M30 specs & connection limits
Parent: Atlas cluster tiers (M30) & major-version upgrade mechanics · Published reference · snapshot 2026-10-23 · viability 3.45 / 5 (research)
↓ Facts as markdownall context files
Source-anchored research for M30 specs & connection limits. 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
- M30 General has a documented admission rate of 24 new connections per second per node when rate limiting applies. For M30 and higher it is always on with MongoDB 9.0 or later and always off below 9.0. Overload queues attempts and sustained overload can drop connections; this is separate from the 3,000 concurrent-connection ceiling. (high confidence) (verified as of 2026-10-20) [source]
Measurements and reference values
- M30 has nominally 8 GB RAM and 2 vCPUs: the AWS reference and a practitioner comparison agree on this size. Actual available RAM can be slightly lower because the kernel reserves memory; provider and generation still matter. (medium confidence) (verified as of 2026-10-18) [source]
- For AWS Atlas Core M30, the documented default storage is 40 GB and the configurable range is 10–512 GB. Storage capacity is configurable independently within tier constraints; 40 GB is not a universal fixed M30 disk size. (low confidence) (verified as of 2026-10-18) [source]
- For AWS Gen2 General-class M30, MongoDB lists the preferred m8g.large instance with 2 vCPUs and 8 GiB RAM. This mapping applies specifically to Gen2 and is not a universal M30 instance-family guarantee. (low confidence) (verified as of 2026-10-18) [source]
- M30 storage performance is deployment-specific. The AWS reference lists Atlas Infinite M30 at a reference 5,000 IOPS and 80 MB/s, assuming 16 kB I/O, while stating that Atlas Core storage performance scales with storage capacity. These figures are not a universal M30 throughput guarantee. (low confidence) (verified as of 2026-10-18) [source]
- A standard M30 Atlas Core replica set has 8 GB RAM per data-bearing node. AWS Gen2 maps M30 to m8g.large with 2 vCPUs and 8 GiB RAM. The tier name describes per-node resources, not a single shared cluster memory budget. (high confidence) (verified as of 2026-10-18) [source]
Problems, failure modes and limitations
- The general-class M30 concurrent incoming connection ceiling is 3,000 per node on AWS, Azure and Google Cloud. In a sharded deployment, the documented limit applies per mongos router. (high confidence) (verified as of 2026-10-20) [source]
- A three-node replica set does not give its primary a 9,000-connection allowance: the 3,000 M30 limit is per node, and secondary-read capacity depends on read preference. Atlas reserves some connections for internal services, so applications should leave headroom. (medium confidence) (verified as of 2026-10-20) [source]
- Mongoose documents maxPoolSize=100 by default and one operation per socket at a time. A separate commenter reports that the default maxPoolSize was 5 in Mongoose 5 and 100 in Mongoose 6, and that their connection count fell after setting it to 10; this is a single report. Reducing the pool can relieve connection pressure, while increasing it can help operations blocked behind slow queries; size it against workload and the M30 connection budget. (medium confidence) (verified as of 2026-10-23) [source]
Comparisons and alternatives
- Atlas scales on compute and memory use rather than connection count. An application should budget the combined connection pools of its clients and leave room for monitoring and reserved connections; pool tuning is a documented mitigation before selecting a larger tier. (medium confidence) (verified as of 2026-10-20) [source]
- Connection pools cache reusable database sockets. Each MongoClient manages its own pool, so budget total application pools against the Atlas limit rather than interpreting maxPoolSize as a cluster-wide cap. Mongoose explicitly warns that total sockets across pools must remain within server and Atlas limits. (high confidence) [source]
Facts and statements
- M20 and M30 both have a documented general-class ceiling of 3,000 concurrent connections per node; moving from M20 to M30 alone does not increase that ceiling. M40 general class raises it to 6,000. (high confidence) (verified as of 2026-10-20) [source]
- Connection limits apply per node, or per mongos router for sharded clusters. Inference: summing three M30 node ceilings into a 9,000-connection budget for traffic directed entirely to one primary is invalid. Atlas also reserves some connections for its own services. (low confidence) (verified as of 2026-10-20) [source]
- An M30 issue records primary connections rising from about 800 to about 1,300 and stabilizing after an upgrade from Mongoose 5.13.14 / Node driver 3.7.3 to Mongoose 6.2.5 / Node driver 4.3.1 on MongoDB 4.4.13. The original reporter later retracted the suspected upgrade attribution, saying a concurrent application change caused rapid instance restarts and improperly closed Mongoose connections to remain until TTL expiry. This is one reported incident, not evidence of a general Mongoose or driver leak. Measure pool behavior after client upgrades. (low confidence) [source]
- In the M30 incident recorded in Mongoose issue #11596, the original reporter retracted the suspected driver-upgrade leak: another application change caused repeated instance death and restart, leaving improperly closed connections until expiry. This is an individual incident, not evidence of a general Mongoose leak. (low confidence) [source]
- Deyan Petrov reported that an Azure upgrade to M30 did not resolve intermittent disk latency in March 2022; his November 2024 update reports much steadier operation durations after Premium SSD v2. Equal M30 RAM and CPU labels therefore should not be treated as a workload performance guarantee. (low confidence) [source]
Children
- No children recorded.