[myleshorton]

Stacked on #688. Merge that first, and GitHub will retarget this PR to main.

Why

Advanced stats shows speed, session data and per-country bytes for Unbounded volunteers, but nothing for direct sharing (Share My Connection): radiance had no byte counts for peer connections. getlantern/lantern-box#371 attaches a live peerconn.Counter to every samizdat connection, and this PR turns those counters into something the app can display.

What changes

• peer.relayStats sums the counters per session and per client IP. Open connections are read live. A close folds in the connection's final count, so the totals keep it after the connection ends.
• Open connections are keyed by their *Counter, not by source. samizdat multiplexes streams over one TCP connection, so several open streams can share an ip:port. The same counter returns on each stream's close, so verifier streams are now matched by counter too, and peer: hide lantern-cloud's verify callback from connection events #688's per-source count is only the fallback for events without one.
• Tallies are pruned as new connections arrive as well as on each snapshot, so a session no one polls doesn't keep every client it has seen. On Stop, end() starts a retention timer for connections still open, whose closes are dropped while draining. They're read live for 60s, then folded into the totals and released.
• Client.RelaySnapshot() → LocalBackend.PeerRelaySnapshot() → GET /peer/snapshot → ipc.Client.PeerRelaySnapshot(). The JSON matches unbounded.Snapshot (epoch, running, peers, bytesDown, bytesUp, peerBytes[{ip,down,up}]), so the app can fold it with its existing relay counter code.
• Like Unbounded's, the epoch is fixed for the process, and totals are cumulative across sessions. A Stop and Start between two polls loses nothing, and connections a stopped session left open are folded in at the next begin. Each session has a generation that also advances at end, so a listener callback that resumes after Stop (or after Stop and Start) isn't counted.
• A departed client's totals stay in the snapshot for 60s, matching unbounded.peerRetention, so the app's 1s poller sees its final count.
• The verifier connections peer: hide lantern-cloud's verify callback from connection events #688 filters never reach the tally.

sequenceDiagram
    autonumber
    participant SB as samizdat inbound<br/>lantern-box#371
    participant L as peerconn listener<br/>peer.go
    participant R as relayStats<br/>snapshot.go
    participant IPC as /peer/snapshot<br/>ipc/server.go
    participant App as lantern-core poller

    SB->>L: accept {Source, Destination, Bytes: *Counter}
    L->>L: isVerifierConn? drop (#688)
    L->>R: record(+1, source, counter) ⚠️
    Note over SB: wrapConn Read/Write<br/>add to the counter
    App->>IPC: GET every 1s
    IPC->>R: snapshot(): closed totals + live counters
    SB->>L: close {Source, Bytes}
    L->>R: record(-1): fold final count

Loading

Dependency bump

lantern-box goes from v0.0.154 to v0.0.166 (the release with #371), and the sing-box-minimal replace moves to the version that lantern-box main requires (mux.ContextWithSessionConn). This picks up lantern-box v0.0.155 through v0.0.165 as well. go mod tidy was run, and go.mod and go.sum are committed together. #371 is merged and released as v0.0.166, which this now pins.

Testing

• relayStats tests: live and closed counts; per-IP sums across one client's connections; retention pruning; a new epoch per session; the totals surviving Stop; uncounted and unmatched events ignored.
• TestClient_RelaySnapshot: end to end through a started client, with the verifier excluded and a poll after Stop still seeing the bytes.
• go test -race ./peer/ ./ipc/ ./backend/ passes, and go test passes for every package except cmd/lantern, whose ipc.NewClient call is already broken on main.

Pre-PR review (local Codex gate)

• Rounds: 3 — final: CODEX GATE: verdict=approve critical=0 important=0 minor=0 reviewed=23
• Exit: approve
• Fixed:
  • peer/snapshot.go:52 — open connections were keyed by ip:port, so a second multiplexed stream from one source overwrote the first one's counter (found while rebasing on peer: hide lantern-cloud's verify callback from connection events #688's fix). Now keyed by counter, with TestRelayStats_StreamsSharingASource.
  • peer/snapshot.go:130 (r1, important) — tallies were pruned only on snapshot, so an unpolled session grew with every client it saw. Now also pruned on accept, with TestRelayStats_PrunesWithoutSnapshots.
  • peer/peer.go:535 (r1, minor) — connections open at Stop were never retired. end() plus time-based retirement, with TestRelayStats_RetiresOpenConnectionsAfterStop.
  • peer/peer.go:536 (r2, important) — per-source verifier matching could hide an ordinary stream's close instead, leaving its counter held. Now matched by counter, with TestClient_OrdinaryCloseBeforeVerifierOnSharedSource (fails on the old code).
• Dismissed: none

🤖 Generated with Claude Code

https://claude.ai/code/session_01LQG2dRhaocR2UY5RPamm39

Summary by CodeRabbit

• New Features
  • Added relay traffic snapshots showing cumulative data transferred, per-peer byte totals, and whether sharing is active.
  • Snapshots retain totals across sessions and provide a brief window to view statistics for recently stopped sessions or departed peers.

[coderabbitai]

Note

Currently processing new changes in this PR. This may take a few minutes, please wait...

⚙️ Run configuration

• Configuration used: defaults
• Review profile: CHILL
• Plan: Advanced
• Run ID: e800c6b7-0042-4bdf-bfb9-6ebd8378e833

📥 Commits

Reviewing files that changed from the base of the PR and between 5043a9a and c050d89.

⛔ Files ignored due to path filters: 1

• go.sum is excluded by !**/*.sum

📒 Files selected for processing: 9

• backend/peer_share.go
• backend/radiance_test.go
• go.mod
• ipc/client.go
• ipc/server.go
• peer/peer.go
• peer/peer_test.go
• peer/snapshot.go
• peer/snapshot_test.go

+------------------------------------------+
| 💣 Deploying bug fixes in 3... 2... 1... |
+------------------------------------------+
  \
   \     /)/)
        ( '')
        /  |
      o(___)

✨ Finishing Touches

📝 Generate docstrings

• Commit to this branch
• Create a new PR

🧪 Generate unit tests (beta)

• Commit to this branch
• Create a new PR

• Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

_{Comment @coderabbitai help to get the list of available commands.}

[Copilot]

🟡 Changes recommended

Connection accounting and session lifecycle handling can produce incorrect totals and retain departed peers indefinitely.

3 open issues

• Prevent stale callbacks from recording after restart · New
• Key multiplexed streams by counter identity · New
• Retire open connections during Stop · New

What changed in this PR

Adds direct-sharing traffic snapshots so the app can display session and per-client byte totals alongside Unbounded statistics.

Changes:

• Tracks live and closed connection counters by session.
• Exposes snapshots through the backend and /peer/snapshot IPC endpoint.
• Adds accounting tests and updates supporting dependencies.

File	Description
peer/​snapshot.go	Implements relay accounting and snapshots.
peer/​snapshot_test.go	Tests totals, retention, epochs, and client integration.
peer/​peer.go	Connects accounting to session and connection events.
ipc/​server.go	Adds the snapshot endpoint.
ipc/​client.go	Adds the snapshot client method.
go.sum	Updates dependency checksums.
go.mod	Updates counter-supporting dependencies.
backend/​radiance_test.go	Extends peer controller test doubles.
backend/​peer_share.go	Exposes snapshots through the backend.

_{Review effort: Balanced}

[Copilot]

🔵 Needs a closer look

An unresolved teardown race and stacked networking dependency upgrades require fixes and human integration review.

1 open issue

• Verifier traffic can be counted after Stop during draining · New

3 resolved since last review

• Key multiplexed streams by counter identity
• Prevent stale callbacks from recording after restart
• Retire open connections during Stop

_{Review effort: Balanced}

[Copilot]

🟡 Changes recommended

Restarting can discard counters still receiving bytes during teardown, permanently undercounting relayed traffic.

1 open issue

• Stop→Start drops bytes from still-draining old sessions · New

1 resolved since last review

• Verifier traffic can be counted after Stop during draining

_{Review effort: Balanced}

[Copilot]

🔵 Needs a closer look

Concurrent teardown and restart handling, prerequisite integration, and the planned dependency retagging require final human validation.

No open issues 🎉

1 resolved since last review

• Stop→Start drops bytes from still-draining old sessions

_{Review effort: Balanced}