A USG exported its user count as mobile and its guest count as handheld in unpoller_device_stations. I fixed it in PR #1114, merged within two hours of opening it. It follows my station-count fix and my temperature and storage fix in the same exporter.

When it happens

Every adopted USG did this. The ugw.json example in the unifi module has 41 user, 0 guest, 7 desktop, 2 mobile and 8 handheld stations. The USG exported (other labels left out):

unpoller_device_stations{station_type="mobile"} 41
unpoller_device_stations{station_type="handheld"} 0

It should export 2 and 8. The other three series were right. The Mobile and Handheld panels of the USG Insights dashboard query these two series, so they showed the user and guest counts.

How the counts are passed

exportSTAcount takes five counts in the order user, guest, desktop, mobile, handheld. They are all the same type, so any five values compile. exportUSG passed d.UserNumSta and d.GuestNumSta again for the last two. The USG struct has NumMobile and NumHandheld, and UXG and UDM pass them.

Why the gap existed

Commit 4d48576 (December 2019, first released in v1.6.2) moved five inline entries into exportSTAcount. Before it, the USG exported d.NumMobile and d.NumHandheld. No test checked a unpoller_device_stations series until my station-count fix added one. It checks labels, not values, and calls exportSTAcount directly, so it never saw what exportUSG passes.

The fix

Pass d.NumMobile and d.NumHandheld. The change is one line in pkg/promunifi/usg.go, shown without its context lines:

@@ -112 +112 @@ func (u *promUnifi) exportUSG(r report, d *unifi.USG) {
-		u.exportSTAcount(r, labels, d.UserNumSta, d.GuestNumSta, d.NumDesktop, d.UserNumSta, d.GuestNumSta)
+		u.exportSTAcount(r, labels, d.UserNumSta, d.GuestNumSta, d.NumDesktop, d.NumMobile, d.NumHandheld)

The test

TestExportUSGStationCountsUseTheirOwnFields exports a USG with 41 user, 1 guest, 7 desktop, 2 mobile and 8 handheld stations and checks the value of each station_type.

Before the fix it failed: mobile was 41 and handheld was 1, where it expected 2 and 8. After the fix it passes, and so does the package. I also reverted each of the two arguments alone. Each revert failed the test.

What the maintainer did

@platinummonkey approved and merged it. It was my third unpoller PR.

The lesson

  • A helper that takes several values of one type compiles with any order and any repeat. Check each call against the parameter order, or pass the values as named fields.
  • When you move inline code into a helper, compare the new arguments with the old lines. Here the old lines were right.
  • Give every input in a test a different value. A fixture with equal counts cannot tell which field was read.

If a helper with many same-type arguments sits in your own exporter, reply on X.