Skip to content

OplogConfig

relationalai.config.config_fields

Configure oplog (semantic-model-management) recording on deploy.

enabled is the opt-in toggle: when true, a successful rai models deploy also records the change to the model’s oplog (and pull / branch / merge become available). It defaults to false while the feature rolls out — set oplog.enabled: true in raiconfig to turn it on. backend selects where the oplog lives: snowflake (the real _INTERNAL_OPLOG table) or jsonl (a local file, reserved for tests/demos).

table_type picks the Snowflake table type the oplog is provisioned as. hybrid is the default because appends are markedly faster on it: a single-record append measures ~1.2s against a Hybrid Table versus ~2.9s against a standard one, and the gap is fixed per-transaction overhead, so it does not shrink for small changesets. Set standard on accounts where Hybrid Tables are unavailable (e.g. Tri-Secret Secure). Correctness is equivalent either way: the concurrent-append fork guard is the compare-and-swap on _INTERNAL_OPLOG_HEAD (RAI-51225), which does not rely on PRIMARY KEY enforcement. This only affects tables at creation; an already-provisioned oplog keeps the type it was created with.

Inheritance Hierarchy

OplogConfigpydantic.BaseModel

Referenced By

RelationalAI Documentation
├──  Build With RelationalAI
│   └──  Understand how PyRel works > Configure PyRel
│       └──  Configure a model
└──  Release Notes
    └──  Python API Release Notes
        └──  What’s New in Version 1.16.0
            └──  New Features and Enhancements