The below indicates that whatever is taking the extra time is not strongly tied to processor performance. www/gitlab@ee and www/gitlab@ce turn out to be good for seeing larger scale elapsed time multiplication factors. That allows seeing this.
I'll note that when, for example, www/gitlab@ee is built with no other builders active, the large multiplication factor has not happened in my experiments. Based on pkg 2.1.0 use on official package builder machines (so longer than normal times but noting pre-2.1.0 example times as well): ....... @ee @ce @ee @ce beefy13: 01:54:17 01:26:19 (pre: 00:07:36 00:06:24) beefy14: 02:01:45 01:17:36 (pre: 00:07:59 00:06:48) beefy15: ??:??:?? 02:00:48 beefy15: 02:28:31 02:18:46 (pre: 00:09:23 00:07:43) (yes: 2nd 2.1.0 run on 15) beefy16: 02:35:53 03:00:42 (pre: 00:08:49 00:06:55) beefy17: 05:35:01 03:32:05 (pre: 00:09:40 00:07:24) beefy18: 04:18:07 02:33:34 (pre: 00:08:01 00:06:48) beefy19: 01:02:46 00:38:10 (pre: 00:08:21 00:05:53) New, much faster machines: beefy20: ??:??:?? ??:??:?? (pre: 00:04:52 00:04:19) beefy21: 01:56:01 01:31:49 beefy21: 01:39:27 00:38:49 beefy21: 02:02:59 01:50:03 (pre: 00:08:27 00:06:43) (yes: 3rd 2.1.0 run on 21) beefy22: 02:15:04 02:14:47 (pre: 00:06:47 00:05:38) aarch64, slower machines (Also: no armv7 examples): ampere1: ??:??:?? ??:??:?? (pre: 00:13:38 00:12:24) ampere2: ??:??:?? ??:??:?? (pre: 00:13:47 00:16:49) ampere3: ??:??:?? 03:34:43 (pre: 00:23:32 00:09:41) Some beefy21 examples are rather similar to some for beefy13 and beefy19 examples. But there is also wide variability in how large the large timings are, even for more more similar machines --or so I would guess since I do not have information about the machines beyond beefy20 and beefy21 and beefy22 being rather new and much faster. That also points to something not tracking processor performance or the like. It also does not track the number of cpus: beefy20, beefy21, and beefy22 having far more. Near live-lock conditions or contention issues? Something else? I do not know which official build machines run debug kernels vs. non-debug ones (15000??). Any idea how someone with access could find out what the issue is during official builds? === Mark Millard marklmi at yahoo.com
