skip to content

MongoDB Atlas

MongoDB's managed service and what it layers on top: cluster tiers, automated operations, Atlas Search and Vector Search. Operational answers often assume it, because many deployments run on it.

on this pageshow

explore

questions

17

What are MongoDB Atlas Free and Flex clusters, and what do they give up compared with a dedicated M10+ cluster?

level: juniorimportance: must knowfreq 65%

answer

  1. entry shapes, not production shapes
  2. a former name for one of them
  3. two storage caps, one connection cap
  4. private networking and workload isolation
  5. sharding has its own tier floor

basics

~20 s

Atlas Free and Flex clusters are entry shapes for learning and small workloads. Against dedicated M10+ clusters they give up private networking, configurable and auto-expanding storage, dedicated analytics and search nodes, customer-managed encryption and, from M30, sharding.

solid answer

~40 s

A Free cluster, formerly called `M0`, is a sandbox replica set, normally one per project, capped at 0.5 GB of data and 100 operations per second. A Flex cluster is the low-cost step up for development and small applications, at 5 GB and 500 operations per second. Both stop at 500 connections and 50 aggregation stages, and neither lets you configure memory or storage. Compared with dedicated `M10+` clusters they give up private endpoints, network peering, auditing, customer-managed encryption keys, auto-expanding storage, dedicated analytics and search nodes, and sharding, which on dedicated starts at `M30`. A team leaves when a security review or data growth hits one of those walls. `M2` and `M5` are retired: Atlas migrated them all to Flex.

go deeper

for a junior

Recall the three shapes by their current names: Free, Flex and dedicated. Have the headline numbers ready, 0.5 GB and 5 GB of data and 500 connections, and name private endpoints and sharding as things neither entry shape offers.

for a middle

Explain why the limits matter in practice: no temporary disk files changes what a big aggregation can do, and a fixed storage size means growth ends in a move. Separate capability gaps from numeric ceilings when you list them.

for a senior

Show that you read the limits page before committing a workload, and that you plan the move to dedicated before a security review or data growth forces it. Know that M10 and M20 still cannot shard, so the target tier matters.

for a principal

Treat the entry shapes as a policy question: which environments may run on them, and what event triggers a move to dedicated. Cost is one input, while audit, network and restore requirements often decide it.

## What the three shapes are **MongoDB Atlas** is MongoDB's managed cloud service. It runs your database as a **cluster**, and the docs name three shapes a team can pick today: - **Free cluster:** a free sandbox replica set, formerly known as `M0`. A project gets one, unless a sales contract with MongoDB allows more. It runs three nodes and you cannot change that. - **Flex cluster:** a low-cost type the docs aim at teams learning MongoDB or building small proof-of-concept applications, and at infrequent or sparse workloads. - **Dedicated cluster:** a tier you choose, starting at `M10`, for production work. `M10` and `M20` suit development and low-traffic production; the docs recommend `M30` and higher for production. Older material also shows `M2` and `M5` as rungs between the free shape and dedicated. Those are retired: Atlas stopped supporting them on January 22, 2026 and migrated every remaining one to a Flex cluster. ## The numbers worth remembering Each figure belongs to one shape's limits page. | Limit | Free | Flex | Dedicated | |---|---|---|---| | Data storage | 0.5 GB | 5 GB | Configurable from `M10` | | Connections | 500 | 500 | Per node by tier: 1500 on an AWS General-class `M10` | | Operations per second | 100 | 500 | None listed on the limits pages | | Snapshots | Daily once enabled, up to four kept | One daily, always on | Configurable | Data storage counts the uncompressed BSON documents in all collections plus their indexes, so compression does not stretch the cap. ## What both entry shapes give up The two limits pages agree on most of the list: - **Private networking:** no private endpoints and no network peering connections. - **Security controls:** no database auditing, and no encryption at rest using your own key management. - **Horizontal scale:** neither can be deployed as a sharded cluster. Dedicated does not hand you this at the entry tier either: `M10` and `M20` are replica set only, and `M30+` adds sharded deployments. - **Workload isolation:** no dedicated analytics or search nodes, which the docs allow only on `M10+`. - **Sizing:** you cannot configure memory or storage size, and storage does not auto-expand. - **Diagnostics:** no Performance Advisor, no Real-Time Performance Panel, no log downloads, and the Metrics view shows only connections, logical size, network and opscounter. - **Query work:** no temporary files on disk, so `allowDiskUse` does not help, in-memory sorts stop at 32 MB, and server-side JavaScript such as `$where` is unsupported. - **Failure testing:** no primary failover tests and no regional outage simulation. The Flex page adds one pointed recommendation. Flex has no continuous backup or point-in-time restore (restoring to any chosen moment), and a workload that needs them, or needs private endpoints, should switch to a dedicated cluster. ## What Flex adds over Free The docs list three things: backups, increased storage and API access. Read the backup line with care. The Free limits page says you can enable backups on a Free cluster, with a daily snapshot and up to four retained, so "a Free cluster has no backups" is wrong. A Flex cluster takes a single daily snapshot at a fixed time, starting 24 hours after creation. It has no custom policies and no on-demand snapshots, and its snapshots restore only to replica sets. The Free page also lists two limits with no matching row on the Flex page: 10 GB in and 10 GB out per rolling seven days, and an automatic pause after 30 days with zero connections. ## What happens at a limit, and how a team leaves Going over the operations limit shows up as delay. On a Free cluster Atlas throttles network speed and applies a one-second cooldown before resuming operations on a connection, and queued operations can wait longer than a second. The Flex page describes a sub-second cooldown and the same queueing. Leaving is a short sequence: 1. Name the wall: a capability gap such as private endpoints, or a ceiling such as 5 GB. 2. Move up: a Free cluster can scale to Flex or to dedicated at any time, and a Flex cluster can upgrade to a dedicated tier later. 3. Pick the tier by need: `M10` for configurable storage and isolated nodes, `M30+` for sharding and the docs' production recommendation. MongoDB says it plans to add more configurations and capabilities to Flex over time, so treat this list as what the docs state at the time of writing, not as a permanent fact. These comparisons also describe Atlas Core; Atlas Infinite, in public preview, has its own gaps, including no sharded clusters.

  • What happens when an Atlas Free cluster goes over its limit of 100 operations per second?
    Atlas throttles the cluster's network speed and applies a one-second cooldown before resuming operations on a given connection. If the queue exceeds the limit, operations can wait more than a second. Once the rate drops below the limit, Atlas works through queued operations on each connection before taking new ones.
  • How much control does an Atlas Flex cluster give you over its snapshots?
    Very little. Atlas always takes one daily snapshot at the same time, starting 24 hours after the cluster was created. Custom policies and on-demand snapshots are not supported, and you can restore Flex snapshots only to replica sets, not to a sharded cluster. The docs point teams wanting finer-grained backups to `M10` or larger.
  • Which upgrade paths do the Atlas docs describe out of the Free and Flex shapes?
    A Free cluster can scale to a Flex cluster or to a dedicated cluster at any time. A Flex cluster can start a project and upgrade to a production-ready dedicated tier later. You pick the Atlas edition when you create either shape, and it applies only once you scale to dedicated, where you can still choose differently.

  • Atlas still offers M2 and M5 as the paid step between Free and dedicated.
  • A Free cluster is a single node because it costs nothing.
  • A Flex cluster can be sharded once its data passes 5 GB.
  • Private endpoints work on Flex clusters and are missing only on Free.
  • Every dedicated cluster from M10 upward can be deployed as a sharded cluster.
  • Setting allowDiskUse rescues a large sort on a Free or Flex cluster.

context

open as a page

In MongoDB Atlas, how do database users differ from the users who sign in to the Atlas console?

level: juniorimportance: must knowfreq 66%

basics

~20 s

Atlas console accounts are organization and project members who manage the deployment through the UI or Admin API. Database users are separate credentials, created per project, that authenticate to the cluster itself and carry MongoDB roles such as readWrite on named databases.

open as a page

In Atlas Backup, how do scheduled cloud snapshots differ from continuous cloud backup?

level: middleimportance: must knowfreq 58%

basics

~20 s

Scheduled cloud snapshots are cloud-provider volume snapshots taken on a policy you define, so you can restore only to a snapshot instant. Continuous cloud backup additionally captures oplog entries, letting you restore to any timestamp inside a configured restore window.

open as a page

Why must $search be the first stage of an Atlas aggregation pipeline?

level: middleimportance: must knowfreq 70%

basics

~20 s

$search runs on mongot, a separate Lucene process, and produces the pipeline's result stream from its own index instead of filtering documents handed to it. Being the source of the stream, it cannot follow another stage.

open as a page

In $vectorSearch, what does numCandidates control and how do you choose it?

level: middleimportance: must knowfreq 65%

basics

~20 s

numCandidates is how many near neighbours the approximate search examines before the best limit are returned. Raising it improves recall and costs latency; it must be at least limit, and a common starting point is ten to twenty times limit.

open as a page

What happens to a running application while an Atlas cluster tier change is applied?

level: seniorimportance: must knowfreq 46%

basics

~20 s

Atlas applies a tier change as a rolling replacement: secondaries are resized and resynchronised one at a time, then the primary steps down so a resized member takes over. Expect a brief election with no primary, plus higher latency afterwards while caches refill.

open as a page

How does Atlas cluster tier and storage auto-scaling decide when to resize a cluster?

level: middleimportance: should knowfreq 52%

basics

~20 s

Atlas watches sustained CPU and memory utilization and moves the cluster one tier at a time within the minimum and maximum you configure. Scale-down is optional and far more conservative. Storage auto-scaling grows the disk when usage approaches about 90 percent, and only ever grows it.

open as a page

When is an Atlas Flex cluster the wrong choice against a dedicated cluster, and what happened to the retired M2, M5 and Serverless shapes?

level: middleimportance: should knowfreq 35%

basics

~20 s

An Atlas Flex cluster is the wrong choice once a workload needs private networking, point-in-time restore, sharding, isolated nodes or configurable scaling. M2 and M5 were migrated to Flex; Serverless instances went to Free, Flex or dedicated by usage.

open as a page

What does the Atlas IP access list control, and why does a blocked client see a timeout?

level: middleimportance: should knowfreq 64%

basics

~20 s

The Atlas IP access list is a per-project allowlist of public source addresses, CIDR blocks, or AWS security groups permitted to open a connection. Traffic from anything else is dropped before authentication, so the driver reports a server-selection timeout rather than an authentication error.

open as a page

How do you return Atlas Search facet counts alongside the matching documents?

level: middleimportance: should knowfreq 38%

basics

~20 s

Use the facet collector inside $search and read the buckets from the $$SEARCH_META variable in a later stage, or run $searchMeta when you want the counts alone. Facet paths must be mapped as stringFacet, numberFacet or dateFacet.

open as a page

How do you restrict a $vectorSearch to documents matching a metadata condition?

level: middleimportance: should knowfreq 55%

basics

~20 s

Pass a filter document to the $vectorSearch stage, and declare every path it touches as a filter field in the vector index definition. The condition is then applied during the search, so all limit results already satisfy it.

open as a page

In an Atlas Search index, when do you replace dynamic mapping with static field mappings?

level: middleimportance: should knowfreq 45%

basics

~20 s

Dynamic mapping indexes every supported field with default settings, which is ideal for prototyping. Switch to static mappings once you need a chosen analyzer, autocomplete, facet or filter field types, or want to stop indexing fields nobody searches.

open as a page

Latency on a production Atlas cluster spiked after a release — which Atlas tools do you open, and in what order?

level: seniorimportance: should knowfreq 52%

basics

~20 s

Start with the Real-Time Performance Panel to see what is running now, use Metrics for the shape and timing of the regression, then the Query Profiler for the slow operations in the last day, and Performance Advisor for ranked index suggestions — validating any suggestion with explain before shipping it.

open as a page

When would you choose an Atlas private endpoint over VPC peering, and what changes for clients?

level: seniorimportance: should knowfreq 42%

basics

~20 s

Choose a private endpoint when you want one-way, non-routable access with no CIDR coordination: the endpoint lives in your VPC and Atlas cannot initiate back. Peering joins two networks bidirectionally, needs non-overlapping CIDRs, and still requires access-list entries. Clients use an endpoint-specific connection string.

open as a page

How do you combine $search and $vectorSearch results in a single MongoDB pipeline?

level: seniorimportance: should knowfreq 45%

basics

~20 s

Run one search as the outer pipeline's first stage and the other inside a $unionWith sub-pipeline, assign each result its rank within its own list, then merge on _id and combine the ranks. Recent server versions add a dedicated rank-fusion stage.

open as a page

How would you distribute Atlas cluster nodes across regions to survive a full region outage?

level: principalimportance: should knowfreq 40%

basics

~20 s

Place electable nodes in at least three regions so a surviving majority can still elect a primary; two regions cannot survive losing the larger half. Use region priority to steer the primary, and add read-only or analytics nodes elsewhere without giving them votes.

open as a page

How does Atlas encrypt data at rest by default, and what changes with your own KMS key?

level: seniorimportance: nice to knowfreq 30%

basics

~20 s

Atlas encrypts all cluster storage at rest by default using cloud-provider disk encryption, with no configuration required. Customer key management on dedicated tiers instead wraps the storage key with a key you hold in AWS KMS, Azure Key Vault, or Google Cloud KMS, giving you custody and a revocation switch.

open as a page