mgr: use ipv4 default when ipv6 was disabled

[kungf]

Fixes: https://tracker.ceph.com/issues/40023
Signed-off-by: kungf yang.wang@easystack.cn

• References tracker ticket
• Updates documentation if necessary
• Includes tests for new functionality or reproducer for bug

[votdev]

I do not like the usage of the ip command, isn't it possible to check that using a different solution, e.g. by using the procfs information, e.g. like this code.

[votdev]

I vote to use a different solution, e.g. by using procfs.

[kungf]

hi, @votdev what's the advantage of procfs?

[sebastian-philipp]

I'm agreeing with Volker here that calling an external program is not my preferred solution.

[sebastian-philipp]

Also this is the wrong place for this if statement, because it's part of the dashboard configuration. Instead it should be located somewhere in _configure()

[wjwithagen]

A Python "universal" solution would be preferable here.
Otherwise part of it will degrade into a `if (os_platform == .....) ... else .... " version

[sebastian-philipp]

Also this is the wrong place for this if statement, because it's part of the dashboard configuration. Instead it should be located somewhere in _configure()

[sebastian-philipp]

Relates to #26635

[kungf]

#26635 may need be reverted, wait_for_occupied_port check is right to check the port.

[jan--f]

Additionally to what other said: Imho this hsould be part of the MgrModule implementation instead of adding this code to every module extending MgrModule

[kungf]

hi, @votdev @sebastian-philipp @jan--f thanks for your reviews, code modified, please review again

[votdev]

Maybe something like this will work? Have not tested how it behaves on a IPv4 only system.

def is_ipv6_enabled():
    import socket
    try:
        socket.AF_INET6
        socket.IPPROTO_IPV6
        socket.IPV6_V6ONLY
        return True
    except AttributeError:
        return False

[kungf]

def is_ipv6_enabled():
    import socket
    try:
        socket.AF_INET6
        socket.IPPROTO_IPV6
        socket.IPV6_V6ONLY
        return True
    except AttributeError:
        return False

@votdev not work too

comment addressed

[kungf]

Hi, reviewers, now we have two function to check whether ipv6 enabled, what's your proposition，socket or netifaces, i think netifaces is more simple，no need to create a socket, but it requires new dependence

def is_ipv6_enabled():                                                                                                                                                                                             
    """ Returns True if the system can bind an IPv6 address. """                 
    sock = None                                                                  
    has_ipv6 = False                                                             
                                                                                 
    if socket.has_ipv6:                                                          
        try:                                                                     
            sock = socket.socket(socket.AF_INET6)                                    
            sock.bind(("::1", 0))                                                    
            has_ipv6 = True                                                          
        except Exception:                                                            
            pass                                                                     
                                                                                     
    if sock:                                                                         
        sock.close()                                                                 
    return has_ipv6

def is_ipv6_enabled():                                                           
    """                                                                          
    Check whether IPv6 is enabled.                                               
    :return: Return True if IPv6 is enabled, otherwise False.                    
    :rtype: bool                                                                 
    """                                                                              
    interfaces = netifaces.interfaces()                                          
    for i in interfaces:                                                         
        if netifaces.ifaddresses(i).has_key(netifaces.AF_INET6):                 
            return True;                                                         
    return False

[sebastian-philipp]

I'm ok with both of them. ( @smithfarm is netifaces already a transitive dependency? )

stale

[tchaikov]

i am in favor of the first solution. and the reason is that it has less dependencies. and we can improve it a little bit:

import contextlib
# ...
def is_ipv6_enabled():
    try:
        sock = socket.socket(socket.AF_INET6)
        with contextlib.closing(sock):
            sock.bind(("::1", 0))
            return True
    except (AttributeError, socket.error) as e:
           return False

changelog:

• the function is self-explained. no need extra doc string.
• their is chance that the constant of socket.AF_INET6 is not defined per python document[0], so AttributeError is caught as well
• use context manager to simplify the exception handling

---

[0]

These constants represent the address (and protocol) families, used for the first argument to socket(). If the AF_UNIX constant is not defined then this protocol is unsupported.

[smithfarm]

is netifaces already a transitive dependency?

I grepped the Ceph codebase for netifaces and came up with nothing, so I would be careful with it...

UPDATE: On a single-node nautilus cluster installed from RPMs, there is no netifaces (python-netifaces, python3-netifaces) at all, so I'd say the short answer is: "no".

[kungf]

hi, reviewers, according your advises, ipv6 check with socket was commit, please review, thanks

[kungf]

hi reviewers need review again,ths!

[votdev]

jenkins retest this please

[tchaikov]

jenkins test dashboard

[tchaikov]

• http://pulpito.ceph.com/kchai-2019-06-02_15:00:54-rados-wip-kefu-testing-2019-06-01-2346-distro-basic-mira/
• http://pulpito.ceph.com/nojha-2019-06-04_22:13:54-rados-wip-kefu-testing-2019-06-01-2346-distro-basic-smithi/

the valgrind failure is a known issue tracked by http://tracker.ceph.com/issues/38403, and it should be addressed by #28397

[tchaikov]

@jan--f and @votdev i think the author has address your comments in his latest change, could you please take another look?

comments addressed