| commit | d9a9180a4f51af7990f69576e66ab9076a88d04b | [log] [tgz] |
|---|---|---|
| author | Alan Stokes <alanstokes@google.com> | Fri Sep 24 12:54:54 2021 +0000 |
| committer | Alan Stokes <alanstokes@google.com> | Thu Oct 07 16:37:58 2021 +0100 |
| tree | a489ecdb8ffda3af28f0ed61051da3389dba6c11 | |
| parent | 0b32546a481c370639a22c986c2a1a0a296b7835 [diff] |
Test CompOS more directly. Rather than using "vm --daemonize ..." to test, which is a slightly unrealistic environment and doesn't exercise much of our client-side code, use composd where we can. A side-effect of this is that we no longer need to disable selinux for the tests. This is a partial revert of commit 6aff4f827839325647a43d3a884e6e3d97075d54, and thus a partial re-submit of commit a00d792a8f7f4bca3d4deae34abbbc0dcca02f2b, changing only ComposTestCase. This doesn't address the remanining changes to ComposKeyTestCase, fixed, since they are higher risk and less urgent. More significantly I've restricted the check for exact binary match to vdex files only. The other artifacts currently differ because the fd numbers used in the two runs are baked into them and differ. Additionally I've tried to handle adb disconnection. In the old tests the VM kept running until after we did our checks when we killed it. In the new tests it gets killed by the service manager at some indeterminate time after composd_cmd exits, and when it does the adb connection drops, causing various weird failures. To try to avoid that I'm deliberately killing it when we're done with it, trying to guard against it exiting spontaneously in the meantime by doing the ADB reconnection incantation slightly obsessively. Bug: 200924405 Test: atest ComposTestCase (locally, on acloud, in presubmit) Change-Id: Ia56ce4b7305d14d6b759ede0d612eb3095559576
This repository contains userspace services related to running virtual machines on Android, especially protected virtual machines. See the getting started documentation and Microdroid README for more information.