►
From YouTube: GitLab 12.9 Kickoff - Verify:Runner
Description
A review of the key issues planned for release in GitLab Runner 12.9.
A
Hello,
everyone
Desmond
here
broke
mind.
You
forget
that
Mona
today,
let's
spend
some
time
reviewing
a
few.
The
key
issues
that
we're
working
to
deliver
in
the
twelve
at
nine
release.
The
first
miss
Sherman
enjoy
attention
to
its
issue.
Number
four
one-one-nine
the
job
market
success
when
it
terminates
Midway
Inc
you,
bananas
issue,
as
you
may
know,
we
happen
to
be
in
this
issue
for
the
past
few
min
these
milestones,
however,
we're
at
a
point
now
where
we
do
expect
a
fix
to
ship
a
solution
for
this
issue
and
12.9.
A
This
is
a
issue
with
the
cube
news
executor
where
a
job
marked
a
success
to
meet
during
execution.
An
incorrectly
report
that
successfully
completed
the
approach
of
solving
this
bug
is
that
we
will
no
longer
be
using
cubic
zi
to
execute
scripts
cubic
zi
does
not
allow
for
differentiating
between
a
script
that
acted
correctly
versus
one
that
exited
due
to
due
to
a
connection
issue.
So
the
open,
mr
mr
number,
175
775
will
implement.
A
The
use
of
cube,
attach
cube
attach
is
a
cube
command
option
for
attaching
to
a
process
that
is
already
running
inside
an
existing
container.
So
the
cube
attached,
we
can
execute
scripts
through
the
main
shell,
that's
giving
the
container
alive
as
well
as
sends
tenant
input
to
the
running
process.
So
this
is
the
approach,
then,
that
we
were
taken
to
resolve
this
issue
whose
roots,
at
a
time
to
how
cube
exec
handle
script
errors
that
have
failed
due
to
connection
issues.
A
The
Alysha
should
I
want
to
draw
attention
to
and
as
related
to
work,
we're
doing
to
resolve
four
one
one,
four
one
one
nine
is
issue
number
three,
two
four
seven,
and
this
is
the
error
dialing
back
in
the
issue,
specifically
with
a
Google
Google
communities
engine.
A
quick
recap
of
this
book
on
gke
uses
reported
from
time
to
time.
At
the
github
run
about
generate
an
error,
job
failed
system,
failure,
error,
darling,
back-end.
A
We
have
been
able
to
reproduce
this
bug
and
I
found
that
there
two
scenarios
in
which
this
bug
can
occur
scenario
number
one.
We
execute
a
command
inside
of
a
container
and
during
command
execution,
which,
in
these
cases
are
typically
long-running
commands.
The
container
becomes
unreachable
since
kind
of
thing,
using
cue,
exact
to
execute
a
command
inside
a
container.
The
proposed
change
can
are
one
seven,
seven
five
to
adjust
issue
for
109
should
address
in
this
scenario
for
the
cell
that
uses
experiencing.
A
The
second
can
scenario.
Conduct
can
also
generate
the
error,
their
users
reporting
here
in
the
blog
description
as
well.
We
need
to
ng
key
master
pot
for
eviction
if
a
part
is
math
for
eviction
and
we
issue
a
command
against
that
part.
Then
Cuban
eyes
will
respond
with
the
error,
as
we
can
no
longer
connected
ippolit,
since
obviously
it's
not
for
eviction.
So
for
this
second
scenario,
we're
currently
investigating
the
behavior
in
GTE,
as
it
relates
to
how
G
key
is
handling
some
leaving
that
we
apply
to
the
cluster.
A
So,
in
short,
then,
the
merchant
press
to
resolved
issue,
409
module
press
177
5-
will
help
us
resolve
resolve
the
first
scenario,
as
relates
to
the
3
to
4,
7
bug
and
I
will
continue
investigating
the
second
scenario
as
well
during
the
12th
at
9
release
milestone
the
other
should
I
want
to
call
your
attention
to
as
been
working
on.
12.9
is
1
0
for
2.
We
have
been
working
this
issue
for
few
release
milestones
it's
a
feature.
A
That's
had
a
high
level
of
interest
by
the
user
community
for
some
time
now,
as
it
addresses
a
key
use
case
related
to
service
discovery
and
network
communication
between
the
pen
and
connect
containers.
For
example,
you
know
the
job
and
services
service
container
today
we're
using
doctors,
doctors
link
flag
for
enabling
communication
between
containers.
So
not
only
is
that
a
dislocated
feature
one
of
his
limitations
is
that
only
allows
for
one-way
network
communication.
At
this
time
there
is
an
open
mood
request
that
once
completed
will
implement
a
capability.
A
That's
been
proposed
here
in
issue
1
0
for
2.
The
goal
is
for
us
to
finally
wrap
this
in
the
review
of
this
community
amount.
Specifically,
M
are
number
one
5
6
9,
so
we
can
ship
this
feature
in
12
that
9
the
nexus
you'll
enjoy
attention
to
that
we're
working
on
in
12.
That
9
is
issue
number
3,
3,
7
6.
This
is
actually
above
arm.
It's
the
long-running
jobs
can't
seem
to
get
you
I
again.
A
A
As
we
mentioned,
the
issue
description,
the
proposal
that's
being
implemented,
but
it
related
merger
quest
is
that,
instead
of
immediately
to
meeting
the
process
using
SiC,
kill
and
ignoring
the
result,
will
attempt
to
gracefully
shut
down
the
process
by
first
selling
Citroen
and
only
at
the
specific
to
specific
time
Magicians
expired.
10
minutes
will
be
then
cement
process
using
cycle.
So
three,
three
seven
six,
it's
pretty
close
to
getting
over
the
finish
line
as
we
have
an
active
military
class.
A
Our
goal
is
to
get
this
ship
in
twelve,
that
nine
they
are
the
future
that
we're
also
working
on.
That's
also
a
little
bit
from
some
from
twelve
that
eight
they'll
be
having
with
you
in
the
twelve
eight
months
and
then
quite
complete
is
issue
number
four,
nine
one
seven,
and
this
is
around
supporting
grace
for
shutdown
on
Windows,
as
I
described
in
a
previous
video.
A
We
are
implementing
in
many
a
method
for
emanating
sick
put
in
Windows,
and
so
when
this
quiz
for
shadow
feature
finding
ships,
the
wrong
will
wait
for
all
executed
jobs,
exit
before
the
run,
a
process
itself
so
mates
and
then
and
then
well
that
we
working
to
deliver
in
twelve.
That
nine
is
issue
number
of
four
four
five
zero
and
this
book
has
to
do
with.
Aaron
is
generated
that
FEMA
says
job
failure,
a
job
field
system
failure
and
no
such
container.
This
particular
bug
has
proven
to
be
quite
challenging
for
us
to
elaborate
reproduce.
A
The
public
port
indicates
that
the
error
occurs
randomly
and
only
for
approximately
ten
percent
of
the
jobs,
so
our
previous
investigation
and
for
those
of
you
that
have
been
following
can
see
the
history
of
the
investigation
in
the
actual
issue
itself.
I'm,
a
previous
investigation
into
this
file
included
releasing
a
patch
version,
the
runner
that
included
additionally,
both
statements,
the
goal
of
which
was
to
determine,
if
there's
anything
in
a
run,
a
code
that
removes
a
container
while
it
is
still
in
use.
A
We
didn't
find
anything
conclusive
after
reviewing
the
various
love
entries
submitted
by
the
user
community
after
they
implemented
or
tested
the
runaway
that
particular
patch.
Our
approach,
then,
for
twelve,
that
nine
is
to
implement
a
retry
mechanism
in
the
codebase.
This
basically
means
that
we
will
retry
each
request
sent
to
the
doctor
API,
where
a
response
code
of
four
four
is
returned
for
every
four
for
return.
It
will
append
all
of
the
container
Chinese
that
the
runner
has
and
all
the
container
a
means
that
the
container
demon
has.
A
So
then
this
will
also
save
a
snapshot
of
the
current
state.
If
the
retry
mechanism
does
not
completely
solve
the
bar,
then
the
additional
Long's
that
we
have
my
implementing
this
new
try
mechanism,
an
additional
lovin
capability
well,
hopefully,
could
find
us
to
T
that
we
need
to
determine
the
root
cause
then
come
up
with
a
final
fix
again.
If
this
new
solution
doesn't
in
fact
address
the
issue
conclusively.
A
So
with
this
feature,
he
uses
that
management
own
kidnap
she
run
a
fleet
on
AWS
cannot
auto
skate,
get
that
runs
on
AWS
foggy.
We
plan
to
get
rid
on
this
feature
and
future
releases,
so
we're
looking
forward
to
your
feedback
bonus
new
capability
I
mean
can
on
your
feedback
directly.
Here
is
a
comment
into
this
issue.
Awareness
issues
released.
You
can
add
additional
feedback
as
well
directly
as
as
previous
as
issues
in
get
lab,
work
and
get
level
on
the
project.