Skip to content

Arteria 2.1.0

Latest

Choose a tag to compare

@github-actions github-actions released this 15 Nov 23:00
· 6 commits to main since this release
c6c4a04

Fixes for what the SNBForums thread reported after 2.0.0: AiMesh nodes that crashed while Arteria read them, a page that took seconds to go live, a menu that stopped updating, and ports out of order on the RT-AX88U. Also a page that follows ROG and TUF routers' colours, room for more nodes than the map has zones, and four devices drawn anew. Arteria is now in amtm (7.2 and later) as ar. Thanks to everyone in the thread who tested development versions and posted their output.

Fixes

  • The page is live about a second after it opens. 2.0.0 spaced samples by how long each took, so the first sample, which does the most work (reading the ports, starting the node sessions), held the second one back by 3 to 6 seconds, and the page waited on Connecting… for it. The second sample now comes a second after the first, the names are written alongside instead of before it, and the page checks for samples every second until it is live.
  • AiMesh nodes where reading crashed ("Segmentation fault") — reported on GT6, XD6, RT-AX56U, RT-AX58U, RT-AX86S and RT-BE9700 nodes:
    • Arteria no longer reads a node's uplink. On a node with a Wi-Fi backhaul, the radio's own interface is a client of the router, and the node's devices are on wl0.1-style interfaces; Arteria read the uplink as if devices were on it.
    • A node on an Ethernet backhaul still names its radios as possible uplinks in its settings, but its devices are on them, so Arteria reads them while they work as access points.
    • The node's login shell now runs Arteria's script, as it would run it typed in a terminal there, instead of sh -s.
    • The trace nodes test and doctor save when something crashes no longer needs sh -x, which those nodes' shells refuse (that is why node-trace.txt held only "invalid option -- 'x'"). It names the command that crashed and the node's shell.
    • doctor checks a node the way nodes test does. It said a node was fine when the script had never run there.
  • A node's part of Flow adds up. The bar of a node on a Wi-Fi backhaul could be taller than the ribbons to its devices, for a few seconds after traffic changed. The router counts the backhaul every sample, but the node counts its devices once a round, and those figures reach the page a round later. In the expanded view the minimum width of each quiet device's ribbon hid the gap. The node now also counts its own uplink in the same round, and its part of Flow and the map use that figure. A node on a wired backhaul keeps the router's. A node that joins its radios' uplinks into one interface (dpsta, as on the RP-AX56) counts that interface alone, since it counts every packet its radios do.
  • No phantom node. An old entry in AiMesh's own list with an address outside the router's LAN (often 192.168.50.x, ASUS's default) showed as a node that's down. Such entries are ignored.
  • Ports are numbered as printed on the router. Arteria takes the numbers from the firmware's own port map, where it has one. On the RT-AX88U, LAN 1 to 4 showed in reverse, and LAN 5 to 8, which sit behind one link to a second switch chip, showed as a device on LAN 5 even with nothing plugged in, at that link's speed. They now show as the port that's in use (LAN 7), at its own speed, and nothing shows when none of them is.
  • The menu is updated when it's mounted twice. Every addon with a page mounts the menu copy addons share (/tmp/menuTree.js) over the router's own. An addon that mounts it again without unmounting it first leaves two mounts, one over the other. The router's pages still show the menu, but the copy can then no longer be replaced, so Arteria's update failed ("mv: can't rename … Device or resource busy"). The menu stayed as it was without a word, and a temporary copy was left in /tmp/arteria. Arteria now takes the extra mounts off, replaces the copy, and mounts it once; if even that fails, it leaves the menu as it was and says so. doctor reports a menu mounted more than once, and doctor fix mounts it once.
  • update says when it replaces a development version with the release of the same number (Updating 2.0.0 (development version: … ) -> 2.0.0, the release). It said Updating 2.0.0 -> 2.0.0.
  • "Weak backhaul" only when it is. The verdict is now based on the backhaul link that carries the node's traffic. While the node is quiet, it's based on its fastest link that isn't 2.4 GHz. The speed of an idle link's last packets is no longer counted: they are control frames, sent slowly on purpose. A 6 GHz link is allowed a weaker signal (-78 dBm), as it reads lower for the same quality.
  • Network names instead of "guest". A device on one of the router's guest networks, or on another of Guest Network Pro's networks, shows that network's name. Devices on a main network show none, even when the router serves it on a wl0.1-style interface: Arteria used to call every such device a guest.
  • The page's requests reach Arteria first. The router runs the lines in /jffs/scripts/service-event in order, so a line before Arteria's that ended the script, or waited, kept the page's requests out: the page said Connecting… for ever. Arteria's lines in service-event and services-start now come first, right after #!/bin/sh. They only start Arteria in the background, for its own events, so they never hold up the lines after them. Installing or updating moves them.

Safer on the router

  • nodes test and doctor take at most about 1 MB from a node, as the page's own node sessions already did: a node that sent output without end could fill the router's RAM.
  • The menu copy addons share is edited only when it is root's own file with a single name. A hard link to another file in /tmp passed the old checks, so that file's content could end up readable as the menu. Such a copy is now replaced with the firmware's menu.
  • What doctor and debug show of other programs: another program's line in service-event by the name of the program it runs only (the rest of the line can hold a password or a token), the program holding the sampler's lock by its name only, and the system log only from root's own files.
  • Arteria's lines in service-event and services-start are added and removed safely. The new file is made and checked in full first, and the old one is put back if writing it doesn't come out whole, so a full JFFS can't cut it short and lose other addons' lines.
  • The port map leaves out the AiMesh nodes' ports, which the firmware lists after the router's own: a node's port can't light up one of the router's.

Doctor and debug

  • nodes test explains a node that reads 0 devices. It lists the node's interfaces it reads and the ones it skips as the node's uplink, with the devices on each, and doctor does the same. Devices on a skipped interface mean Arteria took the wrong interface for the uplink on that node.
  • doctor tests what the page does when it opens. It sends the same request as the page, start_arteria, and says whether it reached Arteria and how long it took, and whether samples followed. It then stops sampling again. When something goes wrong it shows:
    • the router's system log lines about requests it delayed or dropped while busy with another one;
    • the sampler's own errors, which used to go nowhere;
    • the other lines in service-event that wait for their program on every request.
  • doctor also finds and fixes a hook file saved with Windows line endings, which the router can't run at all, Arteria's line not being first, and a process left holding the sampler's lock (every sampler the page starts would give up at once).
  • debug shows more, still without names or addresses:
    • each Wi-Fi interface's band setting and channel, mode and bridge, and whether its network is a main one;
    • the firmware's port map;
    • each node's backhaul links (band, signal, the speed of the last packets each way);
    • which sections the router's AiMesh client list has;
    • the sampler's last errors;
    • the system log's lines about the page's requests.

The page

  • Labels stay off other devices on a crowded map. With many devices, in the expanded view especially, the last labels to be placed found every spot near their device taken, and one could end up over another device's drawing or over the zoom buttons. Once all labels are placed, a label still covering a device now moves to a free spot further out, with a longer line.

  • More AiMesh nodes than the map has room for. The map has a zone for two nodes, three in the expanded view. With more, the first ones in the list got the zones, and the others weren't drawn at all, without a word. Now the busiest nodes get the zones, and keep them unless a node without one gets much busier. The tray names the nodes without a zone (+1 node), and its list says their devices are in it. Flow shows every node, as before.

  • Colours that follow the router's theme. On ROG and TUF routers, whose pages are black with red or amber, Arteria's controls, highlights and room turn black with red or amber too. The traffic's colours stay as they are everywhere.

  • Less motion. When the device asks for less motion (Windows' animation effects off, macOS's Reduce motion, Android's Remove animations), the packets stand still. That used to happen without a word; now the map's legend says so, with an Animate button, which the browser remembers.

  • Four devices drawn anew:

    • the VR headset: a curved glass visor with a soft band;
    • the car: a smooth electric saloon with a glass roof;
    • the washing machine and dryer: round porthole doors;
    • the smart plug: a round plug in a wall socket.

    Their icons match.

Install or update

Over SSH on the router:

d=/tmp/arteria-install; rm -rf "$d" && mkdir -m 700 "$d" && curl -fsL --retry 3 "https://github.com/IfSugar/arteria/releases/download/2.1.0/arteria.sh" -o "$d/arteria.sh" && sh "$d/arteria.sh" install; rm -rf "$d"

Already installed? Run /jffs/addons/arteria/arteria.sh update. Your settings and node access are kept.

Full diff: 2.0.0...2.1.0