mirror of
https://github.com/fiddyschmitt/File-Tunnel.git
synced 2026-10-21 13:52:05 +00:00
4f105ba913
The test machines are now defined as code in provisioning/ and built from public images: settings.yml (the one file to edit), a Vagrantfile for the VirtualBox VMs, an Ansible playbook for the Linux VMs, a PowerShell provisioner for the Windows VMs, and scripts that set up the Mac and its two Android emulators over SSH. Everything is defined on this PC and pushed to the machines, and any machine can be rebuilt identically. provisioning/README.md covers how to use it and why it's built this way. The tests read every machine's role, address and login from settings.yml (ft_tests/Lab.cs, checked by LabTests) instead of hard-coded addresses and user-secrets. Only the Dropbox keys stay in user-secrets. What replaced what: - ft_test_env's menus: vagrant up / halt / reload / destroy, .\mac.ps1 up / halt and .\check.ps1. The Linux VMs persist instead of resetting at every boot, and every VM runs a smoke test after each up and provision. - setup_debian.sh, the cloud-init seeds and mounts.sh: ansible/. The shares are mounted at boot and again whenever the tests start. - the hand-built Windows gold image, its re-addressed clones and the hand-built server: the public gusztavvargadr/windows-11 box and windows/provision.ps1. Every VM made from the box has its own machine SID, which SMB and RDP between the VMs need. - mac_android_setup.sh, MacEmulator.cs and the tests' own setup of the Mac's SMB share: mac/setup.sh, mac/emulators.sh and mac.ps1. - build_host_setup.sh and build-bionic.sh: the build VM's role and build-android.ps1, which build this working tree rather than GitHub's. - the Linux VMs' SMB mount of this PC's r share, for ft's logs: a VirtualBox shared folder, so the VMs need no login to this PC. - Windows client2 moves from .85, which another lab on this PC uses, to .88. Found while bringing it up: - The Windows box logs on automatically only twice, then switches autologon off. The provisioner keeps it on, so RunRemote always runs. - Android 14's emulator kernel (6.1) lacks the 2023 Linux fix "tcp: tsq: relax tcp_small_queue_check() when rtx queue contains a single skb". After any retransmission timeout, a connection out of the emulator sent one segment per ACK for the rest of its life, which now and then stalled the Android sshfs rows. The emulators now run Android 16 (kernel 6.6). It lands an adb-pushed tree without execute bits, so the tests chmod the sshfs toolchain after pushing it. - Both emulators' eth0 have QEMU's default MAC and answered ARP for their Wi-Fi address too, so the LAN sometimes sent one emulator's traffic to the other. .\mac.ps1 up sets arp_ignore and arp_announce. - The lab runs its own pinned emulator (37.2.12, with -feature -WiFiPacketStream so Wi-Fi uses the bridge) and never updates the Mac SDK's, which other emulators run from. Nor does it restart the shared adb server. A full run on this lab (Debian 6.12.111, Android 16) passed 286 of 287 tests. The one failure, S3 Mac-Linux-Mac, passed 3 of 3 when run again. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01SjajJ7d3BhSZGfsMrGRS5S |
||
|---|---|---|
| .. | ||
| emulators.sh | ||
| setup.sh | ||
| stage-termux-sshfs.py | ||