Luci app qosify counters by choppyc79 · Pull Request #9051 · openwrt/luci

choppyc79 · 2026-09-19T17:54:09Z

Pull request details

Description

Depends on #9046, #9047, #9048, #9050. Only the last commit is this PR; draft until those are merged.

New Counters tab, polled at LuCI's interval and only while it is open:

Traffic by Class: ubus call qosify get_stats per class (dscp, packets,
bytes, share, total), bars cube-root scaled, grouped and coloured by the
CAKE tin the class's egress codepoint lands in
Traffic by CAKE Tin: the qosify-status tc output summed tin by tin across
egress and ingress (pkts, bytes, drops; marks on hover)
DNS Entries: ubus call qosify dump dns entries with hits/packets/bytes from
the get_stats dns table

Works with both get_stats shapes: master (beeb87e) and 1501e09 as shipped
in 24.10/25.12, which has per-class packets only and no dns table. There,
the daemon-info and DNS boxes stay hidden and dump is not called.
Everything is cleared while qosify is stopped.

ACL: read group gains qosify get_stats and dump. Traffic by CAKE Tin uses
the existing qosify-status exec grant, so a read-only session does not
see it.

Screenshot or video of changes

Maintainer

@choppyc79

Tested on

OpenWrt version: SNAPSHOT r36237-2ed2c210c3
LuCI version: 26.257.37274~f4eebd4
Web browser(s): Google Chrome

Checklist

(Nice to have) Includes what Issue it closes (e.g. openwrt/luci#issue-number).
(Nice to have) Includes what it depends on (e.g. openwrt/packages#pr-number in sister repo).

openwrt-ai

Commit checks

7253035 "report an unanswered status call as unknown" — body says the start, stop, restart and reload buttons "report the failing call's own error", but they now throw a synthesized "rpcd is not answering ... service state is unknown" when waitForState() returns null. The waitForRunning/waitForStopped → waitForState unification and the new applyService abort are also undescribed; mention both, or drop the "own error" sentence.
aeff225 "build the page from stock LuCI markup" — body says "No behaviour or string changes", but the diff adds a No classes defined in %s empty-state row to the Config reference class box (qos-cfg-cls), which had none before. Either note that empty state in the body or move it to its own commit.

Generated by Claude Code

openwrt-ai · 2026-09-20T17:08:55Z

 	applyService:function(){
 		var self=this;
-		return L.resolveDefault(callServiceList('qosify'),{}).then(function(r){
+		return callServiceList('qosify').catch(function(){throw new Error(_('rpcd is not answering for qosify, so the service state is unknown.'));}).then(function(r){


An unanswered service.list now aborts before any rc.init call, so a config writeConfig()`` has already written to disk is left unapplied and reported as Save failed. The stale-ACL case this commit's message names grants `rc` but not `service`, and there the old path still reached `rc.init start`. Better to fall through to `rc.init` with the running state unknown than to throw before trying.

Generated by Claude Code

openwrt-ai · 2026-09-20T17:09:05Z

+		if(ctx.active==null&&enChecked){el.className='label warning';dom.content(el,_('Status Unknown'));}
+		else if(ctx.active){el.className='label success';dom.content(el,_('Active'));}
+		else if(ctx.running&&enChecked){el.className='label warning';dom.content(el,_('Enabled — Not Shaping (check config)'));}


ctx.running===null with ctx.active===false — service.list unanswered while qosify.status answered, which gatherCtx() now produces since the two facts go null independently — falls through to Enabled — Not Running. That contradicts the Unknown svcNodes()`` puts on the Running row from the same ctx.

Suggested change

if(ctx.active==null&&enChecked){el.className='label warning';dom.content(el,_('Status Unknown'));}

else if(ctx.active){el.className='label success';dom.content(el,_('Active'));}

else if(ctx.running&&enChecked){el.className='label warning';dom.content(el,_('Enabled — Not Shaping (check config)'));}

if(ctx.active==null&&enChecked){el.className='label warning';dom.content(el,_('Status Unknown'));}

else if(ctx.active){el.className='label success';dom.content(el,_('Active'));}

else if(ctx.running==null&&enChecked){el.className='label warning';dom.content(el,_('Status Unknown'));}

else if(ctx.running&&enChecked){el.className='label warning';dom.content(el,_('Enabled — Not Shaping (check config)'));}

Generated by Claude Code

openwrt-ai

Reviewed 5 new commits; no new issues found.

Generated by Claude Code

rpc.declare() resolves a ubus error status unless reject is set, so a status code (6 once the session's ACL no longer covers the object, 4 when the object is gone) stays in result[0] and expect {'': {}} rewrites it to the same {} a working call with nothing to report returns. L.resolveDefault() cannot tell the two apart, so an ACL left over from an older install, a stale session or a stopped rpcd reads as a stopped, unshaped qosify with no init script. Declare rc.list, service.list and qosify.status with reject: true and catch each rejection to null. running, enabled, hasInit, active and shaped are then null rather than false when the call was not answered, and the Overview service table shows an amber Unknown badge for each, with the cause named once, on the first row that reads Unknown. The qosify object goes with the daemon, so an unanswered status call while qosify is stopped still reads as not shaping. The autostart button carries the state as its label, so it shows Unknown and is disabled while the state is unknown. Its click now reads the state from the last refresh rather than from when the button was built, so a tick that changes the state also changes the action. The start, stop, restart and reload buttons stay clickable. When the service.list poll after a start, restart or stop goes unanswered, the action fails with "rpcd is not answering" rather than reporting a result it could not check, and a stop then skips /usr/share/qosify-luci/cleanup instead of cleaning up. A save no longer warns that qosify is not shaping when the check could not run; it says shaping could not be checked. When service.list goes unanswered while a save applies, qosify is restarted rather than reloaded or started, since the config is already written and the running state is unknown. That restart also starts a qosify that was stopped. The Status tab says rpcd is not answering rather than that qosify is not running, and says the interface summary is unavailable when only qosify status went unanswered. updateEnBadge() gives the Config tab's QoS Enabled badge the same treatment: with the section enabled and the status call unanswered it reads Status Unknown, rather than the Enabled — Not Shaping (check config) or Enabled — Not Running a null active used to fall through to. A section disabled in UCI still reads Disabled — that state comes from the config, not from the call, so an unanswered call does not put it in doubt. po/templates/qosify.pot regenerated with build/i18n-scan.pl. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Signed-off-by: Ash Clarke <clarkeaj@hotmail.co.uk>

A rules edit only needs the mapping files re-read, but the only way to get that from the page is Reload, the init script's reload_service(), which pushes the whole config. Add Reload Rules next to it: ubus call qosify reload runs qosify_map_reload(), which re-reads the files in the defaults list and leaves the qdiscs, the interface config and the dynamically added entries alone. Add Check Devices to the Advanced tab: ubus call qosify check_devices runs qosify_iface_check(), which starts a section whose device now exists and stops one whose device has gone, so a device that appears after qosify started is picked up without a restart rebuilding every qdisc. The method arms a 10 ms uloop timer and returns before the pass runs, so the page waits before refreshing the service state. Both methods return an empty reply, so they are declared with reject: true; without it a failed call would report success. Both are in qosify 1501e09 (24.10, 25.12) and master. Grant them in the write ACL group. po/templates/qosify.pot regenerated with build/i18n-scan.pl. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Signed-off-by: Ash Clarke <clarkeaj@hotmail.co.uk>

Both pollers pass a fixed 10 s to poll.add(), so the page ignores luci.main.pollinterval. Leave the interval out so poll.add() takes L.env.pollinterval (5 s unless set), which the user can change. On the default setting the Status tab now forks qosify-status every 5 s rather than 10 s. It still does so only while that tab is open, and Poll.step() still holds the next tick until the last one settles, so a fork slower than the interval skips ticks instead of stacking up. A target that cannot afford the fork at that rate raises luci.main.pollinterval, which slows the rest of LuCI's polling with it, instead of the page choosing its own rate. Pausing with the header refresh toggle is unchanged, as it stops the whole poll loop. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Signed-off-by: Ash Clarke <clarkeaj@hotmail.co.uk>

The page draws its own badges, key/value tables, reference boxes and muted notes with qos-* classes and inline styles, inside fieldset and legend sections, so it looks the same under every theme and ignores the theme's own table, label and section styles. Use the markup the themes already style instead: - sections are div.cbi-section with an h3 title - status and file validity use .label success, warning and danger - key/value tables are .table with .tr and .td rows - Quick Settings and the Advanced file rows are .cbi-value rows, with hints as .cbi-value-description - notes are cbi-section-descr, <p> or <em>, and code is <code> qosify.css now draws each section as a box with a title bar from the theme's own variables, gives the two editors and the tc output their size, and lays out the Quick Add rows. Bootstrap and OpenWrt 2020 have no .label.danger, so the sheet supplies it from --error-color-high. The editor width moves from the qos-edit class to the textarea ids with box-sizing: border-box, next to the height, and the editors resize in both directions, so the theme's textarea width cannot win and a wrong width stays correctable. The Config Reference class list becomes a .table like the one on the Rules tab, so it now picks up that tab's existing "No classes defined in %s" row when no class or alias is configured, where the plain div rendered nothing. That is the only behaviour change, and no string is added or altered. po/templates/qosify.pot regenerated with build/i18n-scan.pl for the moved source lines. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Signed-off-by: Ash Clarke <clarkeaj@hotmail.co.uk>

qosify keeps per-class counters and, on master, per-DNS-pattern and eBPF map statistics, but the page shows none of them. Add a Counters tab, polled like the others and only while it is open. Traffic by Class lists the classes from ubus call qosify get_stats with their dscp, packets, bytes and share, and a total row. Bar length is the cube root of each row against the largest, so one bulk flow does not flatten every other row; the share column stays exact. Rows are grouped and coloured by the CAKE tin their egress codepoint lands in under the configured diffserv mode, and AFx1 to AFx3 sort by drop precedence. The tcp_default and udp_default slots are skipped, as a dscp_default_* naming a class is already counted against that class. get_stats has two shapes. qosify master (beeb87e) returns a classes table plus ebpf_map_entries, last_reload_time, dns_cache, dscp and dns; 1501e09 (24.10, 25.12) returns qosify_map_stats() at the top level, one table per class with packets only. Both are read, and only the fields the reply carries are shown. Traffic by CAKE Tin graphs the qosify-status output tin by tin, egress and ingress summed from the pkts, bytes and drops rows tc prints, with marks on hover. Qdiscs running a different diffserv mode get a table of their own. It needs the existing qosify-status exec grant, so a read-only session does not get it. DNS Entries lists the dns entries from ubus call qosify dump with hits, packets and bytes from the get_stats dns table. It is shown only when get_stats carries that table, so dump is never called on 1501e09. While qosify is stopped the figures are cleared rather than left on screen looking live. get_stats and dump are declared with reject: true, so a failed call is reported rather than read as an empty reply. Tin colours use the theme's error, primary, warn and success colours, with light and dark fallbacks for the extra diffserv8 and precedence tins. Grant get_stats and dump in the read ACL group. po/templates/qosify.pot regenerated with build/i18n-scan.pl. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Signed-off-by: Ash Clarke <clarkeaj@hotmail.co.uk>

choppyc79 · 2026-09-21T09:24:11Z

Folded into #9054, which carries this commit unchanged as part of the
series. Closing in its favour.

This was referenced Sep 19, 2026

Luci app qosify service section #9052

Closed

Luci app qosify quick settings #9053

Closed

luci-app-qosify: status, counters and page rework #9054

Open

openwrt-ai reviewed Sep 20, 2026

View reviewed changes

choppyc79 force-pushed the luci-app-qosify-counters branch 2 times, most recently from a4a6b40 to 0b4459e Compare September 20, 2026 10:50

choppyc79 requested a review from openwrt-ai September 20, 2026 11:02

openwrt-ai reviewed Sep 20, 2026

View reviewed changes

choppyc79 force-pushed the luci-app-qosify-counters branch from 0b4459e to 8d0a3e5 Compare September 20, 2026 18:36

choppyc79 requested a review from openwrt-ai September 20, 2026 18:42

openwrt-ai reviewed Sep 21, 2026

View reviewed changes

choppyc79 force-pushed the luci-app-qosify-counters branch from 8d0a3e5 to 721d151 Compare September 21, 2026 06:49

choppyc79 and others added 4 commits September 21, 2026 07:58

choppyc79 force-pushed the luci-app-qosify-counters branch from 721d151 to 433ce13 Compare September 21, 2026 07:03

choppyc79 closed this Sep 21, 2026

Navigation Menu

Luci app qosify counters - #9051

Luci app qosify counters#9051
choppyc79 wants to merge 5 commits into
openwrt:masterfrom
choppyc79:luci-app-qosify-counters

choppyc79 commented Sep 19, 2026 •

edited

Loading

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

openwrt-ai left a comment

Uh oh!

openwrt-ai Sep 20, 2026

Uh oh!

openwrt-ai Sep 20, 2026

Uh oh!

openwrt-ai left a comment

Uh oh!

choppyc79 commented Sep 21, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Conversation

choppyc79 commented Sep 19, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Pull request details

Description

Screenshot or video of changes

Maintainer

Tested on

Checklist

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

openwrt-ai left a comment

Choose a reason for hiding this comment

Commit checks

Uh oh!

openwrt-ai Sep 20, 2026

Choose a reason for hiding this comment

Uh oh!

openwrt-ai Sep 20, 2026

Choose a reason for hiding this comment

Uh oh!

openwrt-ai left a comment

Choose a reason for hiding this comment

Uh oh!

choppyc79 commented Sep 21, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

choppyc79 commented Sep 19, 2026 •

edited

Loading