Repository navigation
Node Manager 0.3.1-rc.1 (release candidate)
Pre-releaseRelease candidate for Sentinel Node Manager 0.3.1 (Solana devnet). Built from one commit for every OS: the app bundle is byte-identical on all 6 OS/CPU builds.
| System | File |
|---|---|
| Windows 10/11 x64 | Sentinel-Node-Manager-0.3.1-rc.1-win-x64-setup.exe (or -win-x64-portable.exe) |
| Windows 11 arm64 | Sentinel-Node-Manager-0.3.1-rc.1-win-arm64-setup.exe |
| macOS Apple silicon / Intel | …-mac-arm64.dmg / …-mac-x64.dmg (or .zip) |
| Ubuntu / Debian | …-linux-amd64.deb / …-linux-arm64.deb |
| Fedora / RHEL | …-linux-x86_64.rpm / …-linux-aarch64.rpm |
| Any Linux x64 | …-linux-x86_64.AppImage |
Node image 0.3.7 (also on release node-image-0.3.7); the app downloads it and checks the pinned SHA-256. Check any file with sha256sum -c --ignore-missing SHA256SUMS.
- HNS Resolver works for the node's users: with the switch on, a client connected over WireGuard,
AmneziaWG or OpenVPN can open Handshake names (whatever DNS server it asked, port 53 goes to the
node's resolver); ordinary names still go to 1.1.1.1. V2Ray, Xray and Hysteria2 nodes resolve
destinations on the node, through the same resolver, and DNS a Hysteria2/V2Ray/Xray client sends to
any server is answered by it too. Node page: "Routing users' DNS". Needs the new
node image 0.3.7 (amd64 + arm64, SHA-256 pinned); turning HNS on moves a node to it. In that image
an hnsd crash restarts hnsd only (upstream stops the whole node daemon when hnsd exits), and
hnsd's per-block output (~500k lines while syncing) is summarised once a minute instead of
filling the node log; hnsd keeps its chain checkpoint, so a restart no longer re-syncs
the whole Handshake chain (~3 min) and leaves no temp files behind; its log can never stall it.
After a node's first sync hnsd is restarted once from its checkpoint, which frees ~150 MB of
memory for good (163 MB → 9.5 MB; measured cost of HNS on a 2 GB VPS: ~85 s and 210 MB download for
the first sync, then ~0% CPU and ~1 MB/h).
The node page shows the Handshake sync ("Syncing · 45%", and an error asking for outbound TCP
when hnsd reaches no Handshake peer or gets no new block for 45 min; "Restarting…" during a
restart) instead of "Routing users' DNS" as soon as a process named hnsd exists. The paid
self-test checks HNS names on every protocol (V2Ray/Xray through their proxy).
An HNS node on an older node image moves to the current one at its next Restart, protocol switch
or price change (the page says "Restart to update the node image"), so it gets these fixes. - Chain actions without Docker on this computer: withdraw, bond withdrawal, removal and test-escrow
refunds run on one of your servers when Docker Desktop is off (Settings → Chain-call server). The
removal card says why it is waiting and offers Start Docker Desktop / Use a server instead. - Removal never dead-ends: Retry now, the cleanup reason, "Retry cleanup now" and "Server is gone";
a local node whose Docker is gone is forgotten and its container deleted when Docker returns. - Test escrow comes home by itself: the paid self-test starts the channel close right away and the
app finishes it after the grace period (Wallet → Test escrow card, Reclaim now). - Wallet: SOL Max leaves the fee and rent; errors on a node stay on its page (red card) instead of
only in the log. - Fixes: a damaged app data file is restored from a rolling backup instead of starting empty (that
lost every node and its key); "getting the chain-call image" no longer repeats every minute with
Docker off; the escrow sweep no longer writes the same log line every 10 minutes; node logs are
capped at 3 × 20 MB (hnsd logs ~680k lines while syncing). - Two changes on one node at once (e.g. a protocol switch and the HNS switch) are refused with
"… is still running on this node" instead of racing; while a change restarts the node its page says
"Switching protocol…" instead of "Listing expired", and a server node the app has no SSH access
to after a restart reads "Not connected" (not "Error") until Reconnect, without a "not live" alert
or a Docker install hint at every launch. - After a deploy, the button reads "Save phrase to open node" until the recovery phrase is saved; the
tour explains the 0.06 SOL per node (0.05 to the node key + fees). Live sessions no longer lists
sessions the node has already closed (the paid self-test stayed there for an hour).
Tested: unit 407/407, end-to-end on devnet 109/0 (HNS on all six protocols) + refunds 21/0, clean installs and upgrades on Windows, macOS and Linux (data kept), a ~10 h soak on all three OS with paid tests, an OS reboot and a firewall block (red status, back to green within 60 s of unblocking).
Known: after an app restart a server node needs Reconnect before its chain actions run; the app runs one short-lived helper container per chain read while a node page is open. Devnet only; tokens on Solana devnet have no value.