Odoo performance

In Odoo, wall-clock is dominated by SQL query count, not Python. The ORM already batches writes, prefetches relations, and caches reads across a whole recordset — almost all slowness is fighting it: a search or browse inside a loop turns one query into N. Measure the real count first, then fix the hot path. Delegate measurement to the odoo-introspect skill — trace_flow (Layer D) reports total_sql and per-call sql_count:

odoo-ai --db <DB> trace <model> <record_id> <method>   # total_sql + sql_count per call

Version floor: Odoo 17/18, through Odoo 19 (current LTS). The _read_group tuple API and the flush_* / invalidate_* cache methods are v16/17+; pre-16 uses read_group (dicts) and flush() / invalidate_cache(). Note read_group is deprecated in 18.2 (use _read_group, or formatted_read_group for formatted public output), and the field aggregate attribute group_operator was renamed aggregator in 17.2 — see skills/odoo-introspect/references/version-matrix.md.

Use the built-in, don't hand-roll it