KrishivDocs
Experimental

Maintain an incremental view

Run the experimental IVM CLI over change files and materialize a net result.

The ivm command drives an incremental view from one or more change sources. It can run embedded or against a configured coordinator.

Run a local view

krishiv ivm run \
  --job-id sales-by-key \
  --sql "SELECT k, SUM(v) AS total FROM changes GROUP BY k" \
  --source changes=./changes.csv \
  --source-format csv \
  --sink ./sales-by-key.ndjson \
  --sink-format json

This ordinary CSV input is treated as an insert delta. CDC-aware readers can also carry update and delete row kinds; do not infer those semantics from a CSV file alone. The output file contains the net materialized result for the view.

Remote coordinator

krishiv -c http://127.0.0.1:2002 ivm run \
  --job-id sales-by-key \
  --sql "SELECT k, SUM(v) AS total FROM changes GROUP BY k" \
  --source changes=./changes.csv \
  --sink ./sales-by-key.ndjson

Remote IVM paths exist, but distributed executor behavior is not a stable product contract. Keep the evaluation local unless you are testing that boundary itself.

Expect visible fallback

Some plans can propagate deltas operator by operator. Other plans fall back to recomputation. Treat fallback as a correctness path, then measure whether it still meets the workload goal.

Experimental API

Pin the Engine commit and retain input data that can rebuild the view. Operator support, checkpoint metadata, and public APIs may change before a stable release.

On this page