nc has no option -z in centos 7 about puppet-galera HOT 6 CLOSED

You're right, and yet the centos7.pp in examples seems to work for me under vagrant, which is a bit strange. Thanks for pointing this out.

from puppet-galera.

Hmm, this is spooky. I've been running this module for several weeks without issue, then yesterday I ran puppet against the "$galera_master" and the node went ahead and bootstrapped itself due to exactly this problem. Yikes.

It's a nasty bug as it left me with the remaining nodes in one cluster and one node in a new cluster. The cluster is frontended by haproxy and this continued to healthcheck all nodes successfully. The next result was that one in N queries was against a bad database!

ylockerman's workaround looks pretty solid, but does take significantly longer to timeout than 'nc -z'. It may be preferable to use something like tcping?

from puppet-galera.

Yep gonna change it to nmap or tcping I think.

from puppet-galera.

I looked into using nmap when I was writing my workaround. The issue I ran into is that nmap (and tcping) is not installed by default in Centos. Thus, if you use it, it seems like the package will need to be installed manually.

from puppet-galera.

Requiring packages isn't a big problem as long as they're in the standard repos or epel at least. I think nmap is in the base ones so that's probably better.

from puppet-galera.

@mf-lit you raise a good point as well - the timeouts (and maybe retries) probably need to be configurable so that a second of packet loss doesn't result in a bad bootstrap.

How's this to address the nc issue?

#49

from puppet-galera.