This page lists the known limitations of MongoDB Search and Vector Search when you deploy mongot with the MongoDB Controllers for Kubernetes Operator. Review this page and the release notes to learn about any changes to these limitations in future releases.
Deployment and Topology
Sharded cluster topologies that the MongoDB Controllers for Kubernetes Operator manages are supported for MongoDB Enterprise Advanced only. To learn which topologies the MongoDB Controllers for Kubernetes Operator supports, see Supported Topologies.
A
mongotinstance is bound to a single replica set, or to a single shard in a sharded cluster. You cannot share onemongotinstance across multiplemongoddeployments. Deploy a separatemongotinstance for each replica set or shard.A self-managed load balancer only supports single-cluster deployments. Multi-cluster deployments require the managed Envoy load balancer.
mongotdoes not support cross-cluster query failover. Each cluster's search deployment is independent. If a cluster becomes unavailable, search is unavailable in that cluster, and queries do not fail over to another cluster.When you delete a multi-cluster
MongoDBSearchresource, Kubernetes garbage collection removes only the resources in the central cluster. The MongoDB Controllers for Kubernetes Operator does not delete themongotresources that it created in the other member clusters. Delete those resources manually.If you remove a cluster from
spec.clusters, the MongoDB Controllers for Kubernetes Operator does not remove all of that cluster's resources.For sharded cluster sources, the MongoDB Controllers for Kubernetes Operator removes the stale
mongotStatefulSets, Services, and ConfigMaps, but does not remove the TLS certificate Secrets or the Envoy proxy resources.For replica set sources, the MongoDB Controllers for Kubernetes Operator does not remove any of the cluster's resources. Delete the remaining resources manually.
Security
mongotreads certificates at startup and does not reload them while running. If you rotate the TLS certificate in the Secret thatspec.security.tls.certsSecretPrefixreferences, the MongoDB Controllers for Kubernetes Operator detects the change and restartsmongotautomatically in a rolling fashion, onemongotpod at a time in eachmongotgroup.mongotreads credentials at startup. After you rotate the SCRAM password for thesearch-sync-sourceuser, you must restartmongotso that it reads the new credentials. The MongoDB Controllers for Kubernetes Operator does not restartmongotautomatically for this change.mongotdoes not validate OCSP or CRL certificate revocation. To reject a revoked certificate, remove the corresponding$externaluser onmongod.mongotdoes not provide application-level encryption at rest. To encrypt themongotindex data, encrypt the underlying volume with a Kubernetes-level or cloud-provider mechanism. To learn more, see Secure the Connection from MongoDB to Search.
Configuration and Maintenance
mongotreads its configuration at startup and does not support live configuration reload. If you change theMongoDBSearchresource, the MongoDB Controllers for Kubernetes Operator restartsmongotautomatically in a rolling fashion, onemongotpod at a time in eachmongotgroup, to apply the change.mongotupgrades require a process restart. No in-place binary swap is supported. To learn more, see Upgrade or Downgrade.Index storage is not portable across
mongotmajor versions. Amongotmajor-version downgrade requires a re-sync, which rebuilds the index from the source data. Plan major-version upgrades carefully and back upmongodso that the index can be rebuilt if you need to downgrade.
Observability
- The Metrics UI in Atlas is not available for
mongotdeployments that you manage yourself. To set up monitoring, use the metrics endpoint thatmongotexposes, or forwardmongotmetrics to Ops Manager with the metrics forwarder. The metrics forwarder requires Ops Manager 8.0.25 or later and does not support Cloud Manager. To learn more, see Monitor Your Deployment.
Search and Vector Search Features
Automated embedding is a Preview feature and integrates only with Voyage AI models. To use another embedding provider, generate embeddings in your application and store them in a vector field.
Automated embedding is only available for single-cluster topologies.
Features that are guarded by Atlas-only flags are not available on self-managed
mongot. A feature is available on self-managed deployments only after it ships in amongotbinary release.