PartyKit

Hosted stateful realtime servers, one instance per room, reached over WebSocket.

Provider
default
Category
Edge
Properties · 5 set by this component
propertytypevaluesrequiredkind defaultthis component
ackIntervalSeconds · advancedinteger—no—inherited
autoscalingboolean—nofalse▸ true
channelAssignment · advancedenumrandom | hash | stickynorandom▸ sticky
deployment · component propertyenumpartykit-cloud | self-hosted-cloudflarenopartykit-cloudinherited
drainOnScaleIn · advancedboolean—nofalseinherited
hibernation · component propertyboolean—nofalseinherited
httpEndpoints · component propertyboolean—notrueinherited
managedboolean—nofalse▸ true
protocolenumwebsocket | sse | long-poll | grpc-stream | udp-multicastnowebsocket▸ websocket
pushFallback · advancedboolean—nofalseinherited
roomStorage · component propertyboolean—notrueinherited
sequenceNumbers · advancedboolean—nofalseinherited
stampedeGuard · advancedboolean—nofalseinherited
statefulboolean—notrue▸ true
subscriptionsPerUser · advancedinteger—no—inherited
Rules that apply · 3

Every rule whose selector covers this component's kind or its category. Whether a rule fires depends on the design: its properties, its edges and its groups.

reconnect-stampede-without-guard

When a gateway holding millions of sockets restarts, every client reconnects at once and the gateway asks the coordination service about each of them. Without a stampede guard that lookup storm takes the coordination service down too.

  • — all of these hold:
  • — the node has at least one synchronous outgoing edge to a coordination-service
  • — the node's `stampedeGuard` is false
shared-host-per-channel-without-isolation

With heavy-tail traffic a few channels carry most of the load. Assigning channels at random lands those hot channels on arbitrary hosts, and every tenant sharing a host suffers for it.

  • — all of these hold:
  • — the node's `channelAssignment` is "random"
  • — the design requirement `traffic.skew` is "heavy-tail"
stateful-websocket-scale-in-without-drain

Scaling a gateway down kills every live WebSocket on the removed instance at once. Without a drain step those clients all reconnect together.

  • — all of these hold:
  • — the node's `autoscaling` is true
  • — the node's `drainOnScaleIn` is false
Use it in a design
nodes:
  - { id: "partykit", component: "partykit" }
Other Real-time gateway products · 19
Source
# yaml-language-server: $schema=../../schema/component.schema.json
# PartyKit: hosted stateful realtime servers ("parties"). The documentation describes one server
# instance per room reached over WebSocket, persistent per-room storage, HTTP request handling and an
# opt-in hibernation mode; deployment is to PartyKit's platform or to your own Cloudflare account.
schemaVersion: "1"
type: "component"
id: "partykit"
kind: "realtime-gateway"
provider: "default"
docs: "https://docs.partykit.io/"
defaults:
  managed: true
  protocol: "websocket"
  stateful: true
  autoscaling: true
  channelAssignment: "sticky"
properties:
  deployment:
    type: "enum"
    values: ["partykit-cloud", "self-hosted-cloudflare"]
    default: "partykit-cloud"
  hibernation:
    type: "boolean"
    default: false
  roomStorage:
    type: "boolean"
    default: true
  httpEndpoints:
    type: "boolean"
    default: true

The same page as markdown, for an agent: /catalogue/components/partykit.md