nodepool/tools/functional-test-check.sh

179 lines
6.3 KiB
Bash
Executable File

#!/bin/bash -ex
LOGDIR=/home/zuul/zuul-output/logs
# Set to indiciate an error return
RETURN=0
FAILURE_REASON=""
if [[ ${NODEPOOL_FUNCTIONAL_CHECK:-} == "installed" ]]; then
NODEPOOL_INSTALL=${NODEPOOL_INSTALL:-~/.venv}
NODEPOOL_CONFIG=${NODEPOOL_CONFIG:-/etc/nodepool/nodepool.yaml}
NODEPOOL="$NODEPOOL_INSTALL/bin/nodepool -c $NODEPOOL_CONFIG"
elif [[ ${NODEPOOL_FUNCTIONAL_CHECK:-} == "containers" ]]; then
NODEPOOL="docker exec nodepool_nodepool-launcher_1 nodepool"
else
echo "Running in unknown environment!"
exit 1
fi
cat > /tmp/ssh_wrapper <<EOF
#!/bin/bash -ex
sudo -H -u zuul ssh -o StrictHostKeyChecking=no -i $HOME/.ssh/id_nodepool root@\$@
EOF
sudo chmod 0755 /tmp/ssh_wrapper
function sshintonode {
name=$1
state='ready'
node=`$NODEPOOL list | grep $name | grep $state | cut -d '|' -f6 | tr -d ' '`
/tmp/ssh_wrapper $node ls /
# Check that the root partition grew on boot; it should be a 5GiB
# partition minus some space for the boot partition. However
# emperical evidence suggests there is some modulo maths going on,
# (possibly with alignment?) that means we can vary up to even
# 64MiB. Thus we choose an expected value that gives us enough
# slop to avoid false matches, but still indicates we resized up.
root_size=$(/tmp/ssh_wrapper $node -- lsblk -rbno SIZE /dev/vda1)
expected_root_size=$(( 5000000000 ))
if [[ $root_size -lt $expected_root_size ]]; then
echo "*** Root device does not appear to have grown: $root_size"
FAILURE_REASON="Root partition of $name does not appear to have grown: $root_size < $expected_root_size"
RETURN=1
fi
# Check we saw metadata deployed to the config-drive
/tmp/ssh_wrapper $node \
"dd status=none if=/dev/sr0 | tr -cd '[:print:]' | grep -q nodepool_devstack"
if [[ $? -ne 0 ]]; then
echo "*** Failed to find metadata in config-drive"
FAILURE_REASON="Failed to find meta-data in config-drive for $node"
RETURN=1
fi
# Debugging that we're seeing the right node
/tmp/ssh_wrapper $node -- "cat /etc/os-release; blkid"
# This ensures DIB setup the bootloader kernel arguments correctly
# by looking for the default console setup it does. In the past
# we have seen issues with bootloader installation for all sorts
# of reasons (our errors and upstream changes); but generally what
# has happened is that case grub has silently fallen-back to
# "guessing" from the running build-system what kernel flags to
# use.
#
# DIB images should be booting from a root device labeled
# "cloudimg-rootfs". In the gate, where you're *using* a
# DIB-image to build a DIB-image, it's /proc/cmdline contains
# "root=LABEL=cloudimg-rootfs" and a misconfigured grub can
# actually guess the correct root device. However, when this
# builds an image in production on a totally different host you
# get a non-booting, wrong image. Ergo, although this is mostly
# what we're interested in validating, this is not a reliable
# thing to test directly.
#
# So below, we probe for something else; the console setup that
# DIB will put in. If this is missing, it's an indication that
# the bootloader is not setting up the kernel arguments correctly.
kernel_cmd_line=$(/tmp/ssh_wrapper $node -- cat /proc/cmdline)
echo "Kernel command line: ${kernel_cmd_line}"
if [[ ! $kernel_cmd_line =~ 'console=tty0 console=ttyS0,115200' ]]; then
echo "*** Failed to find correct kernel boot flags"
FAILURE_REASON="Failed to find correct kernel boot flags $node"
RETURN=1
fi
# Ensure glean services have loaded
# The output is like:
# ---
# glean-early.service loaded active exited Early glean execution
# glean@ens3.service loaded active exited Glean for interface ens3 with NetworkManager
# glean@lo.service loaded active exited Glean for interface lo with NetworkManager
# ---
# So if we see anything other than 'loaded active exited' we have a problem.
glean_status=$(/tmp/ssh_wrapper $node -- "systemctl | egrep 'glean[@|-]' | { grep -v 'loaded active exited' || true; }")
if [[ ${glean_status} != '' ]]; then
echo "*** Glean not loaded correctly"
echo "*** saw: ${glean_status}"
FAILURE_REASON="Failed to start glean correctly"
RETURN=1
fi
}
function checknm {
name=$1
state='ready'
node=`$NODEPOOL list | grep $name | grep $state | cut -d '|' -f6 | tr -d ' '`
nm_output=$(/tmp/ssh_wrapper $node -- nmcli c)
# virtio device is eth0 on older, ens3 on newer
if [[ ! ${nm_output} =~ (eth0|ens3) ]]; then
echo "*** Failed to find interface in NetworkManager connections"
/tmp/ssh_wrapper $node -- nmcli c
/tmp/ssh_wrapper $node -- nmcli device
FAILURE_REASON="Failed to find interface in NetworkManager connections"
RETURN=1
fi
}
function waitforimage {
local name=$1
local state='ready'
local builds
while ! $NODEPOOL image-list | grep $name | grep $state; do
$NODEPOOL image-list > ${LOGDIR}/nodepool-image-list.txt
$NODEPOOL list --detail > ${LOGDIR}/nodepool-list.txt
builds=$(ls -l /var/log/nodepool/builds/ | grep $name | wc -l)
if [[ ${builds} -ge 4 ]]; then
echo "*** Build of $name failed at least 3 times, aborting"
exit 1
fi
sleep 10
done
}
function waitfornode {
name=$1
state='ready'
while ! $NODEPOOL list | grep $name | grep $state | grep "unlocked"; do
$NODEPOOL image-list > ${LOGDIR}/nodepool-image-list.txt
$NODEPOOL list --detail > ${LOGDIR}/nodepool-list.txt
sleep 10
done
}
# check that image built
waitforimage test-image
# check image was bootable
waitfornode test-image
# check ssh for root user
sshintonode test-image
# networkmanager check
# TODO(jeblair): This should not run in all cases; in fact, most of
# this checking should move into the dib repo
#checknm test-image
# userdata check
set -o errexit
# Show the built nodes
$NODEPOOL list
# Try to delete the nodes that were just built
$NODEPOOL delete --now 0000000000
# show the deleted nodes (and their replacements may be building)
$NODEPOOL list
if [[ -n "${FAILURE_REASON}" ]]; then
echo "${FAILURE_REASON}"
fi
exit $RETURN