Document the implemented Replicator architecture

- mark capability and lifecycle ADRs as accepted
- add lifecycle, fencing, and provider-extension guidance
- split project terminology into a dedicated glossary
This commit is contained in:
vorotamoroz committed 2026-09-03 06:42:34 +00:00
1 parent d00c5ecc56
commit 2461d37ead
12 files changed
+855 -174

No files matched your search

+21 -7
View File
@@ -1,9 +1,23 @@
---
date: 2026-09-03
commonlib-version: "0.1.21"
self-hosted-livesync-version: "1.0.24"
status: accepted
---
# Architectural Decision Record: P2P Room and Transport Lifecycle
## Status
Accepted — implemented and verified through Commonlib owner tests, the Compose transport suite, and the real-Obsidian setup workflow.
The stable P2P service and room-session owner accepted in
[Replicator Capabilities and Lifecycle Orchestration — Part 2](2026_08_replicator_capabilities_02_p2p_service_lifecycle.md)
supersede only this record's replaceable LiveSync P2P Replicator and ownership
of the replaceable result returned by the `serviceFeature`. This record remains
authoritative for serialised room operations, `room.leave()`, Trystero-owned
physical peers, and relay reconnection.
## Context
Self-hosted LiveSync uses Trystero's Nostr strategy for P2P discovery, signalling, and WebRTC transport. Three related resources have different owners and lifetimes:
@@ -12,7 +26,7 @@ Self-hosted LiveSync uses Trystero's Nostr strategy for P2P discovery, signallin
- Trystero owns the underlying WebRTC peers and may share one physical peer across more than one room; and
- Trystero's Nostr relay manager owns WebSocket clients shared by relay URL.
Closing every `RTCPeerConnection` returned by `room.getPeers()` bypasses Trystero's shared-peer manager. The manager may then retain a stale shared peer and prevent a replacement LiveSync replicator from discovering the same remote peer again.
Closing every `RTCPeerConnection` returned by `room.getPeers()` bypasses Trystero's shared-peer manager. The manager may then retain a stale shared peer and prevent a replacement LiveSync Replicator from discovering the same remote peer again.
Room departure and physical transport destruction are not equivalent. `room.leave()` sends the room-leave action, removes that room's actions and callbacks, and detaches its shared-peer binding. Trystero may retain a healthy physical WebRTC peer for later reuse after the last room binding has gone. The retained peer cannot carry actions for the room which has been left.
@@ -40,7 +54,7 @@ The explicit disconnect operation therefore has the following contract:
This operation is a logical LiveSync disconnection and a physical signalling-server disconnection. It does not promise that every browser-owned WebRTC object has been destroyed synchronously.
An explicit connect resumes relay reconnection before opening a new room. Settings application and database lifecycle replacement close the current LiveSync replicator, discard it, construct a new instance from the current settings, and open that current instance when the configured policy requires it. Commands, event handlers, and panes resolve the current service-feature result at the point of use rather than retaining an obsolete replicator.
An explicit connect resumes relay reconnection before opening a new room. Settings application and database lifecycle replacement close the current LiveSync Replicator, discard it, construct a new instance from the current settings, and open that current instance when the configured policy requires it. Commands, event handlers, and panes resolve the focused views returned by the current P2P `serviceFeature` at the point of use rather than retaining an obsolete Replicator.
Lifecycle operations on one `LiveSyncTrysteroReplicator` are serialised. A close requested while an open is in progress must leave no orphan room serving, and repeated opens must not create parallel rooms. No fixed delay is inserted between close and open: readiness is determined by the actual lifecycle operation and peer discovery.
@@ -50,7 +64,7 @@ P2P setup follows the transport's actual ownership model. Initialising the first
## Ownership
Commonlib owns the LiveSync-specific P2P service, RPC, command, and lifecycle composition. Trystero owns WebRTC peer creation, sharing, reuse, stale detection, and destruction, as well as relay-client reconstruction. The Self-hosted LiveSync host owns the current Commonlib service-feature result and supplies the platform services used by its current replicator.
Commonlib owns the LiveSync-specific P2P service, RPC, command, and lifecycle composition. Trystero owns WebRTC peer creation, sharing, reuse, stale detection, and destruction, as well as relay-client reconstruction. The Self-hosted LiveSync host owns the focused views returned by the current Commonlib P2P `serviceFeature` and supplies the platform services used by its current Replicator.
Self-hosted LiveSync does not add a separate root Trystero dependency. Tests which must observe relay sockets resolve the exact Trystero generation owned by the locked Commonlib package, avoiding two independent transport singletons in one process.
@@ -58,7 +72,7 @@ Self-hosted LiveSync does not add a separate root Trystero dependency. Tests whi
### Close every value returned by `room.getPeers()`
This bypasses Trystero's shared-peer manager and can prevent a replacement replicator from rediscovering the same peer.
This bypasses Trystero's shared-peer manager and can prevent a replacement Replicator from rediscovering the same peer.
### Add a fixed close-to-open delay
@@ -76,15 +90,15 @@ This interferes with Trystero's shared relay clients. The public pause and resum
Commonlib unit tests prove that normal P2P host closure calls `room.leave()` without directly closing Trystero-owned peer connections. Additional package tests cover the action API, replaceable peer-event subscriptions, multiple RPC transport disposers, serialised open and close operations, initialisation of the first device without a central remote, and Fetch running once for an additional device.
Self-hosted LiveSync unit tests prove that settings and database replacement leave panes on the current replicator, and that an explicit P2P rebuild bypasses the policy intended for ordinary replication.
Self-hosted LiveSync unit tests prove that settings and database replacement leave panes on the current Replicator, and that an explicit P2P rebuild bypasses the policy intended for ordinary replication.
The canonical Compose P2P suite uses a real local Nostr relay and WebRTC implementation. It covers ordinary two-peer synchronisation, replacement of the active LiveSync replicator followed by discovery and transfer with the same peer, and explicit relay disconnection followed by paused and resumed reconnection. The lifecycle scenario is exposed only through a Docker test build and an injected CLI command runner; it is not part of the public CLI command surface.
The canonical Compose P2P suite uses a real local Nostr relay and WebRTC implementation. It covers ordinary two-peer synchronisation, replacement of the active LiveSync Replicator followed by discovery and transfer with the same peer, and explicit relay disconnection followed by paused and resumed reconnection. The lifecycle scenario is exposed only through a Docker test build and an injected CLI command runner; it is not part of the public CLI command surface.
The real-Obsidian P2P Setup URI workflow creates the first device, generates the second-device URI from it, accepts each peer visibly, and verifies a two-way note round-trip through a local relay. A separate focused pane test covers the principal connection control and teardown without requiring a remote peer. Transport replacement and relay-socket lifecycle remain owned by the package and Compose tests rather than being duplicated in Obsidian.
## Consequences
- Replacing a P2P replicator no longer leaves host views or commands bound to an obsolete instance.
- Replacing a P2P Replicator no longer leaves host views or commands bound to an obsolete instance.
- Explicit signalling-server disconnection has a testable socket-level meaning without claiming immediate destruction of idle WebRTC objects.
- Settings which change the relay, room, passphrase, or TURN configuration can replace the whole LiveSync room safely.
- Trystero may reuse healthy peers across room lifecycles, reducing unnecessary renegotiation.