# Analytics Node Sizing > Depth-first rabbithole dossier for Analytics Node Sizing; source-anchored research pack. Parent: [MongoDB Atlas Analytics Node](https://llms-explorer.com/tree/mongodb-atlas-analytics-node/) · 5 facets · 57 facts · page: https://llms-explorer.com/tree/analytics-node-sizing/ ## How it works - 1. On M10+ clusters, the analytics tier can be larger or smaller than the tier of the electable and read-only ("operational") nodes. https://www.mongodb.com/docs/atlas/cluster-config/multi-cloud-distribution/ [M H E P] 2. In the UI, you add analytics nodes first, then pick their tier on the **Analytics Tier** tab. The **Base Tier** tab sets the operational nodes. Same URL. [M H] 3. MongoDB's stated purpose is to avoid "over or under provisioning your entire cluster for your analytical workload". Same URL. [P] 4. Before 3 August 2022, analytics nodes "used the same cluster tier as all other nod — source: `~/.global-ai-hub/research-tests/mongodb-full-frontier-20261002/full-frontier-run/analytics-node-sizing-3bd55985af/rabbithole-synthesis.md#a-independent-analytics-tier-and-its-history` - 28. An analytics tier costs the same as the same tier on the Base Tier tab. When the two differ, Atlas prorates the price per node. multi-cloud-distribution URL. [M H E P] — source: `~/.global-ai-hub/research-tests/mongodb-full-frontier-20261002/full-frontier-run/analytics-node-sizing-3bd55985af/rabbithole-synthesis.md#d-pricing` - 29. If the Analytics Tier is "significantly below" the Base Tier, replication lag can follow, and the node "might fall off the oplog altogether". No source puts a number on "significantly". Same URL. [M H E P] 30. Secondaries apply the oplog asynchronously. A secondary can catch up only within the oplog window. https://www.mongodb.com/docs/manual/core/replica-set-oplog/ [M] 31. The oplog grows past its configured size only to protect the majority commit point. Same URL. [M] 32. Only data-bearing members with `votes > 0` count toward the majority. https://www.mongodb.com/docs/manual/reference/w — source: `~/.global-ai-hub/research-tests/mongodb-full-frontier-20261002/full-frontier-run/analytics-node-sizing-3bd55985af/rabbithole-synthesis.md#e-undersizing-replication-lag-and-falling-off-the-oplog` - https://www.mongodb.com/docs/atlas/reference/alert-conditions/ ; https://www.mongodb.com/docs/atlas/reference/alert-resolutions/atlas-autoscaling-alerts/ [H E P] 54. *Inference:* replication lag is not an auto-scaling input. A node that lags without being CPU- or RAM-bound will not scale up. compute-core URL. [E] 55. Sudden bursts may not trigger scaling in time. MongoDB advises scaling up before planned spikes. Same URL. [E] 56. Atlas Infinite uses the same analytics scale-up and scale-down rules. compute-infinite URL. [P] — source: `~/.global-ai-hub/research-tests/mongodb-full-frontier-20261002/full-frontier-run/analytics-node-sizing-3bd55985af/rabbithole-synthesis.md#f-analytics-auto-scaling` - Run: /rabbithole, single mechanism pass, 2026-10-02. Parent: MongoDB Atlas Analytics Node. — source: `~/.global-ai-hub/research-tests/mongodb-full-frontier-20261002/full-frontier-run/analytics-node-sizing-3bd55985af/reports/mechanism.md` - 1. On `M10+` clusters, you can give analytics nodes a cluster tier that is larger or smaller than the tier of the electable and read-only ("operational") nodes. — https://www.mongodb.com/docs/atlas/cluster-config/multi-cloud-distribution/#select-a-cluster-tier-for-your-analytics-nodes 2. The UI has two tabs for this: the **Analytics Tier** tab for analytics nodes and the **Base Tier** tab for operational nodes. You add analytics nodes first, then pick their tier on the Analytics Tier tab. — https://www.mongodb.com/docs/atlas/cluster-config/multi-cloud-distribution/#select-a-cluster-tier-for-yo — source: `~/.global-ai-hub/research-tests/mongodb-full-frontier-20261002/full-frontier-run/analytics-node-sizing-3bd55985af/reports/mechanism.md#a-independent-analytics-tier` - 1. You can give analytics nodes a cluster tier that is larger or smaller than the tier of the electable and read-only (operational) nodes. — https://www.mongodb.com/docs/atlas/cluster-config/multi-cloud-distribution/ 2. Atlas added independent analytics tiers in August 2022. Before that, analytics nodes used the same tier as all other nodes. — https://www.mongodb.com/company/blog/product-release-announcements/introducing-ability-independently-scale-atlas-analytics-node-tiers 3. MongoDB recommends a larger analytics tier when many BI-dashboard users or memory-heavy analytics need it. It recomme — source: `~/.global-ai-hub/research-tests/mongodb-full-frontier-20261002/full-frontier-run/analytics-node-sizing-3bd55985af/reports/practice.md#tier-asymmetry-and-its-purpose` - 14. An Analytics Tier costs the same as the same tier chosen on the Base Tier tab. When the analytics tier is higher or lower than the base tier, Atlas adjusts the price on a prorated per-node basis. — https://www.mongodb.com/docs/atlas/cluster-config/multi-cloud-distribution/#select-a-cluster-tier-for-your-analytics-nodes — source: `~/.global-ai-hub/research-tests/mongodb-full-frontier-20261002/full-frontier-run/analytics-node-sizing-3bd55985af/reports/mechanism.md#c-pricing-mechanism` - - Pass 0 (single mechanism pass, as the brief asked): 35 atomic claims, 2 marked as inference. Saturation was not tested. Verdict: `BUDGET_EXHAUSTED` (single-pass brief). A second pass would probably still find new facts in: per-cloud analytics tier availability, the Admin API `analyticsSpecs` defaults when the field is omitted, and AKO `analyticsSpecs` validation. — source: `~/.global-ai-hub/research-tests/mongodb-full-frontier-20261002/full-frontier-run/analytics-node-sizing-3bd55985af/reports/mechanism.md#pass-log` ## Measurements and reference values - 60. The WiredTiger cache is 25% of RAM on M30 and below, and 50% or more on M40 and above. Stepping down across the M30/M40 line therefore cuts cache more than RAM. Plan for RAM of at least twice the working set. Analytics scans tend to favour a larger tier. https://www.mongodb.com/docs/atlas/sizing-tier-selection/ [E P] 61. Sustained load on M10/M20 degrades performance, and M20 analytics nodes have been reported in use. Same URL, and forum URL from claim 42. [E] 62. Connection limits scale with tier: M30 allows 3,000 and M200 allows 128,000. Many concurrent BI users can therefore cap a small — source: `~/.global-ai-hub/research-tests/mongodb-full-frontier-20261002/full-frontier-run/analytics-node-sizing-3bd55985af/rabbithole-synthesis.md#h-sizing-inputs-that-carry-over-from-general-tier-sizing` - 43. `analyticsAutoScaling` is a separate object with `compute.{enabled, scaleDownEnabled, minInstanceSize, maxInstanceSize}` and `diskGB.enabled`. https://www.mongodb.com/docs/atlas/cluster-autoscaling-compute-core/ ; https://www.mongodb.com/docs/atlas/cluster-autoscaling-compute-infinite/ [M H E P] 44. **Scale up** one tier when average Normalized System CPU **or** System Memory is above 75% for 1 hour. compute-core URL. [M H E P] 45. The operational-only triggers do not apply to analytics nodes: 90% for 10 or 20 minutes, and the CPU-steal and softIRQ rules on M10/M20. Same URL. [M E] 46. The — source: `~/.global-ai-hub/research-tests/mongodb-full-frontier-20261002/full-frontier-run/analytics-node-sizing-3bd55985af/rabbithole-synthesis.md#f-analytics-auto-scaling` - 17. Analytics auto-scaling is configured on its own, in `analyticsAutoScaling.compute`, with separate enabled, scale-down, min and max fields. https://www.mongodb.com/docs/atlas/cluster-autoscaling-compute-core/ In Terraform, `analytics_auto_scaling.compute_enabled` is false by default. https://github.com/mongodb/terraform-provider-mongodbatlas/blob/master/docs/resources/advanced_cluster.md 18. Scale-up rule for analytics nodes: average Normalized System CPU **or** System Memory above 75% for the past hour. The fast 90%-for-10-minutes triggers that apply to operational nodes do not apply. http — source: `~/.global-ai-hub/research-tests/mongodb-full-frontier-20261002/full-frontier-run/analytics-node-sizing-3bd55985af/reports/edge-cases.md#c-auto-scaling-the-analytics-tier` - 18. You configure analytics auto-scaling separately from base-node auto-scaling. The Admin API object is `analyticsAutoScaling.compute` {`enabled`, `scaleDownEnabled`, `minInstanceSize`, `maxInstanceSize`}, plus `analyticsAutoScaling.diskGB.enabled`. — https://www.mongodb.com/docs/atlas/cluster-autoscaling-compute-core/ 19. Scale-up: Atlas moves analytics nodes up one tier when average Normalized System CPU or System Memory Utilization is above 75% for the past hour. The Queued/Rejected Operations criterion does not apply to analytics nodes. — https://www.mongodb.com/docs/atlas/cluster-autosca — source: `~/.global-ai-hub/research-tests/mongodb-full-frontier-20261002/full-frontier-run/analytics-node-sizing-3bd55985af/reports/history.md#analytics-compute-auto-scaling` - 23. Analytics auto-scaling is a separate `analyticsAutoScaling` object. It has its own `compute.enabled`, `scaleDownEnabled`, `minInstanceSize`, `maxInstanceSize` and `diskGB.enabled`. — https://www.mongodb.com/docs/atlas/cluster-autoscaling-compute-core/#enable-auto-scaling-with-atlas-cli-and-atlas-administration-api 24. Terraform requires `analytics_auto_scaling` to have the same values in every `region_configs` entry of a cluster. — https://raw.githubusercontent.com/mongodb/terraform-provider-mongodbatlas/master/docs/resources/advanced_cluster.md 25. Scale-up rule for analytics nodes (any c — source: `~/.global-ai-hub/research-tests/mongodb-full-frontier-20261002/full-frontier-run/analytics-node-sizing-3bd55985af/reports/mechanism.md#e-analytics-node-auto-scaling-separate-from-base-auto-scaling` - 19. Analytics nodes have their own compute auto-scaling settings (`analyticsAutoScaling.compute.enabled / scaleDownEnabled / minInstanceSize / maxInstanceSize`), separate from the base tier's settings. — https://www.mongodb.com/docs/atlas/cluster-autoscaling-compute-infinite/ 20. Scale-up rule: Atlas moves analytics nodes up one tier if average Normalized System CPU **or** System Memory Utilization is above 75% for the past hour. The Queued or Rejected Operations criterion does not apply to analytics nodes. — https://www.mongodb.com/docs/atlas/cluster-autoscaling-compute-core/ 21. Scale-down r — source: `~/.global-ai-hub/research-tests/mongodb-full-frontier-20261002/full-frontier-run/analytics-node-sizing-3bd55985af/reports/practice.md#auto-scaling-of-analytics-nodes` - Verdict: BUDGET_EXHAUSTED (soft stop, not SATURATED-DEPTH). The rate is falling but has not had two passes below 5%. A further pass is likely to add a little: the analytics auto-scaling launch date, and whether the storage-match rule predates August 2022 (via Wayback snapshots or older raw docs on raw.githubusercontent.com/mongodb/docs). — source: `~/.global-ai-hub/research-tests/mongodb-full-frontier-20261002/full-frontier-run/analytics-node-sizing-3bd55985af/reports/history.md#depth-pass-curve` ## Problems, failure modes and limitations - - **How small is safe.** The 2022 announcement presents a smaller analytics tier as a plain cost lever with no caveat (claim 3). The Atlas docs warn that a tier "significantly below" the base tier can cause lag and oplog fall-off (claim 15). A forum expert argues the analytics node needs *at least* base-tier write capacity, and maybe more (claim 21). No source defines "significantly", for example as a number of tiers. Contrast: independent shard scaling has an explicit rule that shards may differ by at most two tiers (https://www.mongodb.com/docs/atlas/cluster-autoscaling-compute-core/#scaling — source: `~/.global-ai-hub/research-tests/mongodb-full-frontier-20261002/full-frontier-run/analytics-node-sizing-3bd55985af/reports/mechanism.md#unresolved-disagreements` - 28. WiredTiger cache is 25% of RAM on M30 and smaller and 50% or more on M40+. Moving the analytics tier down across the M30/M40 line therefore cuts its cache more than its RAM alone suggests. https://www.mongodb.com/docs/atlas/sizing-tier-selection/ 29. MongoDB says sustained load on M10 and M20 can degrade performance over time. M20 analytics nodes, the smallest choice users report trying, fall in that range. https://www.mongodb.com/docs/atlas/sizing-tier-selection/ · https://www.mongodb.com/community/forums/t/low-tier-analytic-node-effects/253955 30. Connection limits depend on the tier (M3 — source: `~/.global-ai-hub/research-tests/mongodb-full-frontier-20261002/full-frontier-run/analytics-node-sizing-3bd55985af/reports/edge-cases.md#d-sizing-inputs-that-carry-over` - - Independent documents used: Atlas multi-cloud and workload-isolation docs; the compute auto-scaling docs; the sizing docs; the alert-conditions and oplog-resolution docs; the server manual (troubleshooting, replication, write concern); the 2022 launch blog; a community forum thread; and the Terraform provider docs on GitHub. That is more than 3 documents, across 2 hosts (mongodb.com, github.com). - **Host independence is weak.** Every authoritative source is written by MongoDB, including the Terraform provider. The only third-party view is one non-staff forum answer, hosted on mongodb.com. O — source: `~/.global-ai-hub/research-tests/mongodb-full-frontier-20261002/full-frontier-run/analytics-node-sizing-3bd55985af/reports/edge-cases.md#quality-gate` - In scope: how you choose and change the cluster tier of Atlas analytics nodes. That covers the independent "Analytics Tier", the limits on mixing tiers, analytics compute auto-scaling, the sizing failure modes, pricing, and how all of this changed over time. — source: `~/.global-ai-hub/research-tests/mongodb-full-frontier-20261002/full-frontier-run/analytics-node-sizing-3bd55985af/reports/history.md#scope` - In scope: how Atlas sizes analytics nodes. That covers the separate Analytics Tier, its coupling to the Base Tier (storage, IOPS, class), its auto-scaling rules, how a tier change reaches the nodes, and the replication limit on undersizing. Out of scope: what analytics nodes are, read-preference routing, BI Connector/Atlas SQL, Search Nodes, read-only nodes, and general cluster sizing. Parent facts are not repeated here except where a sizing claim depends on them. — source: `~/.global-ai-hub/research-tests/mongodb-full-frontier-20261002/full-frontier-run/analytics-node-sizing-3bd55985af/reports/mechanism.md#scope` - 7. Disk size and IOPS must stay the same across all node types. — https://www.mongodb.com/docs/atlas/cluster-config/multi-cloud-distribution/#select-a-cluster-tier-for-your-analytics-nodes 8. Storage size must match between the Base Tier and Analytics Tier tabs. You set it only on the Base Tier tab. — same URL as claim 7 9. The Terraform provider says `disk_size_gb` "must be equal for all shards and node types", even though `analytics_specs` has its own `disk_size_gb` and `disk_iops` fields. — https://raw.githubusercontent.com/mongodb/terraform-provider-mongodbatlas/master/docs/resources/advan — source: `~/.global-ai-hub/research-tests/mongodb-full-frontier-20261002/full-frontier-run/analytics-node-sizing-3bd55985af/reports/mechanism.md#b-invariants-that-tie-the-analytics-tier-to-the-base-tier-atlas-core` - 28. The usual working-set rule applies to whichever tier you pick. On M40 and larger, WiredTiger takes 50% or more of RAM, so RAM should be at least twice the working set. On M30 and smaller, WiredTiger takes only 25% of RAM. Analytics scans often touch more data than OLTP, so this rule usually favours a larger analytics tier, not a smaller one. — https://www.mongodb.com/docs/atlas/sizing-tier-selection/ 29. Connection limits grow with tier (for example, M30 allows 3000 connections and M200 allows 128,000). Many concurrent BI users can therefore limit a small analytics tier through connections — source: `~/.global-ai-hub/research-tests/mongodb-full-frontier-20261002/full-frontier-run/analytics-node-sizing-3bd55985af/reports/practice.md#sizing-inputs` - 21. Disk size and IOPS must be the same for every node type. You set storage only on the Base Tier tab. So the analytics tier can change CPU and RAM, but not storage. https://www.mongodb.com/docs/atlas/cluster-config/multi-cloud-distribution/ [M H E P] 22. Terraform says `disk_size_gb` "must be equal for all shards and node types". Terraform URL from claim 15. [M H P] 23. If the Base Tier uses Local NVMe SSD, the Analytics Tier must use the same tier level. NVMe clusters therefore cannot size analytics independently. multi-cloud-distribution URL. [M H E P] 24. The UI greys out analytics tiers — source: `~/.global-ai-hub/research-tests/mongodb-full-frontier-20261002/full-frontier-run/analytics-node-sizing-3bd55985af/rabbithole-synthesis.md#c-what-ties-the-analytics-tier-to-the-base-tier-atlas-core` - 57. Atlas applies tier changes as a rolling update, one node at a time. Each node is unavailable during its own upgrade. compute-core URL. [M] 58. Reads pinned to `nodeType:ANALYTICS` can fail during an initial sync, a disk scale-down, or an NVMe tier change. Atlas recommends adding a fallback `readPreferenceTags=` entry. https://www.mongodb.com/docs/atlas/reference/replica-set-tags/ [M P] 59. *Inference from claims 57–58:* with one analytics node and strict tags, a tier change interrupts analytics reads. Two or more nodes avoid this. A fallback tag also avoids it, but gives up isolation. [M P — source: `~/.global-ai-hub/research-tests/mongodb-full-frontier-20261002/full-frontier-run/analytics-node-sizing-3bd55985af/rabbithole-synthesis.md#g-what-a-resize-does-at-runtime` - - **How small is safe.** - The 2022 blog presents a smaller tier as a cost lever with no caveat (claim 11). - The docs warn that "significantly below" can mean oplog fall-off (claim 29). - The forum expert says at least base write capacity is needed, possibly more (claim 42). - Shards have an explicit rule: they may differ by at most two tiers (compute-core URL). No equivalent rule exists between the analytics and base tiers. - **Storage independence.** - In 2022, analysts framed it as an IOPS-heavy operational node against a compute-heavy analytics node (siliconangle). - The current docs requ — source: `~/.global-ai-hub/research-tests/mongodb-full-frontier-20261002/full-frontier-run/analytics-node-sizing-3bd55985af/rabbithole-synthesis.md#disagreements-kept-side-by-side` - - A numeric safe gap between the analytics tier and the base tier. - Whether a manual tier change on a non-NVMe cluster causes only a rolling restart or also an initial sync. - Per-region limits on the number of analytics nodes. - Whether an analytics node uses its own tier's connection limit or the base tier's. - The launch date of analytics auto-scaling. - Whether the storage-match rule existed in August 2022. - Defaults when `analyticsSpecs` is left out of an Admin API request. - How the AKO operator validates `analyticsSpecs`. — source: `~/.global-ai-hub/research-tests/mongodb-full-frontier-20261002/full-frontier-run/analytics-node-sizing-3bd55985af/rabbithole-synthesis.md#open-gaps-no-source-in-any-report` - 10. MongoDB documents this risk itself. If the Analytics Tier is "significantly below" the Base Tier, replication lag can follow, and the analytics node "might fall off the oplog altogether". The docs do not define how big a gap counts as "significantly". https://www.mongodb.com/docs/atlas/cluster-config/multi-cloud-distribution/ 11. A node that falls off the oplog logs `We are too stale to use :27017 as a sync source`. It stays in STARTUP2 or RECOVERING for a long time and needs a full initial sync to recover. https://www.mongodb.com/docs/atlas/reference/alert-resolutions/replication-op — source: `~/.global-ai-hub/research-tests/mongodb-full-frontier-20261002/full-frontier-run/analytics-node-sizing-3bd55985af/reports/edge-cases.md#b-failure-mode-an-undersized-analytics-tier` - 1. Until 3 August 2022, analytics nodes used the same cluster tier as every other node in the cluster: "Until now, analytics nodes in MongoDB's Atlas clusters have used the same cluster tier as all other nodes." — https://www.mongodb.com/company/blog/product-release-announcements/introducing-ability-independently-scale-atlas-analytics-node-tiers 2. Analyst coverage of MongoDB World 2022 (8 June 2022) previewed the capability: "Analytical nodes will be able to be scaled individually, allowing independent fine-tuning of performance." — https://siliconangle.com/2022/06/08/mongodb-announces-abilit — source: `~/.global-ai-hub/research-tests/mongodb-full-frontier-20261002/full-frontier-run/analytics-node-sizing-3bd55985af/reports/history.md#history-and-evolution` - 15. MongoDB documents this risk directly. If the Analytics Tier is significantly below the Base Tier, replication lag can follow, and the analytics node can fall off the oplog altogether. — https://www.mongodb.com/docs/atlas/cluster-config/multi-cloud-distribution/#select-a-cluster-tier-for-your-analytics-nodes 16. Mechanism: secondaries copy and apply the primary's oplog asynchronously. A secondary can catch up through replication only within the oplog window, which is the time between the newest and oldest oplog entries. — https://www.mongodb.com/docs/manual/core/replica-set-oplog/ 17. The o — source: `~/.global-ai-hub/research-tests/mongodb-full-frontier-20261002/full-frontier-run/analytics-node-sizing-3bd55985af/reports/mechanism.md#d-undersizing-limit-replication-lag-and-falling-off-the-oplog` - 32. Atlas applies tier changes as a rolling update. It upgrades nodes one at a time, and each node is unavailable during its own upgrade. — https://www.mongodb.com/docs/atlas/cluster-autoscaling-compute-core/#reactive-auto-scaling-for-cluster-tier 33. Atlas warns that reads pinned to `nodeType:ANALYTICS` can fail during initial sync, disk scale-down, and any NVMe tier change. It recommends adding a fallback `readPreferenceTags=` entry. — https://www.mongodb.com/docs/atlas/reference/replica-set-tags/ 34. *Inference from claims 32–33:* with a single analytics node and a strict `secondary` + `nod — source: `~/.global-ai-hub/research-tests/mongodb-full-frontier-20261002/full-frontier-run/analytics-node-sizing-3bd55985af/reports/mechanism.md#f-what-a-resize-does-at-runtime` - 7. If the analytics tier is "significantly below" the base tier, replication lag can result. The analytics node "might fall off the oplog altogether". — https://www.mongodb.com/docs/atlas/cluster-config/multi-cloud-distribution/ 8. The docs give no numeric threshold for "significantly below". No source found states a safe tier gap. — https://www.mongodb.com/docs/atlas/cluster-config/multi-cloud-distribution/ 9. A member that falls off the oplog needs a resync. A logical initial sync clones every collection and builds every index. It also "reduces the performance of the primary from which the d — source: `~/.global-ai-hub/research-tests/mongodb-full-frontier-20261002/full-frontier-run/analytics-node-sizing-3bd55985af/reports/practice.md#risk-of-undersizing` - 12. On Atlas Core, disk size and IOPS must be the same for all node types. Storage size must match between the Base Tier and Analytics Tier tabs, and you set it on the Base Tier tab. — https://www.mongodb.com/docs/atlas/cluster-config/multi-cloud-distribution/ 13. In Terraform, `disk_size_gb` "must be equal for all shards and node types". — https://raw.githubusercontent.com/mongodb/terraform-provider-mongodbatlas/master/docs/resources/advanced_cluster.md 14. On Atlas Core, if you choose Local NVMe SSD on the Base Tier, the Analytics Tier must use the same tier level. You cannot size NVMe analy — source: `~/.global-ai-hub/research-tests/mongodb-full-frontier-20261002/full-frontier-run/analytics-node-sizing-3bd55985af/reports/practice.md#storage-iops-and-cluster-class-rules-atlas-core` ## Comparisons and alternatives - 8. On M10+ clusters, you pick the analytics node tier on an "Analytics Tier" tab. It can be larger or smaller than the tier on the "Base Tier" tab, which sets the electable and read-only (operational) nodes. — https://www.mongodb.com/docs/atlas/cluster-config/multi-cloud-distribution/ 9. An Analytics Tier is priced the same as the same tier on the Base Tier tab. If the analytics tier differs from the base tier, the price adjusts on a prorated per-node basis. — https://www.mongodb.com/docs/atlas/cluster-config/multi-cloud-distribution/ 10. Terraform `mongodbatlas_advanced_cluster` sets the anal — source: `~/.global-ai-hub/research-tests/mongodb-full-frontier-20261002/full-frontier-run/analytics-node-sizing-3bd55985af/reports/history.md#current-sizing-mechanism` - Blog URL from claim 4. [M H P] 12. Third-party FinOps guidance says to move reporting and aggregation work onto analytics nodes with their own tier, rather than sizing every operational node for it. https://www.pointfive.co/efficiency-hub/inefficiencies/overprovisioned-mongodb-atlas-cluster-tier [H P] 13. PointFive's signals for overprovisioning, watched over several weeks, are Normalized System CPU, System Memory, WiredTiger cache, Opcounters and Connections. Same URL. [P] — source: `~/.global-ai-hub/research-tests/mongodb-full-frontier-20261002/full-frontier-run/analytics-node-sizing-3bd55985af/rabbithole-synthesis.md#a-independent-analytics-tier-and-its-history` - 25. MongoDB's guidance is to size the analytics tier above the operational tier when many BI users or memory-heavy analytics need it, because "scaling up just your analytics node tiers helps optimize the cost." — https://www.mongodb.com/company/blog/product-release-announcements/introducing-ability-independently-scale-atlas-analytics-node-tiers 26. MongoDB's guidance is to size the analytics tier below the operational tier for teams "with inconsistent needs" or where analytics is not the top priority. — https://www.mongodb.com/company/blog/product-release-announcements/introducing-ability-inde — source: `~/.global-ai-hub/research-tests/mongodb-full-frontier-20261002/full-frontier-run/analytics-node-sizing-3bd55985af/reports/history.md#vendor-sizing-guidance` - - **Do secondaries get resource priority over analytics nodes?** The inherited pmclain.com post (2022) says secondaries have "higher priority resource access than the analytics nodes", so analytics nodes lag under heavy writes (https://pmclain.com/mongodb/2022/09/28/fun-with-mongodb-atlas-analytics-nodes.html). No MongoDB document fetched in this run supports a resource-priority mechanism. The documented cause of analytics lag is an undersized tier (claim 7). Treat the pmclain claim as unverified. - **Self-managed pattern vs Atlas.** OneUptime (2026-03-31) describes a self-managed "analytics n — source: `~/.global-ai-hub/research-tests/mongodb-full-frontier-20261002/full-frontier-run/analytics-node-sizing-3bd55985af/reports/practice.md#unresolved-disagreements-and-gaps` - 14. In the Admin API, the analytics tier is `analyticsSpecs.instanceSize`, set per `regionConfigs` entry next to `electableSpecs`. https://www.mongodb.com/docs/atlas/cluster-autoscaling-compute-core/ [M] 15. In Terraform, `analytics_specs.instance_size` is required inside the `analytics_specs` block, and node count is set with `node_count`. https://raw.githubusercontent.com/mongodb/terraform-provider-mongodbatlas/master/docs/resources/advanced_cluster.md [M H] 16. Node count (`analyticsSpecs.nodeCount`) is separate from tier. The only cap on analytics nodes is the 50-node multi-region cluster — source: `~/.global-ai-hub/research-tests/mongodb-full-frontier-20261002/full-frontier-run/analytics-node-sizing-3bd55985af/rabbithole-synthesis.md#b-api-and-iac-surface` - 1. On M10+ clusters you can give analytics nodes a cluster tier larger or smaller than the tier of the electable and read-only nodes ("operational nodes"). https://www.mongodb.com/docs/atlas/cluster-config/multi-cloud-distribution/ 2. This feature was announced on 2022-08-03. Until then, analytics nodes always used the same tier as every other node in the cluster. https://www.mongodb.com/company/blog/product-release-announcements/introducing-ability-independently-scale-atlas-analytics-node-tiers 3. On Atlas Core, disk size and IOPS must be the same for every node type. Storage size must also m — source: `~/.global-ai-hub/research-tests/mongodb-full-frontier-20261002/full-frontier-run/analytics-node-sizing-3bd55985af/reports/edge-cases.md#a-independent-tier-selection-limits` - - **Smaller is fine vs. smaller is risky.** The launch blog presents a lower analytics tier as a cost saving for "teams with inconsistent needs" and gives no warning about lag. https://www.mongodb.com/company/blog/product-release-announcements/introducing-ability-independently-scale-atlas-analytics-node-tiers The current docs warn that a tier "significantly below" the base tier can fall off the oplog. https://www.mongodb.com/docs/atlas/cluster-config/multi-cloud-distribution/ A community answer goes further: an analytics node must handle **at least** the same write load as the other nodes and — source: `~/.global-ai-hub/research-tests/mongodb-full-frontier-20261002/full-frontier-run/analytics-node-sizing-3bd55985af/reports/edge-cases.md#unresolved-disagreements` - - **Storage profile, analyst framing vs. docs.** The 2022 analyst framing describes tailoring an IOPS-intensive operational node against a compute-intensive analytics node (https://siliconangle.com/2022/06/08/mongodb-announces-ability-for-atlas-to-separate-nodes-allowing-developers-to-customize-for-specific-workloads-mdbw22/). Current Atlas Core docs require the same disk size and IOPS across all node types (https://www.mongodb.com/docs/atlas/cluster-config/multi-cloud-distribution/). So only the compute/RAM tier is independent, not the storage profile. Both statements are kept; the docs are a — source: `~/.global-ai-hub/research-tests/mongodb-full-frontier-20261002/full-frontier-run/analytics-node-sizing-3bd55985af/reports/history.md#unresolved-disagreements-and-limits` - **In scope:** how to pick and change the cluster tier of MongoDB Atlas analytics nodes. That covers tier asymmetry versus the base tier, storage and IOPS rules, cluster-class mixing, pricing, auto-scaling, the effect of an undersized tier on lag and the oplog, and IaC fields. — source: `~/.global-ai-hub/research-tests/mongodb-full-frontier-20261002/full-frontier-run/analytics-node-sizing-3bd55985af/reports/practice.md#scope` - 26. In Terraform, all `analytics_specs` inside one `replication_specs` must have the same `instance_size`. Analytics tiers can differ from the base tier, but not between regions of the same shard. — https://raw.githubusercontent.com/mongodb/terraform-provider-mongodbatlas/master/docs/resources/advanced_cluster.md 27. With the new sharding configuration, analytics nodes scale independently of base nodes within a shard. Both base and analytics nodes can also scale independently of their equivalents in other shards, and analytics auto-scaling runs per shard. — https://raw.githubusercontent.com/mo — source: `~/.global-ai-hub/research-tests/mongodb-full-frontier-20261002/full-frontier-run/analytics-node-sizing-3bd55985af/reports/practice.md#iac-and-sharded-clusters` ## Facts and statements - **Out of scope:** what analytics nodes are, routing with read preference, BI Connector and Atlas SQL, Search Nodes, and general cluster sizing. Those belong to the parent and sibling items. This report does not repeat facts inherited from the parent. — source: `~/.global-ai-hub/research-tests/mongodb-full-frontier-20261002/full-frontier-run/analytics-node-sizing-3bd55985af/reports/edge-cases.md#scope` - **Out of scope:** what analytics nodes are, read-preference routing, BI Connector, Atlas SQL, Search Nodes, and general Atlas tier selection. Those belong to the parent topic or to sibling topics. Inherited parent claims (M10+, priority 0 / votes 0, read-only, `nodeType:ANALYTICS` tag) are not repeated here. They appear only where a sizing delta depends on them. — source: `~/.global-ai-hub/research-tests/mongodb-full-frontier-20261002/full-frontier-run/analytics-node-sizing-3bd55985af/reports/practice.md#scope` - - Independent shard scaling (the two-tier spread rule). - Atlas Infinite analytics behaviour. - IWM admission control on analytics nodes. - Predictive auto-scaling. - Minimum Oplog Window and how it interacts with analytics scale-up. - Low-CPU and NVMe cluster classes. — source: `~/.global-ai-hub/research-tests/mongodb-full-frontier-20261002/full-frontier-run/analytics-node-sizing-3bd55985af/rabbithole-synthesis.md#handoffs-for-concept-family-explorer-not-chased` - **In scope:** how to choose the cluster tier for Atlas analytics nodes, separately from the base tier. That covers the limits on choosing a different tier, what goes wrong if you pick one too small, how auto-scaling behaves for the analytics tier, and how billing works. — source: `~/.global-ai-hub/research-tests/mongodb-full-frontier-20261002/full-frontier-run/analytics-node-sizing-3bd55985af/reports/edge-cases.md#scope` - - Minimum oplog window setting (`Set Minimum Oplog Window`) and how it interacts with analytics scale-up. - How IWM load shedding behaves on analytics nodes. - Analytics node pricing and node limits on Atlas Infinite. — source: `~/.global-ai-hub/research-tests/mongodb-full-frontier-20261002/full-frontier-run/analytics-node-sizing-3bd55985af/reports/edge-cases.md#handoffs-out-of-scope-not-chased` - Run: /rabbithole, 2026-10-02. Parent: MongoDB Atlas Analytics Node. — source: `~/.global-ai-hub/research-tests/mongodb-full-frontier-20261002/full-frontier-run/analytics-node-sizing-3bd55985af/reports/history.md` - Out of scope: what analytics nodes are, read-preference routing, BI Connector, Data Federation, and Search Nodes. The parent and its siblings cover those. Parent claims are not repeated here as findings. — source: `~/.global-ai-hub/research-tests/mongodb-full-frontier-20261002/full-frontier-run/analytics-node-sizing-3bd55985af/reports/history.md#scope` - 11. Disk size and IOPS must stay the same across all node types. Storage size must match between the Base Tier and Analytics Tier tabs. — https://www.mongodb.com/docs/atlas/cluster-config/multi-cloud-distribution/ 12. Terraform's `disk_size_gb` "must be equal for all shards and node types." — https://raw.githubusercontent.com/mongodb/terraform-provider-mongodbatlas/master/docs/resources/advanced_cluster.md 13. If the Base Tier uses the Local NVMe SSD class, the Analytics Tier must use the same tier level. — https://www.mongodb.com/docs/atlas/cluster-config/multi-cloud-distribution/ 14. If one — source: `~/.global-ai-hub/research-tests/mongodb-full-frontier-20261002/full-frontier-run/analytics-node-sizing-3bd55985af/reports/history.md#constraints-on-mixing-tiers-atlas-core` - 16. If the Analytics Tier is significantly below the Base Tier, replication lag can result, and "the analytics node might fall off the oplog altogether." — https://www.mongodb.com/docs/atlas/cluster-config/multi-cloud-distribution/ 17. By default, Atlas keeps oplog entries for 24 hours. A minimum oplog window is configurable. That window is the margin an undersized analytics node has before it falls off the oplog. — https://www.mongodb.com/docs/atlas/cluster-additional-settings/ — source: `~/.global-ai-hub/research-tests/mongodb-full-frontier-20261002/full-frontier-run/analytics-node-sizing-3bd55985af/reports/history.md#undersizing-failure-mode` - Hosts used: mongodb.com (docs and blog), raw.githubusercontent.com (MongoDB-authored Terraform provider docs), pointfive.co, and oneuptime.com. The inherited pmclain.com page is used only for the disagreement and is not counted. The host count therefore meets the 3-source gate. **Substantive independence is weak.** Every concrete sizing rule comes from MongoDB-owned material. PointFive independently confirms only the asymmetric-tier cost practice. OneUptime adds no Atlas analytics-tier facts. No independent benchmark or field report on analytics tier sizing was found. Treat claims 7–8 and 20–2 — source: `~/.global-ai-hub/research-tests/mongodb-full-frontier-20261002/full-frontier-run/analytics-node-sizing-3bd55985af/reports/practice.md#quality-gate` - **Source gate: met on host count, weak on independence.** Three new hosts carry real claims: mongodb.com, siliconangle.com and pointfive.co. raw.githubusercontent.com and github.com hold the Terraform provider, which MongoDB writes. oneuptime.com supplied only disagreements. pmclain.com is inherited from the parent and is not counted. Every concrete sizing rule comes from MongoDB. — source: `~/.global-ai-hub/research-tests/mongodb-full-frontier-20261002/full-frontier-run/analytics-node-sizing-3bd55985af/rabbithole-synthesis.md` - - https://www.mongodb.com/docs/atlas/cluster-config/multi-cloud-distribution/ - https://www.mongodb.com/docs/atlas/cluster-autoscaling-compute-core/ - https://www.mongodb.com/docs/atlas/sizing-tier-selection/ - https://www.mongodb.com/docs/atlas/reference/alert-conditions/ - https://www.mongodb.com/docs/atlas/reference/alert-resolutions/replication-oplog/ - https://www.mongodb.com/docs/manual/tutorial/troubleshoot-replica-sets/ - https://www.mongodb.com/docs/manual/replication/ - https://www.mongodb.com/docs/manual/reference/write-concern/ - https://www.mongodb.com/company/blog/product-release — source: `~/.global-ai-hub/research-tests/mongodb-full-frontier-20261002/full-frontier-run/analytics-node-sizing-3bd55985af/reports/edge-cases.md#sources` - - Sources used (excluding the inherited pmclain.com page): Atlas docs (workload isolation, compute auto-scaling, auto-scaling overview, replica-set tags, deployment FAQ), MongoDB Server manual (replication, oplog, write concern), the MongoDB product blog, the MongoDB Community Forum (a non-employee answer), and the Terraform provider docs on raw.githubusercontent.com. - **The gate is only partly met.** The run has 5 distinct publications and authors, but only 2 hosts (mongodb.com and raw.githubusercontent.com), and MongoDB controls both. I looked for a source outside MongoDB. https://community — source: `~/.global-ai-hub/research-tests/mongodb-full-frontier-20261002/full-frontier-run/analytics-node-sizing-3bd55985af/reports/mechanism.md#quality-gate` - 18. An analytics tier costs the same as the same tier on the Base Tier tab. If the analytics tier is higher or lower than the base tier, the price adjusts on a prorated per-node basis. — https://www.mongodb.com/docs/atlas/cluster-config/multi-cloud-distribution/ — source: `~/.global-ai-hub/research-tests/mongodb-full-frontier-20261002/full-frontier-run/analytics-node-sizing-3bd55985af/reports/practice.md#pricing` ## Related concepts - Analytics — is a part of Analytics Node Sizing - Node — is a part of Analytics Node Sizing - Sizing — is a part of Analytics Node Sizing