►
Description
Presented by Lars Schneider, Autodesk & Todd O'Connor, Adobe.
Managing GitHub Enterprise is much more than just provisioning an appliance. In this talk we will cover techniques and strategies how to manage a large scale appliance efficiently. We will share best practices for the initial setup, solutions for common issues, as well as debugging recommendations and tips for interacting with the GitHub support team. All topics will be accompanied by hands on tutorials or code examples that are directly applicable by GitHub Enterprise administrators.
About GitHub Universe:
GitHub Universe is a two-day conference dedicated to the creativity and curiosity of the largest software community in the world. Sessions cover topics from team culture to open source software across industries and technologies.
For more information on GitHub Universe, check the website:
https://githubuniverse.com
A
All
right,
everyone
in
the
house,
let's
get
her
going
so
I
stand
between
you
and
the
party.
So
let's
get
through
this
hi
everyone,
I'm
todd,
o'connor,
I'm,
a
senior
SDM
engineer
at
Adobe
I'm
here
today
with
large
Snyder
technical
lead
for
github
at
Autodesk
and
we're
going
to
present
advanced
github
Enterprise
administration.
A
You
might
be
wondering
what
does
Adobe
and
Autodesk
have
in
common.
Besides
the
similarities
in
our
logos
and
I
will
say
that
has
not
gone
unnoticed.
Well
in
working
with
Mars
over
the
last
month
on
this
presentation
we
came
to
realize
we're
trying
to
solve
a
lot
of
the
same
challenges
and
in
talking
to
a
lot
of
you
this
week,
you're
trying
to
solve
those
same
challenges
too.
So
we
hope
you
get
a
lot
out
of
our
information
and
our
presentation
today.
A
So
here's
the
agenda
for
today
we
have
a
lot
of
information
to
cover.
Oh,
yes,
there
is
one
more
thing:
I
think
we
saved
the
best
for
last
Millar's
will
show
us
something
brand
new,
never
seen
before,
and
I
think
you're
going
to
love
it.
The
inspiration
for
this
was
a
talk
by
David
Mittman
from
JPL
two
years
ago,
at
Universe
we
wanted
a
big
shout
out
to
David.
Thank
you.
A
These
are
the
topics
I'm
going
to
cover
in
more
detail,
server
considerations,
the
admin
center,
your
high
availability
and
disaster
recovery,
as
well
as
some
other
techniques,
so
server
considerations.
First
and
foremost,
you
want
to
be
on
a
dedicated
VM
host,
Lars
and
I.
Both
know
from
experience
not
to
be
on
a
shared
hosts.
Shared
VMs
on
a
shared
hosts
are
going
to
steal
your
resources.
You
don't
want
that.
This
is
one
time
where
greed
is
good.
You
want
the
VM
all
to
yourself.
A
The
next
thing
you
want
to
look
at
is
RAM.
You
want
to
allocate
as
much
RAM
as
that
appliance
is
going
to
take.
The
system
makes
heavy
use
of
kernel
disk
cache
to
speed
up
many
good
operations.
Now
in
my
appliance,
I
have
four
hundred
fifty-five
gig
of
ram
at
any
one
time.
I
have
less
than
half
of
that
sustained
use
so
I'm
at
about
180
190,
so
the
remainder
of
that
is
being
used
as
disk
cache
and
that
greatly
speeds
up.
A
My
disk
reads
and
my
I/o
operations
and
I
know
that
by
looking
at
the
disk
I/o
graph,
my
read
operations
are
extremely
low,
so
you
definitely
want
to
allocate
as
much
RAM
as
you
can.
The
next
step.
The
next
piece
to
look
at
is
your
number
and
speed
of
your
CPUs
alone.
With
my
455
gig
of
ram
I
have
48
CPUs.
This
is
something
you're
going
to
have
to
track
over
time.
You're
gonna
have
to
do
a
historical
you're.
Gonna
have
to
look
at
it.
A
You're
gonna
have
to
adjust,
add
more
CPUs
adjust
the
speed
of
those
CPUs.
Now,
if
you're
ever
in
doubt
about
either
of
these
a
great
resource
is
to
take
a
system,
bundle
send
it
into
the
excellent
tech
support
folks
at
github
and
they'll
analyze
and
they'll.
Give
you
some
recommendations
and
later
on
in
the
presentation
Lars
is
going
to
talk
about
how
to
interact
with
the
tech
support.
Folks.
A
The
only
way
that
you
could
resize
that
partition
is
to
completely
rebuild
the
appliance
very,
very
inconvenient,
with
2.11
now
they've,
given
you
the
capability
they've,
given
you
the
capability
to
resize
that
partition,
which
is
really
nice
now
in
a
stopgap
measure,
you
could
keep
fewer
logs.
Github
keeps
seven
logs
on
disk
at
one
time
for
rotation.
So,
if
you're
really
running
tight
work
with
tech
support,
adjust
your
log
rotations
only
until
you
can
get
that
wronging
resized
other
server
considerations
use
a
staging
server.
We
use
a
staging
server.
A
It's
great
for
testing
upgrades,
it's
great
for
testing
that
last
feature.
I
talked
about
resizing
the
root
partition.
You
definitely
want
to
be
testing
and
using
a
staging
server.
Now.
One
little
fact
is
that
your
existing
github
license
will
apply
to
that
staging
server.
So
you
can
spin
it
up.
No
problem
use
the
github
backup
utilities.
A
This
should
just
run
on
a
dedicated
VM
doesn't
have
to
be
anything
fancy
small
little
VM
run
your
backups
often
I
brought
my
backup
six
seven
eight
times
per
day,
and
none
of
my
backups
take
any
more
than
an
hour,
so
you
run
them
frequently
they're
going
to
be
very
small
and,
most
importantly,
make
sure
to
backup
your
backup
server.
You
don't
want
to
have
to
be
standing
in
front
of
a
server
down
situation.
A
VP
is
asking
when's
it
going
to
be
backup,
and
you
don't
know
because
you've
never
tested
your
backup
restore
process.
A
Where
can
you
test
that
backup
restore
process
on
that
staging
server,
long
term?
Log
retention?
Maybe
it's
a
security
policy
at
your
business?
Maybe
you
need
to
retain
those
logs
for
more
analysis,
send
your
logs
off
server
so
to
something
like
spunk,
and
you
would
do
this
in
your
management.
Console
the
admin
Center
Lars
and
I.
Both
look
at
a
lot
of
the
settings
in
the
admin
center
to
adjust
our
appliances
accordingly
force
pushes
do
you
allow
people
to
force,
push
or
not
force
pushing
rewrites
history?
You
might
want
to
consider
that
repository
upload
limit.
A
A
The
large
files
contribute
to
size
over
time
to
your
repos
cloning
operations,
get
slower
and
slower
over
time
as
more
versions
of
those
large
files
come
in,
so
definitely
adjust
your
limits.
Large
file,
support,
LFS,
that's
something
new!
That's
come
out
recently.
You
can
keep
your
large
files
off
of
your
appliance
and
maybe
on
to
a
dedicated,
lambic
storage
or
maybe
in
a
separate
area,
on
your
appliance
as
well,
something
as
you
consider
and
reactivating
suspended
users.
Again.
A
I
saw
a
presentation
earlier
today
where
they
were
talking
about
suspending
on
suspending
users
through
their
self-service
portal.
You
can
reactivate
suspended
users
automatically
by
checking
this
box
as
soon
as
a
user
logs
into
the
UI
through,
like
a
browser,
their
account
is
reactivated
transparent
to
them.
They
don't
even
know
what
happened
saves
also
on
your
licensing,
so
a
server
high
availability
and
disaster
recovery
considerations.
You
definitely
want
to
be
using
a
replica
server
now
in
our
in
our
model,
where
a
single
server
so
I
have
a
single
replica
one
to
one
pairing
I
cluster.
A
A
Ideally,
it's
going
to
be
close
to
your
primary
for
speed
for
recovery,
but
there
are
also
some
other
considerations
right
if
they're,
if
your
h.a
is
in
the
same
data
center
as
your
primary
and
the
whole
data
center,
goes
down
you're
kind
of
out
of
luck,
so
look
to
where
you
want
to
put
that
high
availability
server,
your
dr
system,
this
is
typically
set
up
at
a
remote,
distant
site
and
you
build
that
from
a
backup
restore
process.
And
how
long
does
that
backup,
restore
process?
A
Take
you'll,
know
that,
because
you've
been
doing
that
on
your
staging
server
now
2.11
goes
one
step
further
and
has
some
geo
replication
built-in,
so
you
can
have
read-only
replicas
of
remote
sites,
but
these
would
be
read-only
replicas.
For
your
end-users
to
speed
up
their
read
operations,
so
you
might
need
a
replica
in
your
London
office,
for
instance,
and
the
local
users
in
London
would
be
hitting
that
for
their
read
operations,
so
some
other
techniques
that
we
use.
We
have
an
overarching,
org
owner.
A
This
utility
account
that
we
have
as
a
member
of
every
single
organization
it's
used
for
data
mining.
We
use
it
to
build
out
a
list
of
all
the
org
owners
so
that
people
within
the
company
know
who
to
contact
for
access.
My
team
doesn't
grant
access.
I'll
say:
go
talk
to
this
team
over
here,
because
they're,
the
ones
that
manage
their
orgs
and
they'll
know.
If
you
really
need
access
or
not.
So
that's
something
that's
useful.
We
aggressively
suspend
users
that
have
both
left
the
company
and
that
are
in
a
certain
period
of
time.
A
Our
threshold
for
stale
is
is
pretty
low.
You
might
adjust
it
accordingly
and
then,
if
you're,
using
that
reactivate
suspended
users,
if
they
are
suspended,
they
hit
the
browser
and
login
they're
reactivated
immediately
considering
selective
pre
receive
hooks.
Both
Lars
and
I
use
pre
receive
hooks
at
our
company.
Some
things
that
you
can
look
at
blocking
certain
pala
types,
don't
allow
secrets.
For
instance,
another
thing
would
be
checking
for
bug:
IDs,
that's
a
common
one.
A
Some
of
the
new
features
that
you
might
want
to
investigate
in
the
newer
versions
here
rate
limiting
from
experience,
I
turned
it
on
within
a
couple
hours.
I
had
to
turn
it
off
so
you're
gonna
have
to
check
this
out,
look
at
the
values
and
maybe
use
something
that
we're
gonna,
be
showing
you
a
little
bit
later
on
to
help
you
analyze
your
rate,
limiting
dedicated
the
lambic
storage,
as
I
mentioned
LFS
earlier,
you
can
put
your
LFS
artifacts
on
a
separate
set
of
discs
may
be
cheaper
storage
and
geo
replication.
A
A
I
often
do
that
for
myself
service
portal,
my
wiki
page
of
documentation
and
some
other
things,
especially
a
link
to
my
staging
server
people
might
not
know
where
to
login
I'll
also
have
a
mob
break
and
I'll
put
in
a
second
line,
announcing
upgrades,
self-service
portals
I
mentioned
some.
Other
companies
are
doing
the
same
thing,
building
self-service
portals
and
you
might
think
well.
Why
is
this
under
helping
administrators
Todd?
It's
helping
me
cut
down
an
email
to
me
so
accounts
that
can't
log
in
via
browser,
for
instance,
can
use
this
to
manage
their
SSH
keys.
A
They
can
set
up
integrations
where
I
don't
want
to
share
the
secrets
with
someone
to
set
up
Jenkins
or
JIRA,
or
something
like
that
here
are
some
screenshots
of
our
self-service
portal
really
quickly.
Folks
can
log
in
suspend
on
suspend
integrations
with
fisheye
and
they
can
manage
their
generic
or
utility
accounts.
Another
set
of
screenshots
drilling
down
in
a
little
bit
deeper.
They
can
manage
SSH
keys
and
do
via
SSH
key
Varrick
and
again
suspend
unsuspended
a
quick
little
screenshot
here
of
what
it
looks
like
for
our
managing
fisheye
integrations.
A
We
looked
at
see
who
is
an
owner
of
the
repo
and,
if
you're,
an
owner
of
a
repo
that'll
show
up
here,
you
can
enable
that
and
you
can
also
disable
that
integration-
and
this
is
great
because
now
I
don't
have
to
turn
over
my
secrets
to
you
to
set
this
up.
You
can
just
go
to
our
self-service
portal
and
you're
done
and
now
I'll
turn
the
stage
over
to
Lars
and
he'll
continue
with
the
presentation.
B
Thank
You
Todd
at
first
I
also
want
to
introduce
a
new
tool,
and
at
is
Scotty.
Scotty
is
a
collection
of
scripts
for
get
up.
Enterprise
administrators
and
all
scripts
are
open
sourced
and
you
can
download
them
from
github
right
there
in
case
you
wonder
where
the
name
comes
from
well,
whenever
I
read
get
up,
Enterprise
I
can't
help
it
I
have
to
think
of
the
spaceship
enterprise,
and
you
know
Scotty
is
the
chief
engineer
of
this
spaceship.
So
this
is
where
the
name
comes
from.
So
what
can
Scotty
do
for
you?
B
Scotty
has
two
kinds
of
scripts
conveniently
organized
in
two
different
directories:
there's
the
repo
directory
that
contains
scripts
that
access
your
appliance
in
a
read-only
fashion.
These
scripts
query
to
get
up
database
and
blocks
for
all
kinds
of
useful
information
and
then
there's
the
admin
directory.
This
directory
contains
scripts
that
modify
your
appliance,
sometimes
in
undocumented,
and
not
officially
supported
ways.
So
if
you
use
them
make
sure
you
read
and
understand
them.
So
how
does
a
Scotty
script
look
like?
B
Well,
all
scripts
have
to
age
or
help
option
which
prints
the
description
of
the
script
and
other
available
options.
All
scripts
have
also
the
end
option,
which
would
perform
a
dry
run
of
the
script.
That
means
the
script
would
show
you
what
it
would
do
on
your
instance,
but
it
wouldn't
do
actually
anything.
B
B
This
is
very
useful
in
case,
for
instance,
you
have
a
you,
have
a
git
client
with
a
vulnerability,
and
you
want
to
see
which
of
your
engineers
actually
does.
Does
you
use
this
git
client,
because
you
were
to
make
sure
that
they
all
use
the
the
fixed
version
of
kid
right
and
when
you
run
the
script,
you
would
get
an
output
like
that.
You
see
what
wasn't
and
how
many
users
are
using
it
and
by
default
Scotty
generates
these
scripts
over
the
most
recent
block
data.
That
means
usually
over
the
last
24
hours.
B
If
you
are
interested
in
a
greater
timespan,
then
you
can
pass
the
a
option
to
make
Scotty
look
at
all
available
log
data,
and
that
means
it
means
usually
over
seven
days
all
right
so
much
for
the
introduction
to
Scotty
we'll
get
back
to
these
scripts
in
a
bit.
Let's
talk
now
about
get
up
support
tickets.
If
you
rent
run
into
any
kind
of
problem
on
your
appliance,
then
you
can
reach
out
to
the
get
up
enterprise
support
team.
B
B
Second:
explain
the
problem,
kindly
short
and
precise,
put
yourself
into
the
shoes
of
the
support
engineer
when
you
write
the
ticket.
That
means
in
particular
avoid
company
specific
slang
or
observations
or
code
names,
because
the
support
engineer
has
no
way
to
understand
them.
Also
at
details
that
might
be
relevant
to
the
investigation,
in
particular.
Add
steps
to
reproduce
the
problem.
B
If
you
know
the
time
of
the
incident
and
provide
that,
ideally
in
UTC,
because
this
helps
the
support
engineer
to
find
the
relevant,
lock
data,
also
if
a
client
is
involved,
then
at
all
kinds
of
information
around
the
client,
the
IP
address
of
the
client
or
the
versions
of
the
software
that
the
client
is
using.
So
the
Gipper
version
get
LFS
version,
browser
version,
operating
system
versions,
these
of
things
and
also
if
the
problem
is
related
to
certain
users,
then
give
the
support
engineer
the
user
names
that
are
uninvolved
the
repository
names.
The
pull
request.
B
B
So
after
you
submitted
your
ticket,
then
get
up
will
give
you
a
support
ticket
ID,
and
you
can
use
that
to
upload
support
bundles
a
support
button
is
basically
a
zip
file
containing
all
relevant
users
unlock
data
of
your
appliance
to
upload
a
support
banner.
You
need
to
log
into
the
administrative
shell
and
run
the
gh3
support,
bundler
command
and
two
hints
here.
First,
your
appliance
requires
internet
access
to
upload
the
support
button
to
get
up
so
make
sure
that
your
firewall
allows
this
kind
of
access.
B
Second,
the
upload
may
type
might
take
a
while,
in
our
case,
it
can
take
up
to
30
minutes
to
upload
a
support
button.
So
if
you
ssh
into
the
appliance-
and
this
sh
connection
from
a
local
machine
to
other
appliance
drops,
then
the
upload
to
get
up
would
fail.
Well,
therefore,
I
recommend
to
run
the
duty
support
banner
command
in
a
screen
session
on
your
github
Enterprise
appliance,
then
you
are
basically
safe
from
this
kind
of
error,
so
a
normal
support
panel
would
usually
cover
the
Loctite
of
the
last
24
hours.
B
That
contains
log
data
from
all
your
different
for
me
into
a
cluster
and
then
there's
another
case,
and
sometimes
you
might
want
to
upload
only
a
single
file
to
make
it
available
to
a
get
up,
support
engineer
and
you
can
also
use
the
GG
support
banner
command
for
that
just
pass
the
dash
H
up
from
F
option
with
a
file
name
and
then
there's
one
more
case
to
investigate
certain
problems.
You
might
need
to
upload
files
from
your
local
machine.
B
For
instance,
if
you
capture
a
TCP
traffic
to
analyze,
network
issues
and
get
up,
has
no
command
for
you
here,
but
don't
worry
Scotty
has
you
covered
Scotty.
Has
the
support
bundle
local
attach
command
that
you
can
use
to
upload
any
file
from
your
local
machine
to
a
get
up
support
ticket,
and
while
we
edit
I,
also
want
to
introduce
another
Scotty
command,
that
I
think
is
very
useful
and
it
just
sent
ghe
diagnostics.
So
for
some
tickets
a
full-fledged
spot
bundle
might
be
too
much,
as
I
said
it
takes
a
while
to
upload
it.
B
B
B
Okay,
so
some
users
might
complain
that
their
commits
are
not
probably
associated
to
them
on
get
up.
They
would
look
like
this.
You
know
your
name
with
this
little
blue
question
mark
behind
it.
Well,
as
an
experience
could
have
enterprise
admin,
you
would
know
that
the
local
git
config
is
not
configured
with
the
correct
email
address
and,
of
course,
this
is
easy
to
fix
for
a
single
user.
B
First,
we've
installed
a
previous
six
pre
receive
hook
on
our
appliance,
and
this
hook
checks
every
commit
for
proper
outlets
email
addresses.
If
the
check
fails,
then
we
will
still
accept
a
push.
However,
we
will
show
to
use
a
warning
message
like
this
and
in
this
morning
message
we
briefly
explained
the
problem
and
we
suggest
the
elders
get
up
and
what
the
auditors
get
environment
as
solution.
B
The
out
of
the
skit
environment
is
a
script
that
configures
I
use
a
git
client
properly.
That
means
it
ensures
that
name
and
email
address
is
set
correctly
for
the
user,
but
it
doesn't
stop
there.
It
also
checks
that
you
are
using
the
correct,
get
version
the
correct
get
out
of
S
version.
It
makes
you
a
get
out
of
SS
installed.
If
it's
not
installed,
it
will
install
it.
It
will
also
check
your
authentication
mechanism.
It's
your
credential,
her,
but
correctly
set
up.
It
will
set
up.
B
We
create
I,
am
a
personal
access
token
for
you
to
access,
get
up,
and
all
these
things
and
many
more
so
this
order
s
kid
environment
will
also
work
on
on
all
platforms.
It
works
on
Windows,
Mac
and
Linux,
and
we
have
made
very
good
yeah.
It's
a.
It
was
very
helpful
for
us
to
get
our
the
clients
of
our
engineers
streamlined
with
that
tool
and
both
of
these
tools.
B
Ok,
next
I
want
to
talk
about
recovering
branches.
As
you
know,
kit
is
made
for
branching.
However,
sometimes
I
use
the
delete,
sub
branch
accidentally
and
then
their
work
appears
to
be
gone,
but
no
worries
again.
Scotty
can
help
you
here.
Scotty
has
a
script
to
conveniently
retrieve
all
events
of
a
repository,
and
then
you
can
search
these
events
for
your
branch
name
to
find
a
location
event.
If
you
do
that,
then
you
will
get
a
result
like
this.
The
first
line
shows
the
name
of
the
branch.
B
B
The
previous
commit
is
the
interesting
one.
That's
the
commit
we
want
to
recover,
so
we
copy
the
commit
hash
to
our
clipboard
and
go
to
the
repository
on
our
github
Enterprise
appliance
and
then
we
add
tree
slash
the
commit
hash
to
the
URL,
and
this
will
bring
us
to
the
exact
state
of
the
deleted
branch.
And
then
we
use
the
branch
drop
down
menu
to
recreate
the
branch
and
that's
it.
We
will
cover
our
branch,
however,
for
shared
branches
and
in
particular
for
release
planters.
B
Okay,
now
I
want
to
show
you
how
you
delete
a
git
repository
properly.
You
know
giving
it
up.
They
are
pretty
conservative
when
it
comes
to
the
lesion.
In
most
cases,
you
can
recover
things
up
to
90
days
after
bleezin
and
in
general.
This
is
good.
However,
as
an
administrator
you
might
have
to
delete
repositories
right
now.
Maybe
the
source
code
was
sold
and
you
have
to
delete
it
for
legal
reasons
and
again,
Scotty
can
help
you
here.
B
You
can
use
the
repo
purge
archive
script
to
purge
and
script
from
your
instance
after
you
have
deleted.
The
repository,
however,
be
careful.
Even
after
after
running
this
script,
you
still
need
to
delete
Forks
clones
and
backups
of
the
repository.
So
don't
forget
that,
okay,
let's
jump
to
the
next
section,
let's
talk
about
debugging
get
and
get
up.
First
I
want
to
show
you
how
your
back
get
problems
on
a
user
client.
Let's
use
this
git
clone
command
as
an
example
usually
get
prints,
only
minimal
output.
B
Here,
for
instance,
we
see
some
statistics
about
a
clone
in
case
of
problems.
You
want
to
exactly
know
what
is
going
on
under
the
hood
and
you
set
the
good
trace
environment
variable.
Then
it
will
show
you
all
the
sub
commands
that
it
runs,
and
this
can
be
very
valuable
to
understand
a
given
problem
and
if
you
set
the
gate
trace
performance
variable
then
get
would
also
tell
you
how
long
every
sub
command
took,
and
this
is
useful
to
investigate
performance
related
issues.
B
If
you
want
to
understand
why
a
certain
gate
command
is
slow
on
your
machine,
but
to
be
honest,
the
majority
of
problem
that
you
were
investigators
get
up
administrator,
our
likely
network
related,
and
there
are
two
more
environment
variables
that
I
want
to
show
you.
If
you
get
client
communicates
via
HTTP
or
HTTPS,
then
you
can
set
the
gate.
Curl
the
boss
variable
to
see
what
is
going
on
between
you
and
the
server.
B
B
The
first
component
is
H
a
proxy
this
component
and
it's
all
traffic
to
the
gear
up,
Enterprise
appliance
and
you
can
find
the
log
file
at
bar
log
h
a
proxy
door
lock,
and
this
is
a
very
interesting
log
file.
You
can
learn
what
kind
of
requests
were
made
against
your
appliance
and
how
many
of
them
and
where
these
requests
come
from
and
Scotty
can
help
you
to
analyze
this
log
to
the
H
I
proxy
request.
Script
will
print
statistics
about
the
different
requests
that
have
been
made.
B
B
You
can
also
learn
what
clients
generated
these
get
errors
and
again
there
are
Scotty
scripts
to
analyze
the
locks.
This
pier
can
be
quite
useful
because
us
administrator,
you
will
be
able
to
see,
get
problems
before
your
users
do,
and
the
barber
D
component
serves
the
good
content
to
the
client.
However,
it
will
only
do
that
after
the
user
has
successfully
authenticated
against
get
up
and
babidi
uses
to
get
off
component
to
check
the
credentials
in
the
get
off
lock.
B
You
can
learn
about
HTTP
or
ssh
authentication
problems,
and
you
can
also
learn
who
of
your
users
authenticate
with
username
and
password,
and
this
is
in
particular
relevant
if
you
have
an
LDAP
setup,
as
we
have
at
Autodesk.
If
you
have
configured
l-dub
then
get
up
needs
to
check
the
username
and
password
always
against
l-dub.
B
B
Behind
nginx
we
find
a
lambic
Alembic
serves
other
terminators
release,
binaries
and
uploads
on
these
kind
of
things
behind
nginx
we
also
find
unicorn,
and
this
is
the
actual
get
up
Enterprise
website.
So
a
very,
very
important
component
and
again
we
have
Scotty
scripts
to
analyze
the
errors
in
this
component.
And
last
but
not
least,
we
have
the
LFS
server,
and
this
component
serves
all
the
large
files
to
your
users.
So
this
is
a
very
high-level
overview
of
the
get
up
and
oppress
architecture.
B
All
right
in
the
beginning
of
this
talk,
taht
promised
you
one
more
thing
and
I
want
to
show
you
that
now
I
want
to
introduce
to
you
Hubbell
enterprise,
a
tool
to
visualize
collaboration,
usage
and
health
data
of
your
get
up,
Enterprise
appliance
Hubble
shows
a
variety
of
metrics
and
I
want
to
highlight
a
few
of
them
in
the
following.
You
can
follow,
along
with
our
interactive
demo,
on
get
up
calm
on
your
own
machine
right
there.
B
Let's
start
with
collaboration,
Hubble
shows
the
collaboration
on
your
appliance.
It
will
render
a
chart
that
visualizes,
how
many
members
of
one
organization
contribute
to
other
organizations,
and
this
includes
contributions
made
by
a
direct
pushes
and
by
a
pull
requests.
If
you
have
a
over
an
organization,
and
you
can
conveniently
see
the
contribution
relationships,
this
gives
you
a
good
overview
of
what
product
and
teams
are
the
top
collaborators
on
your
appliance,
and
this
is
especially
true
if
the
organization's
have
a
sufficiently
different
topics.
B
I'm
mentioning
this,
because
sometimes
you
will
also
discover
relationships
like
this.
These
two
organizations
they
cover
the
same
topic
and
therefore
the
collaboration
is
not
surprising
at
all,
and
in
these
cases
it
might
make
sense
to
consolidate
these
two
organizations
into
one
and
with
trouble.
You
can
learn
even
more.
You
can
ask
Hubble
to
visualize
all
relationships
of
a
given
organization,
and
this
is
especially
interesting
for
organizations
that
share
or
at
hosts
shared
code.
B
That,
however,
doesn't
stop
there.
It
also
visualizes
how
you
engineer
SEOs
get
up
here.
For
instance,
we
see
the
percentage
of
repositories
that
use
pull
requests
on
our
appliance,
and
this
can
be
an
indicator
for
the
usage
of
quote
review
in
your
company,
and
paraca
scanners
can
tell
us
even
more
to
Hubble
shows
the
percentage
of
repositories
that
use
pull
requests
with
status
checks.
Configured
status
checks
can
be
an
indicator
for
CI
adoption
at
your
company
and
Hubble
with
Hubble.
You
can
also
look
at
clones
and
fetches
on
your
appliance
and
here's
a
typical
example.
B
B
However,
if
we
look
at
the
extra
amount
of
data
that
was
transferred,
then
we
can
see
that
clones
caused
much
much
more
traffic
than
fetches
and,
if
you
think
about
it,
this
actually
makes
sense,
because
a
fetch
only
transfers
the
latest
changes.
A
clone,
on
the
other
hand,
will
transfer
the
entire
repository
with
all
its
history
and
in
many
cases
users
perform
clones
where
they
should
rather
use
fetch.
But
how
do
you
identify
these
users?
Well,
Hubble
helps
you
hit.
B
It
prints
a
table
of
the
repositories
and
users
that
cause
the
most
traffic
on
your
appliance.
Let's
use
the
Akamai
repository
as
an
example.
The
repository
is
just
2
megabyte
in
size.
However,
a
misconfigured
CI
process
clones,
the
repository
1200
times
a
day,
and
this
causes
250
gigabyte
of
traffic
a
day,
and
these
kind
of
things
happen
and
believe
me,
a
lot
and
a
users
behavior
like
this
will
stress
even
the
most
powerful
get
up
and
the
price
appliance
and
with
Hubble.
B
You
can
address
these
kind
of
issues
by
contacting
the
responsible
teams
and
solving
the
issue,
and
after
fixing
these
issues,
the
CI
processes
will
usually
even
be
faster,
as
fetching
is
always
more
efficient
than
cloning,
and
the
overall
performance
of
your
appliance
will
also
be
better
because
you
have
more
resource
available
for
users,
so
everybody
wins
some
there's
more
to
come.
Really,
we
released
Hubble
just
two
weeks
ago
and
we
have
lots
of
ideas
for
the
future.
We
are
also
interested
in
your
ideas.
B
Okay,
so
let's
wrap
up
what
tools
forget
up
Enterprise
administrators
have
we
discussed
today
we
have
discussed
Hubble
Enterprise,
a
dashboard
for
your
gear,
up,
Enterprise,
appliance
to
monitor
and
visualize
collaboration,
users
and
health
data.
We
have
discussed
Scotty
a
collection
of
helper
scripts
for
you
get
up,
Enterprise
appliance.
We
have
briefly
discussed
Enterprise
config,
forget
a
script
that
configures
to
get
client
consistently
in
your
company,
and
we
have
discussed
get
up
previously,
folks
that
you
can
get
from
the
get
up
platform
samples
repository.
B
B
A
B
A
The
platform
samples
it's
actually
a
default,
one,
that's
already
out
there
and
you
just
changed
the
default
names
so
yeah
you
have
to
keep
in
mind
with
those
previous
ebooks.
You've
got
like
five
seconds
that
the
thing
has
to
run
in
if
it
runs
any
longer
than
that
is
just
gonna
block
the
whole
thing.
So
it's
got
to
be
really
really
fast.
So
we
just
look
at
really
simple
file
names,
yep,.
B
That,
okay,
there
was
a
comment
that
you
can
disable
user
and
password
authentication
and
within
the
github
enterprise
admin
settings.
That
is
true,
however,
two
problems.
First,
many
people
are
using
it
still
and
it
would
break
many
things
and
second
username
and
password
authentication
is
useful
at
times.
If
you
just
log
into
a
server
clone
repository,
it's
nice
to
have
it
it's
just.
We
don't
want
to
have
people
making
like
30,000,
username
and
password
requests
a
day,
and
we
have
this
a
lot
so.
A
C
B
A
I'm,
sorry,
how
do
we
QA
changes
to
our
pre
receive
hooks
that
staging
server,
so
we
test
everything
over
there
and
we'll
also
go
through
a
code
review
process?
Probably
much
like
yourself.
You
know
we
always
get
bit
by
the
the
single
person.
Writes
it
the
single
person
reviews
it
single
person
tests
it
right.
So,
if
Lars
run
my
team,
I'd
say:
hey
Lars,
I've
just
wrote
this
cool
proceed
book.
Could
you
you
know
check
it
out
and
go
through
there
yeah.
A
B
Is
that
is
totally
true
and-
and
you
can
for
the
previous-
if,
if
you
could
put
them
in
kind
of
a
testing
mode
so
that
they
won't
affect
anything,
it
would
just
make
an
output,
and
this
is
something
that
will
come
to
Hubble,
that
you
see
how
many
of
your
peers
see
folks
actually
fail
would
fail
yeah,
because
we
can
see
that
in
the
locks
as
well.
So
if
you
make
a
change
to
a
previously
folk-
and
it
fails
all
the
time
in
this
demo
mode,
then
you
would
no
question
over
here.
B
Okay,
so
yeah,
so
the
question
was:
how
do
we
actually
validate
the
email
addresses
in
this
previously
folk
and
the
answer
is
we
cheat,
so
we
just
check
that
is
in
our
case,
we
just
check
that
is
an
autodesk
email
address.
That's
our
experience
showed
us
that
this
is
sufficient
to
for
our
purpose
and-
and
this
check
is
very
quick.
So,
as
Todd
mentioned,
we
need
to
be
under
five
seconds,
and
this
is
how
we
do
it
kind.
A
B
A
A
How
did
we
learn?
The
question
is:
how
do
we
learn
about
sizing
our
instance,
the
caching
and
all
of
the
things
that
go
into
making
a
well
a
healthy,
github,
Enterprise,
appliance
and
I?
Think
going
back
to
my
talk
from
last
year?
It's
that
thing
called
experience,
the
thing
that
we
get
when
who
don't
get
what
we
wanted
and
I
know
over
time.
We've
had
to
rebuild
our
instance
several
times,
and
it's
also
working
with
github
support.
A
I'd
send
in
support,
bundles
and
they'd,
get
back
to
me
and
say:
hey
Todd,
you
know,
you're
you've
got
a
lot
of
CPU
load,
you
need
to
add
more
CPUs
or
you're
really
high.
On
your
RAM
usage.
You
really
need
to
get
more
RAM
on
your
instance,
so
we'd
go
through
and
we
adjust
those
thresholds
accordingly
with
the
help
of
support,
but
now
with
Hubble
scouting
a
lot
of
these
other
tools,
we
can
find
out
some
analysis
and
some
details
and
some
trends
within
our
appliance.
How
about
you?
What
how
did
you
go
about?