Router¶

A typical deployment has one shared queue whose messages carry only the tenant and an event. The message does not name a script; the tenant's router script decides.

Tenant config: router¶

Name of the script whose on_route handler chooses a script for messages that don't name one.

{ "router": "router" }

Flow¶

queue message {tenant, event}
        │
        ▼
 <tenant>/config.json  →  "router": "router"
        │
        ▼
 router.on_route(ctx, event)  →  "parse-order"   (or { script, version?, event? })
        │            same VM, same deadline and call budgets
        ▼
 parse-order.on_queue(ctx, event)

Router script¶

--!strict
--!runtime 1
local function on_route(ctx: Context, event: { key: string? })
  local key = event.key
  if key and string.sub(key, 1, 16) == "incoming/orders/" then
    return "parse-order"
  end
  return { script = "fallback", event = { original = event } }
end
return { on_route = on_route }

• The router's handler is on_route (or route) and it runs with ctx.trigger == "route".
• Return a script name, or a table { script, version?, event? } to pin a version or rewrite the event the target receives.
• The target then runs with the original trigger (queue, http, …).
• The router and the target run in one Luau VM: one native-API install, one heap, one deadline, one set of call budgets. If the target pins a different --!runtime version than the router, it gets a fresh VM with the remaining time instead.
• A router cannot route to another router. With no router configured, a message that names no script fails with a no route error (queue: retried, then dead-lettered).

What gets routed¶

• Any envelope without a script: queue messages, POST /v1/{tenant} over HTTP, direct invokes.
• S3 notifications, which never name a script.
• Callbacks created without script.
• A batch job's done.json, which the router hands to native.batch.route.