Alternative UE-uniquement à MongoDB Atlas.
MongoDB Atlas is the managed MongoDB offering from MongoDB Inc., a publicly-traded US corporation. Atlas runs on AWS, Azure or GCP — meaning the data sits on a US-jurisdictional hyperscaler, managed by a US-jurisdictional database vendor. Two layers, both US. For sovereignty, the answer is either self-managed MongoDB on EU infrastructure (which we operate for clients) or migration to a different document/JSON-capable database under EU jurisdiction (typically PostgreSQL with JSONB, which covers 90% of MongoDB use cases).
Une "région UE" n'est pas la souveraineté. Quatre questions tranchent.
La résidence des données indique où sont les bits. La souveraineté indique quel système juridique peut contraindre à l'accès. La réponse doit tenir sur les quatre points — sinon la stack n'est pas souveraine.
Où les données sont-elles physiquement stockées ?
Pas "dans le cloud" — quel datacenter, dans quel pays, sous quelle juridiction.
Qui d'autre est dans votre chemin de données ?
Chaque fournisseur qui touche les données : le CDN, le relais e-mail, le tracker d'erreurs, le pipeline analytics.
Quelles lois peuvent contraindre à la divulgation ?
Un fournisseur dont le siège est aux États-Unis relève du FISA 702 et du CLOUD Act — même lorsque les données se trouvent à Francfort.
Qui détient réellement les clés de chiffrement ?
Si le fournisseur cloud détient à la fois les données et les clés, il peut les lire — quel que soit le DPA.
Échoue sur la juridiction et la garde des clés.
Bits en UE, maison mère américaine, sous-traitants américains dans le chemin par défaut, clés gérées par le fournisseur.
Réussit sur les quatre.
Hébergé en UE sur une infrastructure au siège européen. Zéro sous-traitant américain dans le chemin par défaut. Clés détenues par le client ou par un KMS européen. Nommés dans votre DPA Article 28.
Pourquoi les équipes partent MongoDB Atlas
MongoDB Atlas exits we've scoped come from two angles: regulated workloads (healthcare, fintech) where the AWS-via-Atlas double-hop fails compliance, and cost reviews where Atlas's per-cluster pricing is genuinely high vs self-managed. The migration target depends on use case. For document-heavy workloads with complex aggregation, self-managed MongoDB on EU compute preserves the API surface. For workloads that are using MongoDB as a JSON store, migrating to PostgreSQL with JSONB is often simpler and cheaper long-term.
MongoDB Atlas services et leurs équivalents UE-uniquement
Une migration n'est pas "échanger une boîte contre une autre". La cartographie ci-dessous est ce que nous exécutons pour les clients qui quittent MongoDB Atlas pour des raisons Schrems II — pleine juridiction UE, pas de maison mère US dans le chemin des données.
| MongoDB Atlas service | Alternative UE-uniquement | Note d'ingénierie |
|---|---|---|
| Atlas clusters (M10+) | Self-managed MongoDB on EU compute (Hetzner, OVH dedicated), or PostgreSQL with JSONB on EU managed | For pure MongoDB compatibility, self-managed on EU bare metal is the production pattern. PostgreSQL JSONB is the cheaper alternative if you don't need MongoDB-specific features. |
| Atlas Search (Lucene-based) | Elasticsearch on EU compute (Bonsai EU, self-managed), Meilisearch self-hosted, OpenSearch | Atlas Search is Lucene under the hood; standard Elasticsearch / OpenSearch handles equivalent workloads. |
| Atlas Vector Search | Qdrant (DE-headquartered) self-hosted or cloud, pgvector on EU PostgreSQL, Weaviate | Qdrant is the strongest EU sovereign vector database — production-ready and explicitly EU-jurisdictional. |
| App Services (Realm) | Self-hosted backend on EU PaaS, Supabase self-hosted with Realtime | App Services is being deprecated in 2025-2026 anyway; the migration target is a custom backend on EU infrastructure. |
| Atlas Stream Processing | Apache Kafka + ksqlDB on EU compute, Apache Flink on EU K8s | For stream workloads, Kafka + Flink is the industry standard. |
| Atlas Triggers | PostgreSQL triggers + LISTEN/NOTIFY, application-layer event handling, MongoDB change streams self-managed | Self-managed MongoDB supports change streams natively; PostgreSQL has its own LISTEN/NOTIFY mechanism. |
| Atlas Online Archive | Application-managed cold storage to EU object storage, scheduled archival to compressed BSON in OVH Object Storage | Online Archive is essentially scheduled tiering; replicate the pattern with EU object storage as the cold tier. |
| Charts (BI) | Metabase self-hosted, Apache Superset, Grafana with PostgreSQL data source | Metabase is the most Atlas-Charts-like; runs anywhere. |
| Data API / GraphQL | Hasura self-hosted, PostgREST, custom API layer | For PostgreSQL backends, Hasura provides instant GraphQL APIs. |
| Atlas Backup (Continuous) | mongodump scheduled to EU object storage, point-in-time recovery via oplog backup | For self-managed MongoDB, oplog-based PITR is the production pattern; we operate this for clients. |
Comment nous migrons depuis MongoDB Atlas
Une migration typique de mid-market se déroule en trois phases. Les chiffres ci-dessous supposent une équipe d'ingénierie de 6 à 10 personnes et une stack applicative modérément complexe.
Use-case decision: Mongo or Postgres
Audit query patterns. If you use Mongo-specific features (aggregation pipelines, change streams, Atlas Search complex queries) → self-managed MongoDB on EU. If you treat Mongo as a JSON store → migrate to PostgreSQL JSONB. Output: target architecture decision.
Cluster provisioning + replication
Self-managed MongoDB cluster (3-node replica set, EU bare metal) provisioned. Initial sync via Atlas Live Migration tool. Monitoring and backup configured.
Application cutover
Connection string changed. Read-only validation period. Cutover during low-traffic window. Atlas cluster decommissioned after verification.
5-year TCO on Atlas → self-managed MongoDB on EU bare metal: typically 70-85% cheaper at production scale. A typical M30 Atlas cluster ($2k+/month) is replaced by a 3-node Hetzner dedicated MongoDB cluster (€300-500/month) with comparable or better performance. The trade-off is operational responsibility, which is what we take on as the managed partner.
Questions fréquemment posées
Atlas has EU regions on AWS Frankfurt — does that solve sovereignty?
No. Two layers of US jurisdiction: MongoDB Inc. (US-headquartered) and AWS (US-headquartered). The CLOUD Act applies to both. For Schrems II–strict workloads, both must be eliminated.
Should we move to PostgreSQL or self-managed MongoDB?
Depends on usage. If you use MongoDB-specific features (complex aggregation pipelines, $lookup, change streams across collections, Atlas Search), self-managed MongoDB on EU bare metal preserves the API. If MongoDB is mostly storing JSON documents with simple queries, PostgreSQL JSONB is cheaper, more capable for analytical queries, and simpler to operate.
How operational is self-managed MongoDB?
For a 3-node replica set with backups and monitoring, it requires real ops work — typically 2-4 hours per week of attention plus incident response. We operate this for clients as part of the managed-infrastructure relationship; it's a known pattern.
What about MongoDB's own self-hosted enterprise version?
MongoDB Enterprise Advanced runs on your infrastructure but the licence is from MongoDB Inc. — a US contracting party. For pure sovereignty, MongoDB Community Edition (open-source, AGPL) is the option. For most production workloads, Community is sufficient.
Atlas Vector Search is critical for our AI features. What's the EU equivalent?
Qdrant (DE-headquartered) is the strongest sovereign vector database. Qdrant Cloud has EU regions; Qdrant self-hosted is the same engine. Migration from Atlas Vector Search to Qdrant is straightforward (vectors are vectors), with the schema mapping being the only meaningful work.
How long does an Atlas exit take?
For a single replica-set workload with moderate data (50-200GB): 2–4 weeks. For sharded clusters or very large data: 6–12 weeks. The schedule is dominated by the live-migration phase, not the application-side changes.
Planifiez votre sortie de MongoDB Atlas.
Appel de cadrage de 30 minutes. Nous cartographions votre stack par rapport aux alternatives UE-uniquement, estimons l'effort de migration et vous disons si c'est le bon choix.