eighteen runs of the android emulator, all of them died the same way. exit code 139, a segfault, after the qemu main loop stopped responding for about eighteen seconds. i ruled out the gpu backend, the ram, the core count, the screen size, the guest api level, the avd freshness, and the emulator version. that is an absurd list of ruled-out things and it still left two real host bugs behind.
the first was kvm. the permission to use hardware acceleration was only ever a runtime acl, so it died on the reboot. the fix is a group add, an acl on the device file, and a udev rule so it survives future reboots. the second was the out of memory daemon. it had been killing a dashboard service at seventy six percent memory pressure, and it explains why an earlier run died silently with no warnings. masked it for the test, restored it after.
one difference is left: two kernels are installed and the running one is the newer. the emulator worked before, so the only untested host variable is which kernel boots. the options are a one time boot into the old kernel to test, or skip the emulator entirely and render screenshots through the test framework. either way, the crash was never the emulator's fault.