Denorm

Denormalized and aggregated tables for PostgreSQL.

Keywords: PostgreSQL, denormalization, aggregation, incremental view maintenance, materialized view

• Overview
• Install
• Usage
• Operations
• Performance
• Migration
• Limitations
• Developing

Overview

Dernom is similar to PostgreSQL's REFRESH MATERIALIZED VIEW, except that it updates materialized table incrementally.

How it works: Define the query, the tables, and their relationships in JSON. Denorm generates the SQL DDL statements that create the necessary functions and triggers. Apply the generated SQL statements to the database. Now, the materialized target is automatically kept in sync with the source tables.

Features

• Efficient incremental updates
• Arbitrarily complex SQL features and expressions
• Configurable consistency
• Deadlock-free

Install

Pip

pip3 install denorm

Usage

For CLI usage, see Usage.

Operations

Denorm has two operations:

Aggregate

Create a materalized aggregate of a single table.

See Aggregate.

Join

Create a materialized join of several tables.

See Join.

Performance

Materialized views exchange slower write performance for higher read performance.

While it's impossible to escape the fundamental trade-off, Denorm is as fast or faster than hand-written triggers. It uses statement-level transitions tables to make batch updates especially efficient.

In deferred mode, Denorm uses temp tables to defer updates until the end of the transaction. Using temp tables and ON DELETE COMMIT reduces I/O overhead and obviates the need for vacuuming. Since PostgreSQL does not support global temporary tables, the tables are created as necessary for each session. Thus the first saliant update in a session may have several millseconds of overhead as the trigger creates the temporary tables. Pool connections to reduce overhead, and vacuum reguarly to prevent system tables from bloating.

Migration

Denorm does not generate migration scripts.

Consider a tool like migra to help generate migration scripts.

Limitations

Denorm mangles names for generated objects, Long IDs and table names may run into the PostgreSQL limit of 63 characters for identifiers.

Developing

Install: make install

Generate JSONSchema: make schema

Test: make test

Generate documentation: make doc

Format: make format

Publish

1. Update denorm/version.py.
2. Create commit Version <version>.
3. Tag v<version>.
4. Push master branch and tag.
5. Publish to PyPI: make publish.