yahoo-eng-team team mailing list archive

[Takashi Kajinami <1926693@xxxxxxxxxxxxxxxxxx>]

Public bug reported:

Currently we use socket.getshostname() to determine the heyervisor
hostname assuming that is what is used in libvirt.

However libvirt actually ues the following steps to generate hypervisor
hostname and socket.gethostname() doesn't match in some cases.

/* Who knew getting a hostname could be so delicate.  In Linux (and Unices
 * in general), many things depend on "hostname" returning a value that will
 * resolve one way or another.  In the modern world where networks frequently
 * come and go this is often being hard-coded to resolve to "localhost".  If
 * it *doesn't* resolve to localhost, then we would prefer to have the FQDN.
 * That leads us to 3 possibilities:
 *
 * 1)  gethostname() returns an FQDN (not localhost) - we return the string
 *     as-is, it's all of the information we want
 * 2)  gethostname() returns "localhost" - we return localhost; doing further
 *     work to try to resolve it is pointless
 * 3)  gethostname() returns a shortened hostname - in this case, we want to
 *     try to resolve this to a fully-qualified name.  Therefore we pass it
 *     to getaddrinfo().  There are two possible responses:
 *     a)  getaddrinfo() resolves to a FQDN - return the FQDN
 *     b)  getaddrinfo() fails or resolves to localhost - in this case, the
 *         data we got from gethostname() is actually more useful than what
 *         we got from getaddrinfo().  Return the value from gethostname()
 *         and hope for the best.
 */

https://github.com/libvirt/libvirt/blob/ec2e3336b8c8df572600043976e1ab5feead656e/src/util/virutil.c#L454-L531

Acutally we do see that in TripleO deployment socket.gethostname() and virsh hostname doesn't agree, and this causes an issue when we set up resource_provider_bandwidths
~~~
[heat-admin@compute-0 ~]$ python
Python 3.6.8 (default, Dec  5 2019, 15:45:45) 
[GCC 8.3.1 20191121 (Red Hat 8.3.1-5)] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> import socket
>>> socket.gethostname()
'compute-0'
>>> exit()
[heat-admin@compute-0 ~]$ sudo podman exec -it nova_libvirt virsh hostname
compute-0.redhat.local
~~~

** Affects: neutron
     Importance: Undecided
         Status: New

-- 
You received this bug notification because you are a member of Yahoo!
Engineering Team, which is subscribed to neutron.
https://bugs.launchpad.net/bugs/1926693

Title:
  Logic to obtain hypervisor hostname is not completely compatible with
  libvirt

Status in neutron:
  New

Bug description:
  Currently we use socket.getshostname() to determine the heyervisor
  hostname assuming that is what is used in libvirt.

  However libvirt actually ues the following steps to generate
  hypervisor hostname and socket.gethostname() doesn't match in some
  cases.

  /* Who knew getting a hostname could be so delicate.  In Linux (and Unices
   * in general), many things depend on "hostname" returning a value that will
   * resolve one way or another.  In the modern world where networks frequently
   * come and go this is often being hard-coded to resolve to "localhost".  If
   * it *doesn't* resolve to localhost, then we would prefer to have the FQDN.
   * That leads us to 3 possibilities:
   *
   * 1)  gethostname() returns an FQDN (not localhost) - we return the string
   *     as-is, it's all of the information we want
   * 2)  gethostname() returns "localhost" - we return localhost; doing further
   *     work to try to resolve it is pointless
   * 3)  gethostname() returns a shortened hostname - in this case, we want to
   *     try to resolve this to a fully-qualified name.  Therefore we pass it
   *     to getaddrinfo().  There are two possible responses:
   *     a)  getaddrinfo() resolves to a FQDN - return the FQDN
   *     b)  getaddrinfo() fails or resolves to localhost - in this case, the
   *         data we got from gethostname() is actually more useful than what
   *         we got from getaddrinfo().  Return the value from gethostname()
   *         and hope for the best.
   */

  https://github.com/libvirt/libvirt/blob/ec2e3336b8c8df572600043976e1ab5feead656e/src/util/virutil.c#L454-L531

  Acutally we do see that in TripleO deployment socket.gethostname() and virsh hostname doesn't agree, and this causes an issue when we set up resource_provider_bandwidths
  ~~~
  [heat-admin@compute-0 ~]$ python
  Python 3.6.8 (default, Dec  5 2019, 15:45:45) 
  [GCC 8.3.1 20191121 (Red Hat 8.3.1-5)] on linux
  Type "help", "copyright", "credits" or "license" for more information.
  >>> import socket
  >>> socket.gethostname()
  'compute-0'
  >>> exit()
  [heat-admin@compute-0 ~]$ sudo podman exec -it nova_libvirt virsh hostname
  compute-0.redhat.local
  ~~~

To manage notifications about this bug go to:
https://bugs.launchpad.net/neutron/+bug/1926693/+subscriptions