From cb3b9b92b1464bd1e4c17a2ca3b78f8dbc9b8b1f Mon Sep 17 00:00:00 2001
From: Michael Prokop <mika@grml.org>
Date: Sat, 17 Aug 2024 14:14:08 +0200
Subject: [PATCH] tests: Try to capture screenshot via VNC if
 serial-console-connection fails

While debugging the failing GitHub Actions (see #278), we noticed that
it might serve useful to have a screenshot of the booted system
available. We can capture such a screenshot via vncsnapshot.

NOTE: we move the files inside `results` only if the directory as such
exists in $PWD, otherwise we'd fail the build when being executed as
`tests/build-vm-and-test.sh test` from inside the grml-debootstrap.git.
This happens when manually reproducing the GH tests, though then
`build-vm-and-test.sh` is supposed to be executed from the parent
directory of the `tests` directory (being the grml-debootstrap.git
checkout).
---
 tests/build-vm-and-test.sh      |  2 ++
 tests/serial-console-connection | 25 ++++++++++++++++++++++++-
 tests/test-vm.sh                | 13 +++++++++++--
 3 files changed, 37 insertions(+), 3 deletions(-)

diff --git a/tests/build-vm-and-test.sh b/tests/build-vm-and-test.sh
index 1fb604d..c88c6c5 100755
--- a/tests/build-vm-and-test.sh
+++ b/tests/build-vm-and-test.sh
@@ -34,6 +34,8 @@ fi
 if [ "$1" == "setup" ]; then
   sudo apt-get update
   sudo apt-get -qq -y install curl qemu-system-x86 kpartx python3-pexpect python3-serial
+  # vncsnapshot might not be available, though we don't want to abort execution then
+  sudo apt-get -qq -y install vncsnapshot || true
   [ -x ./tests/goss ] || curl -fsSL https://goss.rocks/install | GOSS_DST="$(pwd)/tests" sh
   # TODO: docker.io
   exit 0
diff --git a/tests/serial-console-connection b/tests/serial-console-connection
index 1aaf5d4..f2ffbc1 100755
--- a/tests/serial-console-connection
+++ b/tests/serial-console-connection
@@ -1,8 +1,11 @@
 #!/usr/bin/env python3
 import argparse
 import multiprocessing
+import os
 import pexpect
 import serial
+import shutil
+import subprocess
 import sys
 import time
 from pexpect import fdpexpect, TIMEOUT
@@ -19,6 +22,8 @@ parser.add_argument('--user', default="root",
                     help='user name to use for login (default: root)')
 parser.add_argument('--password', default="grml",
                     help='password for login (default: grml)')
+parser.add_argument('--screenshot', default="screenshot.jpg",
+                    help='file name for screenshot captured via VNC on error (default: screenshot.jpg)')
 parser.add_argument('--tries', default="12", type=int,
                     help='Number of retries for finding the login prompt')
 parser.add_argument('--timeout', default="300", type=int,
@@ -64,6 +69,21 @@ def login(ser, hostname, user, password, timeout=5):
     child.expect("%s@%s" % (user, hostname), timeout=timeout)
 
 
+def capture_vnc_screenshot(screenshot_file):
+    if not shutil.which('vncsnapshot'):
+        print("WARN: vncsnapshot not available, skipping vnc snapshot capturing.")
+        return(0)
+
+    print("Trying to capture screenshot via vncsnapshot")
+
+    proc = subprocess.Popen(["vncsnapshot", "localhost", screenshot_file])
+    proc.wait()
+    if proc.returncode != 0:
+        print("WARN: failed to capture vnc snapshot :(")
+    else:
+        print("Screenshot file '%s' available" % os.path.abspath(screenshot_file))
+
+
 def main_login(args, ser, hostname, user, password, port, commands):
     success = False
     for i in range(args.tries):
@@ -102,6 +122,7 @@ def main():
     user = args.user
     commands = args.command
     timeout = args.timeout
+    screenshot_file = args.screenshot
 
     ser = serial.Serial(port, 115200)
     ser.flushInput()
@@ -112,8 +133,10 @@ def main():
     process.join(timeout=timeout)
 
     if process.is_alive():
-        print("Reached timeout of %s seconds, aborting." % timeout)
+        print("Reached timeout of %s seconds" % timeout)
+        capture_vnc_screenshot(screenshot_file)
         process.terminate()
+        print("Giving up serial-console-connection execution now")
         sys.exit(1)
 
 
diff --git a/tests/test-vm.sh b/tests/test-vm.sh
index d838801..e59a1d9 100755
--- a/tests/test-vm.sh
+++ b/tests/test-vm.sh
@@ -114,13 +114,15 @@ if [ "$success" = "0" ] ; then
   exit 1
 fi
 
+RC=0
 "$TEST_PWD"/tests/serial-console-connection \
+  --screenshot "$TEST_PWD/tests/screenshot.jpg" \
   --timeout 300 \
   --tries 60 \
   --port "$serial_port" \
   --hostname "$VM_HOSTNAME" \
   --poweroff \
-  "mount -t 9p -o trans=virtio,version=9p2000.L,rw $MOUNT_TAG /mnt && cd /mnt && ./testrunner"
+  "mount -t 9p -o trans=virtio,version=9p2000.L,rw $MOUNT_TAG /mnt && cd /mnt && ./testrunner" || RC=$?
 
 if [ ! -d results ] || [ ! -f ./results/goss.tap ] || [ ! -f ./results/goss.exitcode ]; then
   echo "Running tests inside VM failed for unknown reason" >&2
@@ -134,7 +136,14 @@ fi
 
 echo "Finished serial console connection [timeout=${timeout}]."
 
-mv results/* "$TESTS_RESULTSDIR/"
+# in case of errors we might have captured a screenshot via VNC
+if [ -r "${TEST_PWD}"/tests/screenshot.jpg ] ; then
+  cp "${TEST_PWD}"/tests/screenshot.jpg "${TESTS_RESULTSDIR}"
+fi
+
+if [ -d results ] ; then
+  mv results/* "$TESTS_RESULTSDIR/"
+fi
 
 bailout $RC