Atlas database and Search load-shedding eligibility boundaries
Parent: MongoDB Atlas · Published reference · snapshot 2026-10-27
↓ Facts as markdownall context files
Source-anchored research for Atlas database and Search load-shedding eligibility boundaries. 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 eligible database clusters, the mongod ingress limiter queues excess operations. Enabling Load Shedding bounds that queue and rejects newly arriving operations when it is full during sustained overload. Disabling the setting leaves the ingress queue unbounded; it does not disable all IWM protections. (medium confidence) (verified as of 2026-10-14) [source]
- On eligible M30+ replica sets running MongoDB 9.0+, ingress queueing and execution control prioritization are always on even when database Load Shedding is disabled. Atlas does not provide these two protections on M10/M20 or M30+ sharded clusters in the documented matrix. Other protections, including cache/memory tuning, connection rate limiting and write blocking, apply more broadly. (low confidence) (verified as of 2026-10-14) [source]
- The database Load Shedding default is Atlas Managed, whose effective state is Disabled on MongoDB 9.0. Operators must distinguish the configured override from effectiveAdaptiveSettings.LOAD_SHEDDING; an override takes effect only when cluster eligibility permits it. (medium confidence) (verified as of 2026-10-14) [source]
- Search Load Shedding does not increase or decrease the fixed query execution pool. It bounds the admission queue; only a full queue causes rejection. Atlas sets execution-pool and bounded-queue capacities internally and can change them at any time. Off leaves the queue unbounded. (low confidence) (verified as of 2026-10-27) [source]
- SERVER-133136 describes query-memory shedding eligibility leaking through DBDirectClient reads inside writes; it reports no active bug because current bucket-reopening reads track no memory. This is an internal query-memory mechanism, not evidence of Atlas ingress-shedding availability. (low confidence) (verified as of 2026-08-10) [source]
- Search queue pressure can raise latency before queries are rejected. The Search Process is Overloaded alert fires after the queue-threshold condition persists for five minutes. Remediation includes reducing concurrent queries, optimizing queries, and adding Search Nodes or increasing their vCPUs. (medium confidence) (verified as of 2026-10-27) [source]
Problems, failure modes and limitations
- A Search queue rejection returns error 489, SearchRequestRejectedDueToOverload, with SystemOverloadedError but without RetryableError. Neither mongod nor the drivers automatically retries that rejection; applications should back off with jitter, cap attempts, and reduce query concurrency. (low confidence) (verified as of 2026-10-27) [source]
Comparisons and alternatives
- The Atlas database Load Shedding resilience test requires an M30+ replica set on MongoDB 9.0+, no pending configuration changes, no other concurrent shedding test, and Project Cluster Manager access. It temporarily rejects a small random percentage of operations regardless of load, then restores the prior setting; this random rejection is test behavior rather than ordinary production activation. (low confidence) (verified as of 2026-10-14) [source]
Facts and statements
- Atlas database Load Shedding takes effect only on M30+ replica set clusters running MongoDB 9.0 or later. It is unavailable on M10/M20 and M30+ sharded clusters. The IWM section can appear on M10/M20, but selecting Enabled leaves the effective shedding state Disabled. (medium confidence) (verified as of 2026-10-14) [source]
- Database shedding at mongod and Search/Vector Search shedding at mongot have independent controls; changing one does not change the other. (medium confidence) (verified as of 2026-10-27) [source]
- Search Load Shedding protects the mongot admission control queue for both $search and $vectorSearch. It is available on MongoDB 8.0, 8.3, and 9.0 or later. Atlas Managed resolves to Off on 8.0 and 8.3 and On on 9.0 or later. (medium confidence) (verified as of 2026-10-27) [source]
- The Search shedding override is unavailable on Free and Flex clusters; Atlas always manages their setting. The database M30+ boundary must not be applied to Search. (low confidence) (verified as of 2026-10-27) [source]
- The driver specification requires both overload and retryability labels for overload retries; a SystemOverloadedError label alone does not establish safe automatic retry eligibility. (medium confidence) (verified as of 2026-10-27) [source]
- SERVER-131884 describes a query-memory eligibility gap for merge-on-shard pipelines, but its resolution is Gone away with no fix version. It does not demonstrate released Atlas database ingress shedding on sharded clusters. (low confidence) (verified as of 2026-07-21) [source]
- The documented defaults differ by engine on MongoDB 9.0: database Load Shedding under Atlas Managed is Disabled, whereas Search Load Shedding under Atlas Managed is On. The database M30+ replica-set eligibility matrix must not be treated as proof of Search tier or topology restrictions. (medium confidence) (verified as of 2026-10-27) [source]
Children
- No children recorded.