Development Update — April 12

Skywire: Uptime Tracking Overhaul

The uptime system was completely reworked to produce higher-fidelity data and reduce load on the standalone uptime tracker:

0c18f00a9 TPD: record heartbeat on transport registration — every time a visor re-registers its transports (every 90 seconds), the TPD now records a heartbeat for uptime tracking. This piggybacks uptime data collection onto an operation that’s already happening, rather than requiring a separate HTTP call.

0c955313f Visor: send heartbeat to TPD alongside uptime tracker — the visor now sends heartbeats to both the standalone uptime tracker and the TPD, ensuring continuity during the migration.

f13248aa6 TPD: online status based on 2+ p2p transports — a visor is considered “online” in the TPD if it has at least 2 peer-to-peer transports (STCPR or SUDPH). This is a stronger signal than a simple heartbeat — it means the visor is actually participating in the network, not just registering empty entries.

75136e547 TPD: add v3 uptime endpoint with per-day timeline bitmaps — the new /v3/uptimes endpoint returns per-day bitmaps where each bit represents one hour. This enables the CLI to render visual uptime timelines:

2026-04-10: ████████████████████████  24/24h
2026-04-11: ████████████████░░██████  22/24h
2026-04-12: ████████████████████████  24/24h

0f87b4cca dmsg-discovery: add integrated uptime tracking with /uptimes endpoint — DMSG discovery now also tracks uptime from session events (connect/disconnect). The /uptimes endpoint mirrors the TPD format, giving operators a DMSG-layer view of visor availability.

93ff62f60 Add integrated uptime tracking to service-discovery — service discovery joins the uptime tracking pool, recording heartbeats from service registration events.

Skywire: DMSG Reachability Probes

A new subsystem that verifies visors are actually reachable before accepting their registrations:

55b31ce44 SD: probe visor dmsg reachability before accepting registration — when a visor registers a service (VPN, proxy, public visor), the service discovery now probes the visor over DMSG to verify it’s actually reachable. Unreachable visors are rejected, preventing the listing of phantom services.

c5ee6efac SD: cache dmsg probe results for 5 minutes — probing every registration is expensive. Results are cached per-PK with a 5-minute TTL.

261b35ea8 Autoconnect: probe dmsg reachability before transport establishment — before the visor’s autoconnect logic tries to establish a transport to a remote visor, it first probes DMSG reachability. This catches dead entries in the transport discovery and avoids wasting time on doomed dial attempts.

Skywire: Visor Self-Probes

6b03d41e3 Visor: add periodic self-probe for dmsg listener health — the visor now probes its own DMSG listener every 60 seconds by dialing itself on port 136 (route setup port). If the probe fails twice consecutively, ForceReconnect is called to re-establish DMSG sessions. This catches the case where the visor’s DMSG listener becomes unreachable without the visor knowing (e.g., server-side session drop).

bfb0038e9 dmsg: remove negative route cache, add reachability probe — the negative route cache (introduced earlier to skip known-bad routes) was causing false negatives when network conditions changed. Replaced with explicit reachability probes that provide definitive yes/no answers.

ffa4221ad Visor: wait for DMSG HTTP transport instead of fataling — previously, if the DMSG HTTP transport wasn’t ready at visor startup, the visor would fatal. Now waits with a timeout, allowing the DMSG client to finish connecting.

Skywire: RSN Tuning

880354bbf RSN: increase maxConcurrent from 20 to 128 — the embedded route setup node’s concurrency was too low, causing queuing under moderate load.

b494c1a64 RSN: remove probe from CreateRouteGroup — route setup was probing the destination before creating the route group, but the probe itself consumed a route setup slot. Removing it frees capacity for actual route setups.

dc9870b55 RSN: reduce maxConcurrent from 128 to 32 — 128 was too high, causing ephemeral port exhaustion on the DMSG server. 32 is the sweet spot: enough concurrency for throughput, low enough to stay within port budgets.

31661571f RSN: reduce connection deadline to 30s, add inbound conn deadline — route setup connections that take more than 30 seconds are killed. Inbound connections from visors now also have a deadline, preventing slow clients from holding slots indefinitely.

Skywire: Build and CI

3c1273ed7 Remove goreleaser configs — all the old .goreleaser-*.yml files (darwin-amd64, darwin-arm64, linux, windows) were removed. The release workflow was already migrated to a simpler approach.

0082cf45f Visor: add dmsghttp support for route finder and uptime tracker — these two services can now be reached over DMSG HTTP, completing the DMSG-first service access pattern.

958b3e6f3 Visor: fix embedded RSN/TPS with dmsghttp-only config — when the visor was configured for DMSG-only mode (no HTTP), the embedded route setup and transport setup nodes weren’t initializing their DMSG clients correctly.

318a78d3d dmsg: debounce discovery entry updates on session changes — DMSG entry updates now have a 1-second debounce window. Multiple rapid session changes (common during reconnection storms) coalesce into a single discovery update.