Kubernetes in the Second Decade
KubeCon, Chicago
November, 2023
Tim Hockin, Google
@thockin
(with help from many people!)
Slide 2
Slide 2 text
No content
Slide 3
Slide 3 text
No content
Slide 4
Slide 4 text
“...the next
trillion core
hours...”
Jago Macleod
@jagosan
Slide 5
Slide 5 text
AI / ML
Slide 6
Slide 6 text
“AI/ML will increase
compute resource usage
and the requirement to
manage those
workloads effectively
and efficiently.”
Janet Kuo
@janetkuo
Slide 7
Slide 7 text
“Inference is the new
web app.”
Clayton Coleman
@smarterclayton
Slide 8
Slide 8 text
“The impact of AI/ML
will be on the same
scale as the impact of
the internet itself.”
Tim Hockin
@thockin
Slide 9
Slide 9 text
“I hope AI / ML can help
us to optimize the
testing, debugging, and
supportability [of
Kubernetes].”
Dawn Chen
@dchen1107
Slide 10
Slide 10 text
Clusters
Slide 11
Slide 11 text
“Multi-cluster is
unavoidable, and for
many, it's already the
reality.”
Jeremy
Olmsted-Thompson
@JeremyOT
Slide 12
Slide 12 text
“Clusters are limiting,
users need to work at a
higher level.”
Jeremy
Olmsted-Thompson
@JeremyOT
Slide 13
Slide 13 text
“...maybe not
clusterless, but less
cluster.”
Tim Hockin
@thockin
Slide 14
Slide 14 text
“We're over-indexing
[on multi-cluster],
instead of making
individual clusters
reliably run all kinds of
workloads at once.”
Wojciech Tyczynski
@wojtek-t
Slide 15
Slide 15 text
Complexity &
Reliability
Slide 16
Slide 16 text
“...trying to do
everything...”
Antonio Ojea
@aojea
Slide 17
Slide 17 text
“...trying to solve all
issues for everyone...”
Dawn Chen
@dchen1107
Slide 18
Slide 18 text
“All of us [maintainers]
underestimate the cost
[of complexity] for
users, and I perceive it
as the biggest existential
threat.”
Wojciech Tyczynski
@wojtek-t
Slide 19
Slide 19 text
“I don't think the Rails
equivalent for K8s has
emerged yet ... to solve
this complexity by
removing unnecessary
choices.”
Michael Taufen
@mtaufen
Slide 20
Slide 20 text
“The big risk is its
current complexity.”
Tim Bannister
@sftim
Slide 21
Slide 21 text
“We should focus on
keeping the core small
and stable.”
Maciej Szulik
@soltysh
Slide 22
Slide 22 text
“Avoid gridlock or
calcification.”
Jordan Liggitt
@liggitt
Slide 23
Slide 23 text
“Kubernetes won not
because it’s the best at
any specific workload,
but because it can run
almost everything
reasonably well.”
Clayton Coleman
@smarterclayton
Slide 24
Slide 24 text
“Kubernetes won not
because it’s the best at
any specific workload,
but because it can run
almost everything
reasonably well.”
Clayton Coleman
@smarterclayton
Slide 25
Slide 25 text
“We’ve got to say no to
things today, so we can
afford to do interesting
things tomorrow.”
Tim Hockin
@thockin
Slide 26
Slide 26 text
“Next generation
operators are less likely
to be experts in
Kubernetes ... reliability
and security are more
important than new
features.”
Jago Macleod
@jagosan
Slide 27
Slide 27 text
“Workloads on
Kubernetes have
evolved from casual to
critical, including life
safety, healthcare, telco,
and public sector.”
Jago Macleod
@jagosan
Slide 28
Slide 28 text
“Keeping (or increasing)
the [quality] bar in core
is more important than
any other feature that
we may add.”
Wojciech Tyczynski
@wojtek-t
Slide 29
Slide 29 text
No content
Slide 30
Slide 30 text
Ecosystem
Slide 31
Slide 31 text
“Paradox of Choice - too
many choices actually
limits our freedom, [we
need to] help users
choose or avoid the
need to choose.”
Janet Kuo
@janetkuo
Slide 32
Slide 32 text
“We probably gotta
shrink the landscape to
stay sustainable.”
Davanum Srinivas
@dims
Slide 33
Slide 33 text
“It's hard to find what to
plug in to Kubernetes
and the problem gets
worse the bigger the
ecosystem grows.”
Tim Bannister
@sftim
Slide 34
Slide 34 text
“The growing CNCF
landscape indicates a
lack of standards. This
creates compatibility
and operational
complexities.”
Dawn Chen
@dchen1107
Slide 35
Slide 35 text
“The ecosystem has
flourished, but I see
fewer learnings and
patterns being brought
back to Kubernetes.”
Michelle Au
@msau42
Slide 36
Slide 36 text
“...the big tent of
openstack is a good
example of how this can
go wrong.”
Antonio Ojea
@aojea
Slide 37
Slide 37 text
“Kubernetes should stay
unfinished.”
Tim Bannister
@sftim