Development Update — April 16

Skywire: CLI Refactor and Skynet Integration

PR #2313 is one of the largest single changes in recent Skywire history — a complete restructuring of the CLI, a new deployment configuration system, and the foundations of the skynet integration.

CLI Directory Reorganization

The command structure was reorganized for clarity:

• cmd/dmsg-commands -> cmd/dmsg — DMSG-specific commands
• Services moved to cmd/svc/ — all deployment service binaries
• Utilities moved to cmd/util/ — helper tools
• cmd/skywirevisormobile removed — dead Android scaffold that was never completed

Cobra command groups introduced across all command trees: skywire cli now has 8 groups (Network, Transports, DMSG, Visor, Services, Utilities, Config, Debug), visor has 5, tp has 3. The flat list of 60+ commands is now navigable.

A new help command with -r (recursive tree), -t (full tree), and -d (documentation) flags replaces the old standalone tree/doc generation scripts. All command README files regenerated.

Deployment Configuration System

deployment.DeployConfig — a unified JSON config format for deployment services. Instead of scattering configuration across environment variables, compose files, and command flags, a single deploy.json file describes the entire deployment: service URLs, public keys, DMSG servers, per-service operational config (ports, Redis URLs, TTLs, pprof), and secret key file paths.

deployment.Keyring — a single encrypted file holding all public/secret key pairs for a deployment (services, DMSG servers, route setup nodes, transport setup nodes, survey whitelist, network monitors, reward system). Can export individual keyfiles and print public keys for sharing.

Uptime CLI

New ut subcommands (sd, mdisc, tpd) with a shared factory for querying uptime data from any of the three discovery services. Timeline graph mode (-T, --per-day, --hours N) renders the v3 bitmap data as shaded hourly blocks — a visual representation of visor uptime directly in the terminal.

tp commands switched from standalone uptime tracker to TPD — all transport commands that queried the standalone uptime tracker now query the TPD’s integrated uptime endpoint instead. The TPD is the authoritative source since transport re-registration is the primary heartbeat.

POST /transports/edges added to TPD — accepts an array of PK hex strings and returns all transports involving those PKs. More efficient than fetching all transports when only specific visors are needed.

Data Fetch Optimization

The tp add command was one of the heaviest CLI operations, fetching the entire transport list and the entire DMSG discovery entry list just to check a few PKs. Three rounds of optimization:

1. Replace /all-transports with /all-transports/per-key-stats — the command only needs transport counts per visor, not full entries. Orders of magnitude smaller.
2. POST /dmsg-discovery/entries/batch — fetch entries for specific PKs instead of all entries.
3. Remove dmsg discovery pre-check entirely — the visor’s DialStream already returns “entry not found” with a clear error. The pre-check was a redundant HTTP round-trip.

DMSG Probe Command

skywire cli dmsg probe <pk> <port> — explicitly test DMSG port reachability for any PK. The noise handshake completing confirms a listener is active on that port. Reports reachable/unreachable with latency.

tp add now probes the destination on DMSG port 136 before establishing transports. If route setup can’t reach the destination, the transport would be useless. --no-probe disables.

Standalone mode (-s flag) creates an ephemeral DMSG client for probing without a running visor.

Circuit Breaker Fix

Source-side dial failures no longer poison the destination’s breaker. A new DialError type carries the failed PK through the error chain; the collector checks it against src/dst before tripping. This fixed a production pathology where popular public visors were blocked for hours because flaky source visors were tripping the destination’s breaker.

E2E and CI

GODEBUG=madvdontneed=1 added to all Go service containers — without this, Go uses MADV_FREE which marks pages as reclaimable but the OS doesn’t actually reclaim them, making RSS appear to grow indefinitely. The production deployment already had this; now E2E matches.

Deployment compose files updated to match production, with dependency-ordered startup stages to avoid healthcheck timeouts on dual-core CI runners.