Datastores

Scope

Applies to every DynamoAI product. Tier names below are Azure; see Cloud Reference for equivalents.

Configuring datastores

Point the platform at managed services you operate. The bundled PostgreSQL, MongoDB and Redis run as single instances and are intended for evaluation.

PostgreSQL

• Use a General Purpose or Memory Optimized tier. Burstable tiers support neither high availability nor read replicas.
• Enable zone-redundant high availability.
• Create the dfl and keycloak databases before installing. The migration job creates schema inside them but does not create the databases.
• Include PG_PARTMAN, PG_STAT_STATEMENTS, and PGVECTOR in the allowed extensions. These are not enabled by default and the migration job requires PG_PARTMAN.

important

Geo-redundant backup can only be enabled when the server is created. If cross-region recovery is in scope, enable it at provisioning time. Adding it later requires rebuilding the server.

MongoDB

• Use a dedicated cluster tier. Free and Flex clusters offer no point-in-time recovery, no multi-region option, and no private connectivity; Flex keeps daily snapshots only.
• Deploy a three-node replica set with writes acknowledged by a majority. A write is not acknowledged until a majority holds it, so failover loses no acknowledged data.
• Enable point-in-time recovery. MongoDB is the system of record for all policy and guardrail configuration, and snapshot-interval recovery is the difference between losing minutes and losing hours of configuration changes.

Redis

• Enable high availability, then enable persistence. Persistence requires high availability to be enabled first. These rules apply to Redis Enterprise based managed services such as Azure Managed Redis; older cache tiers expose different options.
• Note that high availability can be enabled on a running instance but cannot be disabled again. Decide before provisioning.
• Use TLS with authentication and private connectivity. TLS to Redis is supported from release 3.26.4; see Enable Redis Authentication & TLS.

Redis carries the task queue, job progress, and the stage-completion map that evaluation restarts consult. A Redis failover discards queued and in-flight background work; an operator restarts the affected jobs, which re-run every stage from their checkpoints. Daily request rate-limit counters also reset.