Incremental Rollout
Gradually shift traffic between backends using percentage-based load balancing — no application changes required.
Architectural Value
ProxySQL's weight-based host group routing enables precise, percentage-controlled traffic distribution across backends, making progressive delivery and canary deployments a configuration concern rather than an application one.
Key Capabilities
- → Percentage-Based Traffic Split: Assign weights to backend nodes within a host group to control the exact proportion of queries each receives.
- → Runtime Adjustable: Shift traffic percentages live via ProxySQL admin without restarting, redeploying, or touching application configuration.
- → Per-Query-Rule Granularity: Combine with query rules to apply incremental rollout to specific query types, users, or schemas independently.
Architectural Review
Not sure if this fits your current stack? Our experts can help.
Schedule a Call →Incremental Rollout: Architectural Deep-Dive
The Problem
Deploying a new database version, schema migration, or backend configuration change carries risk. The conventional options are binary: cut over entirely and absorb the blast radius if something is wrong, or maintain a parallel environment and switch via DNS or application reconfiguration — both of which are slow, operationally expensive, and difficult to reverse quickly.
Canary deployments and blue/green strategies are well-understood at the application tier, but the database layer has traditionally been excluded from this model. The database is treated as a monolithic dependency that must be changed all at once, because there is no facility between the application and the database to split traffic by percentage.
The ProxySQL Approach
ProxySQL controls traffic distribution across backend nodes through a weight system applied per host group. When multiple backends are members of the same host group, incoming queries are distributed across them in proportion to their assigned weights. Setting one node to a weight of 95 and another to 5 routes approximately 5% of traffic to the new backend, with no change to application code, no DNS update, and no redeployment.
The weights are adjustable at runtime through the ProxySQL admin interface. A rollout can begin at 1%, be observed, increased to 5%, 10%, 25%, and ultimately 100% — each step taking effect immediately, each step fully reversible by adjusting weights back. If the new backend exhibits problems at any percentage, traffic is shifted back in seconds.
Combined with ProxySQL’s query rule engine, rollout granularity becomes precise. You can direct 10% of read traffic to a new replica, while all write traffic remains on the established primary. Or route a specific user group’s queries to the new backend at full percentage while everyone else stays on the current configuration.
What You Can Do With It
Canary Database Releases allow a new MySQL or PostgreSQL version, or a modified configuration, to receive a controlled slice of production traffic before a full cutover. Real query patterns hit the new backend at low volume, surfacing compatibility issues, performance regressions, or operational problems before they affect your entire user base.
Schema Migration Validation routes a percentage of queries to a backend running the post-migration schema, confirming that application queries remain compatible and that performance characteristics are acceptable under real load — without requiring a maintenance window or full cutover.
Hardware or Cloud Instance Testing distributes traffic across instance types to measure real-world query latency, connection behavior, and resource utilisation differences under production workload, providing data for infrastructure decisions that synthetic benchmarks cannot replicate.
The Result
Incremental rollout through ProxySQL makes progressive delivery a first-class capability of your database infrastructure. Traffic shifts are instantaneous, reversible, and require no coordination with application teams. The risk profile of database changes drops from binary all-or-nothing to a graduated ramp that can be paused or reversed at any point. Changes that previously required maintenance windows can be validated continuously under production load at whatever percentage the risk tolerance allows.