Skip to main content

Optimizer

The optimizer rewrites the parsed plan to make it run faster. The most important rewrite is slice push-down.

Slice push-down​

Naive evaluation of mydb.users[country = "IN"][0:20] would:

  1. Fetch every Indian user.
  2. Filter in memory.
  3. Slice the first 20.

The optimizer detects this pattern and pushes the slice into the storage layer:

  1. Fetch only the first 20 Indian users from the index.

For a table with a million rows, this is a thousand-x speedup.

Patterns the optimizer recognises​

PatternPush-down
Table[filter][slice]Filter + slice → indexed range read
Table._asc(col)[slice]Sort + slice → indexed sort scan
Table[filter]._asc(col)[slice]All three together

What blocks push-down​

  • Filters that include _like, _asc, _desc, or other aggregates
  • Filters that walk into a related table
  • Comparison operators other than =, ==, and, or
  • Filters with row access (e.g. category[0].name)

When push-down is blocked, the slice is applied in memory after the full filtered set is loaded. Correct, just slower.

Disabling the optimizer​

For debugging:

ONQL_OPTIMIZER=off ./onql

The optimizer source lives in dsl/optimizer/.