Ok, I made the following test today: Built this package 1000 times, on 20 different autobuilders.
It built ok this many times in these autobuilders: 67 uranio2 58 uranio1 53 e5 53 e12 48 leaseweb2 48 leaseweb1 47 e16 35 e14 34 e4 34 e17 34 e10 32 e11 30 e20 28 e18 and it failed to build this many times in these autobuilders: 65 e8 65 e6 64 e3 61 e7 52 yoda2 50 yoda1 50 skywalker1 2 uranio2 Now the fun part: The machines on which it fails are the "fast" ones, and the machines on which it builds ok are the "slow" ones. I have good estimations for the CPU speed for these machines. This is the "slowness" of the machines on which it fails to build: e6 0.413 e8 0.413 e3 0.417 e7 0.417 yoda1 0.418 yoda2 0.429 skywalker1 0.446 uranio2 0.540 and this is the "slowness" of the machines on which it builds ok: e12 0.526 e5 0.540 uranio2 0.540 e16 0.646 uranio1 0.734 e14 1.528 e17 1.537 e10 1.545 e11 1.558 e4 1.573 e18 1.977 e20 2.074 The "slowness" is the inverse of the speed. My unit of measure (i.e. slowness 1) is the speed of my i3-3217U @ 1.80GHz at home, which was my first autobuilder. Can you host a virtual machine somewhere which is at least twice as fast as my i3 at home? I guess, but I'm not sure, that the problem could be reproduced there as well. [ Of course, if we can confirm that speed is the issue here, this should be "serious" again, as packages should never fail to build just because the CPU is "too fast" ]. Thanks.