minute interval [source]
Max file size: 100 MB per archive file [source]
Documents deleted from live cluster AFTER successful write to object storage [source]
Schedule window: Archive jobs can be limited to off-hours to reduce production impact [source]

Query Routing

When querying via the cluster's FDI endpoint: [source]
Recent data → live cluster (fast, indexed) [source]
Archived data → Online Archive object storage (slower, partition-based) [source]
Combined queries → both, results merged transparently [source]
Applications do not need to change query patterns after data is archived - same connection string, same MQL. [source]

Cost Model

$5/TB cost applies when QUERYING archived data. Storage itself is much cheaper than cluster storage (NVMe/GP3). [source]
Minimum: 10 MB per archive query (no benefit from very small queries). [source]

Restore / Rehydration

Archive data is immutable (append-only). To restore to the live cluster: [source]

Limitations

No updates/deletes on archived data: Archive is immutable object storage; updateMany or deleteMany against archived documents will not affect them [source]
No Atlas Search on archived data: Full-text $search doesn't work on archived data [source]
No indexes on archived data: All archived queries use partition pruning only [source]
$sample not random: Returns first N documents, not a statistical random sample [source]
No transactions crossing live+archive: Cannot span a transaction across live and archived documents [source]
Maximum query timeout: 6 hours (inherits from Data Federation limit) [source]

Archive Not Moving Data

Check Atlas → Online Archive → [collection] → Activity tab for job status and errors [source]
Verify dateField name is exact (case-sensitive) and matches documents [source]
Confirm correct dateFormat - ISODATE requires actual ISODate values, not Unix epoch integers [source]
Confirm cluster tier is M10+ and MongoDB version 5.0+ [source]

Archive Backlog Growing

Archive jobs process 2 GB per 5-minute window [source]
If data accumulates faster than 2 GB/5 minutes, backlog will grow [source]
Mitigation: narrow the archive rule to reduce concurrent archiving volume, or contact MongoDB for higher throughput [source]

Slow Archive Queries

Partition fields not aligned with query filter → full archive scan [source]
Check partitionFields match the most common query filter fields [source]
Run explain() on the FDI endpoint to check nPartitionsScanned [source]

MACC / Cost Tracking

Archive storage costs appear on the Atlas invoice under Tools & Services → Online Archive. Query processing costs under Data Federation. Monitor in Atlas Billing → Current Invoice. [source]

Anti-Patterns

Choosing partition fields that don't match query patterns: Full archive scan on every query → high $5/TB cost [source]
Using CUSTOM criteria without an index on the filter field: Archive job itself will do slow scans to find matching documents - add an index on filter fields [source]
Querying archived data without partition-aligned filters: Always include a partition field in $match when querying archive [source]
Expecting consistency between live and archive: Documents are copied to archive and then deleted from live - there is no transactional guarantee between the two [source]
Using Online Archive as a backup system: Archive only stores the final document state at archival time - not a substitute for backup snapshots with PITR [source]

References

Online Archive Documentation [source]
Configure an Online Archive [source]
Query Archived Data [source]
Online Archive Billing [source]

Children

Archive Rules (frontier)
Partition Fields (frontier)
Federated Query on Archive (frontier)
Archive Cost Model (frontier)
Archive Restore (frontier)
Archive Monitoring (frontier)
Archive Anti-Patterns (frontier)

Frontier under this node: Archive Anti-Patterns, Archive Cost Model, Archive Monitoring, Archive Restore, Archive Rules, Federated Query on Archive, Partition Fields

← the whole tree · 3D view· how to read this page