SAN FRANCISCO – One of a stone stars of a cloud and DevOps transformation common lessons learned
from regulating Opscode’s Chef pattern government tools.
Phil Dibowitz, systems operative during Facebook, demonstrated during ChefConf here this week his
company’s use of Chef and other open
source utilities
to conduct an sourroundings consisting of mixed clusters of 10,000 or
more servers each.
Dibowitz and Pedro Canahuati, executive of prolongation engineering during Facebook, sat down with
SearchDataCenter for a some-more in-depth contention about DevOps, recommendation for DevOps newbies, how
Facebook comparison Chef, and what’s on a wish list for a tool.
What do we consider of a DevOps term? Is it a real-world judgment or an empty
buzzword?
Phil Dibowitz: I consider people like to make terms. People like to tag stuff. But we think
the judgment underneath it is useful.
When we was a youth admin in a late ’90s and early 2000s, someone pronounced to me that the
difference between a comparison admin and a youth admin is that a senior
admin can review and write code. we suspicion that was useful, and we had a mechanism scholarship degree,
and a judgment of sitting in front of a outrageous physique of formula all a time did not sound interesting
to me, though we could see since it was useful. we consider that was kind of an early seed of what we see
today, and people like to call it DevOps.
I call it being good during my job. Some people call it
DevOps.
Phil Dibowitz
systems engineer, Facebook
At a finish of a day, people who are in operations need to be means to review and write code; they
need to be means to understand
the applications they’re supporting; though moreover, as we scale, if we can’t write code, you
can’t automate effectively.
Logging into any appurtenance by palm only doesn’t fly anymore. We’ve gotten past that scale. So if
you have tens of thousands of machines, SSHing
into any one and regulating a authority only isn’t going to fly, though if we can automate
that, afterwards we can indeed run that sourroundings efficiently.
Everyone will give we a somewhat opposite definition. That’s a heart and essence of it, and I
don’t unequivocally caring what we call it. we call it being good during my job. Some people call it DevOps.
Would we have any recommendation for beginners to a DevOps concept?
Dibowitz: The biggest thing is that it’s a mind-set shift. If you’re in a legacy
infrastructure, it’s unequivocally tantalizing to consider about, ‘I have my DNS
server, and we have my mail server, and we need to record in to x server and do x thing.’ And that
mind-set doesn’t scale. Because one day we need 5 DNS servers or 10 mail servers, or 30 Web
servers — or hundreds, or thousands.
Think about your environment, as against to your servers. Think about how we wish to express
the preferred state of your universe in a useful approach that we can pull that out, and always regenerate
everything we have fast and easily, in an programmed fashion.
Phil, in your speak this morning we mentioned ‘configuration as data.’ What accurately does that
mean?
Dibowitz: The existence of it is that if we wish to nominee some subset of configuring a
system or organisation of systems to a program developer, they don’t indispensably know how to be a
sysadmin, they don’t indispensably know all of a Chef, Puppet or CFengine bits, or whatever it is
you’re operative with. What they do know is, ‘I need x megabytes of this,’ or ‘I need my core files
in a opposite directory’ or whatever. And that’s only data. ‘I wish this volume of this. we want
this in this place.’
And so, if we can give them a approach to demonstrate that in data, if we can give them a crush that
represents settings of complement controls or arrays of packages, we get to a place where developers
can conduct a pieces of pattern a focus needs in their environment, in a approach that
works good with a thing they’re used to; it’s only formula and information during that point, and every
developer has worked with formula and data.
Pedro Canahuati: I consider to take his instance even further, a developer maybe needs to
know, ‘I need some-more common memory.’ But he doesn’t need to know, for example, that he’s regulating on a
Linux box, that a Linux complement has a sysctl record that’s in /etc, doesn’t need to know that after we run that file
you have to run a [command] to make those settings correct.
And so, what pattern as information does for us fundamentally says, ‘Here is a record that tells you
[that] we wish some-more common memory. Here’s one tiny apportionment of that crush and all we have to do is
change it, and afterwards a rest is sorcery underneath.’ Developers adore that since they don’t need to
know systems administration. They only have to know that they need some-more shared memory.
Did we do a bake-off between pattern automation tools?
Dibowitz: We did. We looked during Puppet, Chef and Spine. Spine you’ve substantially never heard
of, though it’s a thing we co-wrote when we was during Ticketmaster. … Rather than give everybody their
favorite tool, we gave everybody a apparatus they suspicion would fail, and said, ‘You have to go and sell
this apparatus as best we can.’ We gave everybody a integrate of weeks, and afterwards we met and implemented a
couple of opposite facilities — sysctl being one of them, as we competence suppose — and a SSH
configs, that are sincerely complicated. And afterwards we came behind and looked during a code, we looked at
the knowledge any chairman had perplexing to use it, and how good it would fit into a model.
All 3 collection were means to do a job, though in a case, a coherence of Chef met our
workflow needs best. You have a full energy of Ruby to demonstrate and cgange your configuration, but
you also have a full energy of Ruby to extend and cgange Chef itself. We mutated node.save, which
is an inner square of Chef — we mutated how that behaved.
We could hook a apparatus to what we wanted it to be, rather than hook a workflow to a tool.
The other unequivocally large pull for Chef was, since all happens in a client, regulating it in a
server mode was an choice for us, where it wouldn’t have been for other tools.
Is there anything on your wish list for Chef?
Canahuati: One of a things we onslaught with many is creation certain that a changes that
we make are not inspiring what we don’t design to be altered — a unintended consequences. Chef
gives us a ability to exam things, though I’d like that to be some-more pure and easier to use at
the beginning, where if I’m going to cgange a cache server, we know we will not be inadvertently
modifying a database server. That’s a really, unequivocally tough problem to solve.
Beth Pariseau is comparison news author for SearchCloudComputing.com. Write to her during bpariseau@techtarget.com or follow @PariseauTT on
Twitter.
Article source: http://www.pheedcontent.com/click.phdo?i=8251ed1956705da152cc62fbddab72e0
