Comment 2 for bug 1836253

I managed to reproduce the error in my environment (on the master branch).

The config needed in devstack:

/etc/neutron/metadata_agent.ini
[cache]
enabled = True
backend = dogpile.cache.memory

I simplified the script a bit:

#!/bin/bash

cat << EOF > /tmp/metadata
#!/bin/sh
if curl http://169.254.169.254/openstack
then
    echo "XXX metadata OK"
else
    echo "XXX metadata ERROR"
fi
EOF

openstack server delete vm0 2>/dev/null
openstack server create \
    --flavor cirros256 \
    --image cirros-0.4.0-x86_64-disk \
    --nic net-id=private,v4-fixed-ip=10.0.0.41 \
    --user-data /tmp/metadata \
    --wait \
    vm0 \
    >/dev/null
openstack server show vm0 -f value -c addresses

sleep 4

openstack console log show vm0 | grep "XXX metadata"

-- end of script --

10.0.0.41 is an arbitrary fixed-ip I selected from the range of the private network. The first run finds "XXX metadata OK". Starting rom the second run (that is re-using the same IP) the following errors can be seen instead in the console log:

checking http://169.254.169.254/2009-04-04/instance-id
failed 1/20: up 3.26. request failed
failed 2/20: up 5.32. request failed
failed 3/20: up 7.45. request failed
failed 4/20: up 9.52. request failed
...