chess example problem about pachyderm HOT 2 CLOSED

Hi, really sorry you ran in to this. I think I have an idea what's going on here.

tl;dr: I think it's a problem with dns.

Pfs uses etcd to announce its services, this is handled by the announce-master-*-*.service services. Those services do etcdctl set /pfs/master/0-1 {HOST_NAME}:port. The router service then uses this to figure out how to contact the master. However this only works if the router can do a dns lookup. How did you setup the CoreOS instances? I remember running in to this problem when I tried to get pfs setup on Vagrant.

There are a few easy ways I can think of to fix this for you short term:

• echo "coreos-1 10.132.128.22" >/etc/hosts is a quick hack that should make things start working. This would also be a good way to confirm the DNS theory.
• GCE and EC2 both have DNS by default, shoot me an email at [email protected] and I'll setup a hosted instance for you to play around with.

Action items for pfs:

• If DNS is indeed the problem this error message should explicitly mention that as a likely cause.
• Docs should do a better job of describing this in the "getting started" section.
• We should look in to making pfs not depend on DNS. I think this will be doable via flannel or something similar.

from pachyderm.

Hi Joe,

Thanks for the great response! The project does have "Alpha" plastered
firmly all over it so don't stress about newb issues :)

Unfortunately setting up /etc/hosts like so (and the other way round as you
suggested, which I assume was a mistake).

SHUTIT_TMP#GusY7I2m>cat /etc/hosts
10.132.128.22 coreos-1
10.132.129.103 coreos-2

I can only assume coreos has its own way of doing name resolution?

Was my send_sample call correct btw? If so, I can update the chess example
documentation. Tho it returned an error code 6, possibly because of the
lookup issues.

I'm not too familiar with coreos, so am without all my usual debugging
tools :(

Don't worry about setting something up for me. I can try the others myself;
it's really a DO automation that I was after.

Thanks again.

Ian

On Sat, Jan 24, 2015 at 10:25 PM, Joe Doliner [email protected]
wrote:

Hi, really sorry you ran in to this. I think I have an idea what's going
on here.

tl;dr: I think it's a problem with dns.

Pfs uses etcd to announce its services, this is handled by the
announce-master--.service services. Those services do etcdctl set
/pfs/master/0-1 {HOST_NAME}:port. The router service then uses this to
figure out how to contact the master. However this only works if the router
can do a dns lookup. How did you setup the CoreOS instances? I remember
running in to this problem when I tried to get pfs setup on Vagrant.

There are a few easy ways I can think of to fix this for you short term:

• echo "coreos-1 10.132.128.22" >/etc/hosts is a quick hack that
  should make things start working. This would also be a good way to confirm
  the DNS theory.
• GCE and EC2 both have DNS by default, shoot me an email at
  [email protected] and I'll setup a hosted instance for you to play
  around with.

---

Action items for pfs:

• If DNS is indeed the problem this error message should explicitly
  mention that as a likely cause.
• Docs should do a better job of describing this in the "getting
  started" section.
• We should look in to making pfs not depend on DNS. I think this
  will be doable via flannel or something similar.

—
Reply to this email directly or view it on GitHub
#32 (comment).

from pachyderm.