Hot reload

Reload your gateway config with SIGHUP -- no restart, no dropped workflows.

Change config without restarting

You’ve got workflows running. A model is mid-way through a deploy pipeline. You need to update a workflow definition, add a new connection, or tweak an executor. Restarting the process means killing those in-flight workflows.

Instead, send SIGHUP:

kill -HUP $(pgrep praxec)

The gateway re-reads your config file, rebuilds everything that changed, and swaps it in. In-flight workflows keep running without interruption.

What gets rebuilt

When SIGHUP arrives, the gateway reloads and rebuilds:

• Workflow definitions – new workflows appear, updated definitions take effect for new instances, removed definitions stop being discoverable.
• Executors – new executor configurations, updated connection strings, changed CLI commands.
• Connections – new MCP server connections are established, removed ones are torn down.
• Discovery index – the search index rebuilds to reflect the new set of capabilities, workflow descriptions, goals, and guidance text.

What survives

• In-flight workflows – any workflow instance that’s already started continues on its current definition. It doesn’t get yanked mid-step. The model can still call praxec.query({workflowId}) (get) and praxec.command({workflowId, transition}) (submit) on existing instances.
• The audit sink – audit events keep flowing without gaps. The reload itself emits a config.reloaded audit event so you have a record of when the config changed.

How the swap works

The gateway wraps all reloadable components behind RwLocks. When the new config is parsed and validated, all new components swap in atomically – definitions, executors, connections, and the discovery index all update together. There’s no window where half the config is old and half is new.

If the new config file has a parse error or validation failure, the swap doesn’t happen. The gateway keeps running with the previous config and logs the error. You fix the file and send SIGHUP again.

Platform note

SIGHUP is a Unix signal. It works on Linux and macOS. It doesn’t exist on Windows. If you’re running on Windows, you’ll need to restart the process to pick up config changes.

Zero-downtime restart

For changes that SIGHUP can’t handle – like upgrading the binary itself – you can do a zero-downtime restart:

1. Start a new praxec instance on a different port with the new binary and config.
2. Update your load balancer or reverse proxy to point to the new instance.
3. Call begin_drain on the old instance (or let it drain naturally). It stops accepting new praxec.command({definitionId}) (start) calls but lets in-flight workflows finish.
4. Once the old instance has no active workflows, shut it down.

This gives you binary upgrades without dropping a single workflow.