virtualbmc binds only to localhost even though each ipmi devices were configured with different ip addresses

[David Hill]

virtualbmc binds only to one IP address even though each ipmi devices were configured with different ip addresses

[root@zappa /]# vbmc list
+------------------+---------+----------------+------+
| Domain name | Status | Address | Port |
+------------------+---------+----------------+------+
| ceph-0-rhosp8 | down | 192.168.122.14 | 623 |
| ceph-1-rhosp8 | down | 192.168.122.15 | 623 |
| ceph-2-rhosp8 | down | 192.168.122.16 | 623 |
| compute-0-rhosp8 | down | 192.168.122.13 | 623 |
| control-0-rhosp8 | running | 192.168.122.10 | 623 |
| control-1-rhosp8 | down | 192.168.122.11 | 623 |
| control-2-rhosp8 | down | 192.168.122.12 | 623 |
+------------------+---------+----------------+------+

This new feature lacks documentation and pxe_ssh is being deprecated too early imho.

[David Hill (david-hill-ubisoft)]

[root@zappa virtualbmc]# 2016-11-26 00:48:04,140.140 19910 ERROR VirtualBMC [-] Error starting a Virtual BMC for domain ceph-0-rhosp8. Error: [Errno 98] Address already in use
Error starting a Virtual BMC for domain ceph-0-rhosp8. Error: [Errno 98] Address already in use

[David Hill (david-hill-ubisoft)]

The problem here is that pyghmi is using hardcoded values on localhost and port 623.

From vbmc.py:
[...]
import pyghmi.ipmi.bmc as bmc
[...]
    def __init__(self, username, password, port, address, domain_name, libvirt_uri, libvirt_sasl_username=None, libvirt_sasl_password=None):
        super(VirtualBMC, self).__init__({username: password}, port=port, address=address)

From pyghmi.ipmi.bmc:
[...]
class Bmc(serversession.IpmiServer):
[...]

From serversession/IpmiServer:
class IpmiServer(object)
    def __init__(self, authdata, port=623, bmcuuid=None, address='::'):

[David Hill (david-hill-ubisoft)]

[root@zappa pyghmi]# netstat -nap | grep python
udp6 0 0 :::623 :::* 22980/python

[David Hill (david-hill-ubisoft)]

If add server=None in this pyghmi function, I can then successfully create many virtualbmc instances.

    def _assignsocket(cls, server=None):
        global iothread
        global iothreadready
        global iosockets
        global ipv6support
        global myself
        server=None <===

root 28860 0.0 0.0 454308 18248 ? Sl 20:31 0:00 /usr/bin/python /usr/bin/vbmc start ceph-0-rhosp8
root 29333 0.0 0.0 454440 18432 ? Sl 20:36 0:00 /usr/bin/python /usr/bin/vbmc start ceph-0-rhosp8
root 29342 0.0 0.0 454308 18428 ? Sl 20:36 0:00 /usr/bin/python /usr/bin/vbmc start ceph-1-rhosp8
root 29393 0.0 0.0 454440 18364 ? Sl 20:36 0:00 /usr/bin/python /usr/bin/vbmc start ceph-2-rhosp8

The problem I now face is that I can successfully start many virtualbmc instances per IP address which is not really wanted.

[David Hill (david-hill-ubisoft)]

But with this patch, the process is still not binding on the specified IP... So my guess here is I don't understand python.

[Lucas Alvares Gomes (lucasagomes)]

Here's my first attempt to fix it in pyghmi: https://review.openstack.org/404837

[Lucas Alvares Gomes (lucasagomes)]

Confirmed with David that https://review.openstack.org/404837 has fixed it.