Development Update — April 22

Skywire: DHT-to-Discovery Pusher

PR #2327/2328 closes the loop on DHT architecture: DMSG servers now push DHT writes back to the HTTP discovery services, enabling a smooth transition where visors can publish to the DHT only while the HTTP APIs stay in sync.

Store.SetOnPut callback — fires after every successful DHT store operation (outside the lock). The DMSG server registers a callback that inspects the salt and forwards the entry to the appropriate HTTP discovery.

DiscoveryPusher — maps salt to discovery endpoint:

• dmsg -> DMSG discovery
• tp -> Transport discovery
• svc -> Service discovery

The transition path is now clear:

1. Today: visors write to HTTP discoveries, services mirror to DHT
2. Next: visors write to DHT, DMSG servers push to HTTP discoveries
3. Eventually: HTTP discoveries become optional read caches

Skywire: DHT-First Lookups

All discovery queries now check DHT first:

• Service discovery (VPN, proxy, public visors) — check local DHT store for “svc” entries before falling back to HTTP. Autoconnect also uses DHT-first for public visor listing.
• DMSG discovery — HybridDiscClient reads from DHT, falls back to HTTP on miss.
• Transport discovery — tp all serves from local DHT store when available.

Lookup counters — dmsg.Client now tracks: cache hits, DHT hits, HTTP hits, HTTP misses. Exposed via dht status CLI command. This lets operators see how much traffic is being served from DHT vs HTTP.

Populate DMSG server cache from DHT — AvailableServers() now reads from the local DHT store, eliminating HTTP round-trips for server discovery.

Skywire: DHT Entry Lookup on DMSG Server

GET /dht/entry/<pk>?salt=dmsg — resolves a PK from the DMSG server’s local DHT store. Any connected DMSG client (including ephemeral ones like dmsgcurl) can query it without needing a DHT node. This eliminates the HTTP discovery round-trip for the most common operation: looking up a DMSG client entry.

GET /dht/entries?salt=dmsg — returns count of entries by salt for monitoring.

Skywire: DHT Backfill on Startup

ed197b47b Add DHT backfill to TPD and SD on startup — when the transport discovery or service discovery starts, it iterates all existing entries and mirrors them to the DHT. This ensures the DHT has the complete dataset immediately after a service restart, not just entries updated since the last restart.

4333c354a Fix TPD DHT Redis mirror: read REDIS_PASSWORD from env — Redis password was hardcoded to empty string, breaking authentication on production Redis instances.

940abc681 Fix Redis DHT Load: pipeline batches + 5min timeout — bulk loading from Redis was sending individual GET commands. Switched to pipeline batches of 100 with a 5-minute overall timeout for large datasets.

2c8004b1c Increase DHT RPC max message size to 4MB — the default 1MB was too small for full sync operations on production-scale data.

fda9f4df8 Fix DHT sync: include storage targets in GetItems response — GetItems was returning items but not their storage targets, making it impossible for the caller to re-store them correctly.

Skywire: DMSG Forward Startup Fix

c841c1b67 Fix/dmsg fwd startup — persisted forwarded ports (from local/forwarded_ports.json) weren’t being restored on visor startup. DMSG listeners for ports with dmsg=true are now created during init. Also fixes:

• Skynetweb: replaced the in-process HTTP reverse proxy (which couldn’t handle WebSocket upgrades) with direct SOCKS5 dial using a per-destination mutex to serialize concurrent DialRoutes to the same PK.
• --whitelist flag added to skynet port add for PK access control.
• Use skynet port as lPort to avoid route descriptor conflicts.
• Fix DHT bootstrap spam: only DMSG servers (which have enable_dht=true) are used as bootstrap peers, not all deployment services.

Skywire: Hardcoded Service URLs Removed

163196280 Remove hardcoded service URLs, use deployment.Prod constants — three files had hardcoded fallback URLs (conf.skywire.skycoin.com, tpd.skywire.skycoin.com). These are now pulled from deployment.Prod constants, making it possible to run against alternative deployments by changing a single config.

Also: DMSG listeners are now created for forwarded ports with dmsg=true, enabling .dmsg:<port> access.

Skywire: Skynetweb Fixes

PR #2323 addresses several issues with the skynet web proxy:

Connection pooling — each SOCKS5 CONNECT was creating a new skynet tunnel, but the routing layer can only maintain one route per (PK, port). Switched to httputil.ReverseProxy backed by an http.Transport that pools connections per destination.

Autoconfig reward display — always show the reward address in output (was hidden behind --verbose, unlike the old autoconfig script).

Hypervisor UI xpub fix — increase maxlength from 40 to 112 for reward address input fields so xpub keys (111 chars) can be entered.

Skywire: CLI Validation and Bug Fixes

PR #2322:

RouteGroup close panic fix — replaced sync.WaitGroup with atomic counter + channel for closeDone. The WaitGroup panicked with “reused before previous Wait returned” when forceCompleteCloseDone returned before the Wait goroutine exited.

ServiceHealth: use direct DMSG client — deployment services run as direct DMSG clients and don’t register in discovery. Now seeds synthetic entries into the DMSG client’s entry cache so health probes can reach them.

DHT full-node status fix — DHTStatus was reading from static config instead of runtime state. SetFullNode changes the store but status was reporting the config value.

FindFullNodes ordering — prioritize peers from the Kademlia routing table (confirmed reachable) over raw bootstrap PKs (which include non-DHT services that fail on dial).

Skynet resolving proxy fixes — removed localhost bypass for self-dial (was causing route descriptor conflicts), fixed HTTP bridge port collision, then replaced the HTTP bridge entirely with direct SOCKS5 skynet dial.