Install a verified Ookla CLI for adaptive SQM - #1093
Open
Jason-Morcos wants to merge 5 commits into
Open
Conversation
Jason-Morcos
force-pushed
the
agent/fix-sqm-speedtest-bootstrap
branch
from
August 3, 2026 04:11
438c172 to
6f24e66
Compare
Collaborator
|
Couple things need fixing here, and then I can work this in for sure. The Ookla CLI install shouldn't fail the whole install script, we want it to log and continue as it does today with the apt install attempt. The other thing, let's centralize the hashes here:
|
… the configuration to plan now
…scrub Dragging the playhead showed one set of figures a second after settling, then another two seconds later. The card debounces seeks, but an answer already in flight for an earlier position was rendered when it landed, then noticed the playhead had moved and asked again. A click seeks once, so it never showed it. - The card cancels the in-flight answer when the playhead (or the selected WANs, or their rate at the playhead) changes, and discards an answer that completes after cancellation instead of rendering it. - The page's historic snapshot carries a sequence number, so a slow older seek cannot overwrite a newer one. - The WAN line reads the rate the tiles carry rather than the answer's echo of it, and a snapshot landing after the card asked re-answers the split, which is allocated against that rate at the playhead.
The lag after a seek was not the debounce and not InfluxDB (every historic query there answers in 10-40 ms). It was three things stacked: - At the playhead the card asked the console for two DPI reports (the quarter hour and the 24 h exclusion window) before looking at the measured WAN, and the console gate spaces those calls two seconds apart. On a site the gateway agent measures the split never reads them. Measured first now, and a covered site skips the DPI, the history, and first-seen entirely. An uncovered site fetches all three as before. - The map's historic fetch asked device_health for every node with a MAC, one at a time, and clients write none: around a hundred empty round trips per seek. Infrastructure nodes only now. - The page renders on its own 3 s tick before a seek's snapshot lands, so the card asked with the previous instant's WAN rate, then cancelled and asked again when the snapshot arrived. The page now stamps the instant its rate describes and the card waits for it to match the playhead before asking. Debounce cut from 400 to 150 ms; the map already spaces seeks 500 ms apart. The service's catch blocks swallowed a cancelled query as a per-stage miss and the request ran on to a degraded answer nobody would see; they let our own cancellation through now. A playhead answer logs its stage timings at Debug.
Static/bench-verified on the UCG-Fiber, the first 6.0.x image for that line: qca-ssdk.ko .text byte-identical to the live-verified 6.0.5 SSDK, every other common kernel module .text-identical, vermagic unchanged, all boot-tweak userland present including MongoDB (unifi-perf-tweaks docs/compat-6.0.7.md). The UXG ceiling stays at 6.0.5.
Jason-Morcos
force-pushed
the
agent/fix-sqm-speedtest-bootstrap
branch
from
September 7, 2026 18:06
6f24e66 to
6f400e1
Compare
Contributor
Author
|
Addressed both requests in 6f400e1:
I also refreshed the change onto current dev. The full Release solution builds, all 272 SQM tests pass, the focused Web deployment-status test passes, and generated Bash syntax is covered. GitHub did not schedule a new check because the current workflow filters pull-request events to main while this PR targets dev; the PR description records that limitation and the bounded macOS Web-suite result. |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Problem
Adaptive SQM currently pipes the packagecloud repository bootstrap into a shell, removes any existing
speedtestpackage, and installs an unpinned candidate withapt. That can fail when UniFi OS changes its Debian base, mutates a vendor-managed package surface, and makes the calibration binary drift independently of Network Optimizer.Fix
/data/network-optimizer/bin1.2.0.84 (ea6b6773cf)build selected by the existing packagecloud install pathjqseparately alongside the existingbcdependency statusValidation
dev: succeededgit diff --check: passedGitHub did not schedule a new Actions run for this head because the upstream workflow currently filters pull-request events to
main, while this pull request targetsdev. The unfiltered Web suite is not claimed as locally green on macOS: two existing Vodafone AES-CCM tests are unsupported on this host, and the remaining run exceeded the bounded local test window.No gateway package, SQM script, cron entry, service, or live installation was changed while validating this fix.