Atlas compute auto-scaling for sharded clusters
Parent: Independent shard scaling and per-shard tiering · Published reference · snapshot 2026-10-23 · skill mongodb-cohort-1d6ef7b4932c
↓ Facts as markdownall context files
Source-anchored research for Atlas compute auto-scaling for sharded clusters. 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
- For Atlas Core sharded clusters, Atlas applies the same compute tier auto-scaling criteria as for replica sets. When independent shard scaling is active, Atlas evaluates shards separately and changes the tier only for the operational or analytics node class on a shard that meets a scaling criterion. (high confidence) (verified as of 2026-10-20) [source]
- Declarative infrastructure controllers can conflict with Atlas auto-scaling. A Crossplane issue reports that reconciliation reset Atlas-scaled cluster tiers to the declared specification. HashiCorp documents `ignore_changes` for attributes managed by an external process, and MongoDB's Terraform provider guide recommends `use_effective_fields` or lifecycle ignore rules for Atlas-managed size changes. (medium confidence) (verified as of 2026-10-20) [source]
- MongoDB's Terraform provider guide says its new sharding configuration enables per-shard instance-size auto-scaling when each shard has a unique `replication_specs` element; this schema removes `num_shards` in provider v2.0.0. Provider PR #2836 reports that a manually tested legacy `num_shards` configuration returned `autoScalingMode: CLUSTER`, while the separate-shard configuration returned `SHARD`. (high confidence) (verified as of 2026-10-23) [source]
- MongoDB's Terraform provider documents `use_effective_fields = true` for dedicated clusters (M10+) as a way to preserve configured specs while exposing Atlas-managed effective values through the data source, avoiding `lifecycle.ignore_changes` when auto-scaling stays enabled. The provider says this setting currently has no effect when a cluster has more than one `replication_specs` entry and recommends not enabling it for multi-shard clusters. (low confidence) (verified as of 2026-10-23) [source]
Problems, failure modes and limitations
- Atlas UI can enable compute auto-scaling so shards scale independently, but it cannot manually set each shard's tier. Per-shard tier changes use the Atlas Administration API, Atlas CLI, or Terraform; after independently scaling shards through those interfaces, the Atlas Cluster Builder UI cannot manage that cluster. (high confidence) (verified as of 2026-10-20) [source]
Facts and statements
- Atlas documents cluster-wide and independent shard scaling as separate modes. It says sharded clusters created before Dec 9, 2024 default to cluster-wide scaling. The current Atlas CLI guide says omitting `--autoScalingMode` defaults to `clusterWideScaling`; it describes independent mode through a `replicationSpecs` object that represents an individual shard. (high confidence) (verified as of 2026-10-23) [source]
- The MongoDB Terraform provider guide ties independent compute auto-scaling to the new sharding schema: each shard needs a unique `replication_specs` element, while the legacy `num_shards` grouping remains cluster-wide. A provider change compared the old schema's `autoScalingMode: CLUSTER` with the new schema's `autoScalingMode: SHARD`. (high confidence) [source]
- Atlas CLI exposes `clusterWideScaling` and `independentShardScaling` through `--autoScalingMode`; its documentation says omitting the option defaults to `clusterWideScaling`. Atlas's sharding page says clusters created before Dec 9, 2024 use cluster-wide scaling by default. (medium confidence) (verified as of 2026-10-20) [source]
- Infrastructure-as-code reconciliation can conflict with Atlas auto-scaling: HashiCorp says Terraform normally plans remote settings back to configuration unless attributes are ignored, and a Crossplane provider user reported Atlas-scaled tier and storage values being reverted during reconciliation. Crossplane maintainers later suggested `spec.initProvider` and closed the issue as completed, but the issue does not identify a released provider version containing the change. (medium confidence) (verified as of 2026-10-23) [source]
Children
- No children recorded.