hawq-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Konstantin Boudnik <...@apache.org>
Subject Re: ssh'ing around the cluster
Date Sun, 28 Feb 2016 03:40:47 GMT
On Sat, Feb 27, 2016 at 07:38PM, Ivan Novick wrote:
> Can we change Greenplum path file to say hawq path?  That's super confusing IMO

Yeah, I've opened a JIRA for it. There's more, really, like GPHOME, etc.

It's all here HAWQ-421

Cos

> 
> Cheers,
> Ivan
> 
> Sent from my iPhone
> 
> > On Feb 27, 2016, at 6:53 PM, Radar Da lei <rlei@pivotal.io> wrote:
> >
> > Hi Konstantin,
> >
> > Never see this error before.
> >
> > 1. Did you do 'source greenplum_path.sh' before init HAWQ?
> > 2. HAWQ does not support 'root' user, so no need to use sudo.
> > 3. Did you set up SSH password-less on the machine?
> >
> > Thanks.
> >
> > Regards,
> > Radar
> >
> >> On Sun, Feb 28, 2016 at 6:44 AM, Konstantin Boudnik <cos@apache.org> wrote:
> >>
> >>> On Thu, Feb 25, 2016 at 09:03PM, Konstantin Boudnik wrote:
> >>>> On Fri, Feb 26, 2016 at 10:53AM, Lei Chang wrote:
> >>>> Hi Konstantin,
> >>>>
> >>>> If I understand correctly, what you requested is in current hawq code
> >> base.
> >>>> To init hawq, "hawq init cluster" is one way, you can also use "hawq
> >> init
> >>>> master" and "hawq init segment" on all cluster nodes. Master and
> >> segments
> >>>> are decoupled in 2.0.
> >>>
> >>> Ah, nice. Let me try that. Thanks!
> >>
> >> Well, unfortunately, it doesn't work as expected. Even as I do
> >>    $ sudo -u hawq /usr/bin/hawq init master
> >>
> >> I see that the process is always going to local_ssh call, which ends up in
> >> an attempt to ssh into localhost.
> >> Traceback (most recent call last):
> >>  File "/usr/lib64/python2.7/runpy.py", line 162, in _run_module_as_main
> >>    "__main__", fname, loader, pkg_name)
> >>  File "/usr/lib64/python2.7/runpy.py", line 72, in _run_code
> >>    exec code in run_globals
> >>  File "/usr/lib64/python2.7/trace.py", line 819, in <module>
> >>    main()
> >>  File "/usr/lib64/python2.7/trace.py", line 807, in main
> >>    t.runctx(code, globs, globs)
> >>  File "/usr/lib64/python2.7/trace.py", line 513, in runctx
> >>    exec cmd in globals, locals
> >>  File "/usr/lib/hawq/bin/hawq", line 133, in <module>
> >>    main()
> >>  File "/usr/lib/hawq/bin/hawq", line 80, in main
> >>    result = local_ssh(cmd)
> >>  File "/usr/lib/hawq/bin/hawq", line 35, in local_ssh
> >>    result = subprocess.Popen(cmd, shell=True).wait()
> >>  File "/usr/lib64/python2.7/subprocess.py", line 1376, in wait
> >>    pid, sts = _eintr_retry_call(os.waitpid, self.pid, 0)
> >>  File "/usr/lib64/python2.7/subprocess.py", line 478, in _eintr_retry_call
> >>    return func(*args)
> >>
> >> Any ideas what I am doing wrong? Thanks
> >>  Cos
> >>
> >>>> On Fri, Feb 26, 2016 at 9:40 AM, Konstantin Boudnik <cos@apache.org>
> >> wrote:
> >>>>
> >>>>> Guys,
> >>>>>
> >>>>> more revelations about the way Hawq is designed to work with the
> >> service
> >>>>> bootstraps, config management, and so on. Looking into hawq_ctl
and
> >>>>> observing
> >>>>> the behavior of 'hawq init cluster' I see that a number of
> >> operations is
> >>>>> intended to be initiated from presumably, a master node, and carried
> >> on all
> >>>>> the nodes of a hawq cluster via the means or ssh (or rsync). While
> >> doing so
> >>>>> might be a convenient shortcut for the development environment,
it
> >> isn't as
> >>>>> much in a real deployment. For one, it requires password-less SSH
> >> access
> >>>>> between nodes, which isn't (generally speaking) how data centers
> >> might
> >>>>> operate.
> >>>>>
> >>>>> Perhaps a better way of separating the concerns here is to have
> >> isolated
> >>>>> functions to perform only local-node operations, and a wrapper to
> >> run the
> >>>>> same
> >>>>> functionality on the all remote nodes, via ssh or else. If such
> >> split is
> >>>>> done,
> >>>>> then an orchestration mechanism (such as a state machine similar
to
> >> Puppet
> >>>>> or
> >>>>> Chef), would execute the scripts on separate nodes in full
> >> isolation. And
> >>>>> if
> >>>>> so desired in a dev environment, the current functionality would
be
> >>>>> available
> >>>>> as well.
> >>>>>
> >>>>> Thoughts? Regards,
> >>>>>  Cos
> >>
> >>
> >>

Mime
View raw message