Figma Multiplayer Editing, the whole design.

%% title: Multiplayer on one page: one owner per file
flowchart TD
  ED@{ icon: "sd:browser", label: "Browser + desktop editors<br/>local copy of the file · outbox" } ==>|"changes, cursors"| GW@{ icon: "tech:websocket", label: "WebSocket gateways · ~60<br/>TLS · session · route by file" }
  ED --> API@{ icon: "sd:cdn", label: "HTTP API + CDN<br/>images, fonts by content hash" }
  API --> META@{ icon: "tech:postgresql", label: "LiveGraph over Postgres<br/>comments · file browser" }
  GW -.->|"who owns file F?"| RT@{ icon: "sd:key", label: "Router + lock table<br/>file → owner, lock UUID<br/>role per file · revocations" }
  GW ==> MP["Multiplayer server · one process per file<br/>whole file in memory<br/>checks + sequences every change"]:::hot
  RT -.->|"revoke edit"| MP
  MP --> PR@{ icon: "sd:cache", label: "Presence<br/>latest cursor per session · memory only" }
  MP ==>|"~0.5 s batches, if lock matches"| JR@{ icon: "tech:dynamodb", label: "Journal · DynamoDB<br/>per-file sequence numbers" }
  MP --> CK@{ icon: "tech:s3", label: "Checkpoints · S3<br/>compressed file · every 10 min" }
  CK --> VH["Version history<br/>autosave every 30 min · named · branches"]:::service
  class ED client
  class GW,API edge
  class META,JR,CK store
  class RT store
  class PR cache

%% title: Inside the owner: one change, checked then sequenced
flowchart TD
  IN(["Change · client 17, change_id 41<br/>button.x = 112"]):::client --> DUP{"change_id already applied<br/>for this client?"}:::service
  DUP -->|"yes"| DROP(["Drop · ack again<br/>a resend after reconnect"]):::ok
  DUP -->|"no"| CHK{"Can edit? Object exists?<br/>Parent change makes a cycle?"}:::service
  CHK -->|"fails"| REJ(["Reject with server values<br/>client rolls back"]):::bad
  CHK -->|"ok"| LWW["Write the property · last writer wins<br/>free pos if taken · seq + 1"]:::hot
  LWW --> OUT["Ack to author · echo to all<br/>one frame per viewer per tick"]:::core
  LWW --> BUF[["Unsaved buffer"]]:::queue
  BUF -->|"every 0.5 s"| JW@{ icon: "tech:dynamodb", label: "Journal write<br/>conditional on lock UUID" }
  JW --> SAVED(["Broadcast saved up to seq<br/>clients clear their outboxes"]):::ok
  class JW store

%% title: One nudge, Priya to Arjun and into the journal
sequenceDiagram
  autonumber
  participant P as Priya's editor
  participant S as Owner server
  participant J as Journal
  participant A as Arjun's editor
  P->>P: apply x = 112 now · mark unacked
  P->>S: change c41 · button.x = 112
  S->>S: may edit? exists? · seq 9,012
  S-->>P: ack c41 = seq 9,012
  S-)A: seq 9,012 · button.x = 112
  A->>A: apply · redraw next frame
  Note over P: incoming values for x ignored while unacked
  Note over S,J: every ~0.5 s
  S->>J: batch 9,001–9,040 · if lock matches
  J-->>S: written
  S-)P: saved up to 9,040
  P->>P: drop c41 from the outbox

%% title: Owner crashes with Priya's changes unsaved
sequenceDiagram
  autonumber
  participant P as Priya's editor
  participant R as Router + lock table
  participant N as New owner
  participant J as Journal + S3
  participant O as Old owner (stalled)
  Note over O: journal has up to 9,412 · 9,413–9,414 echoed, not saved
  R->>N: own file F
  N->>R: write fresh lock UUID
  N->>J: load checkpoint, replay batches after it
  Note over N: recovered to 9,412
  P->>N: hello · old epoch, last seq 9,414
  N-->>P: ahead of recovery · resync
  P->>N: resend outbox · her two unsaved changes
  N-->>P: sequenced again · dupes dropped by change_id
  alt old owner wakes up
    O->>J: write batch
    J-->>O: lock mismatch · refused
    O->>O: stop serving F
  end

%% title: Draining a host: warmed handoff of one busy file
sequenceDiagram
  autonumber
  participant C as Editors in the file
  participant O as Old owner
  participant R as Router
  participant N as New owner
  participant J as Journal
  R->>N: prepare file F
  N->>J: load checkpoint, follow the journal
  Note over N: decode done before cutover
  R->>O: hand F to N
  O->>O: stop sequencing · edits wait in outboxes
  O->>J: flush last batch up to 9,412 · under 1 s at p99
  R->>R: lock F with N's UUID · route to N
  O-)C: moved · reconnect jittered over 2 s
  C->>N: hello, last seq 9,412, then outbox
  N-->>C: nothing missed · acks for resends
  Note over C,N: a few KB each, not 40 MB × 500

%% title: Core entities
classDiagram
  class File {
    +fileKey : string · owner routing key
    +seq : int · last sequenced change
    +lockUuid : string · epoch, fences writes
  }
  class SceneObject {
    +id : string · client:counter e.g. 17:204
    +parent : id · a property like any other
    +pos : fraction · base-95 string
    +props : map · x, fill, text…
  }
  class Change {
    +clientId : int
    +changeId : int · idempotency
    +obj : id
    +props : map
    +seq : int · set by the owner
  }
  class JournalBatch {
    +fileKey : string
    +start : int
    +end : int
    +changes : list · ~200 B each
  }
  class Checkpoint {
    +fileKey : string
    +seq : int
    +blobKey : string · S3, ~10 % of memory size
  }
  class Presence {
    +sessionId : string
    +x : int
    +y : int
    +page : id · memory only, never saved
  }
  File "1" --> "many" SceneObject : tree of
  File "1" --> "many" Checkpoint : snapshotted as
  Checkpoint "1" --> "many" JournalBatch : replay after its seq
  JournalBatch "1" --> "many" Change : holds
  Presence "many" --> "1" SceneObject : on page

%% title: The life of one local change
stateDiagram-v2
  [*] --> LOCAL: applied at once, sent
  LOCAL --> SEQUENCED: ack with seq
  LOCAL --> ROLLED_BACK: rejected, server value
  LOCAL --> RESEND_PENDING: connection lost
  SEQUENCED --> RESEND_PENDING: owner lost it before save
  RESEND_PENDING --> LOCAL: reconnect, outbox resent
  SEQUENCED --> SAVED: saved watermark passes seq
  SAVED --> [*]
  ROLLED_BACK --> [*]
  class SAVED ok
  class ROLLED_BACK bad
  class RESEND_PENDING queue