►
From YouTube: GitLab 15.2 Kickoff - Enablement:Database
Description
Kickoff for the Database Group for the GitLab 15.2 release
Planning issue: https://gitlab.com/gitlab-org/database-team/team-tasks/-/issues/256
Database Group Past Kickoff Videos: https://youtube.com/playlist?list=PL05JrBw4t0KqP3MYrcoQHrqPUqn_jJZSN
Presentation by: Yannis Roussos, Sr. Product Manager, Memory and Database Groups
A
The
core
thing
for
fallout
15.2
for
the
database
group
is
that
we're
planning
to
double
down
on
a
few
select
key
initiatives
for
the
availability
and
reliability
of
github
instances.
So
let
me
take
you
through
our
top
priorities.
The
first
one
are
bad
background
migrations.
This
is
the
new
framework
that
we
have
introduced
for
executing
very
large
data
updates
asynchronously.
A
So
our
top
priority
is
improving
the
patch
handling,
which
will
allow
us
to
execute
even
more
types
of
background
migrations
with
different
types
of
filtering,
as
well
as
a
few
stability
and
improvements.
Our
second
top
priority
is
the
throttling
mechanism
for
large
data
changes.
So
what
are
we
doing
in
this
initiative?
We
are
implementing
a
generic
throttle
mechanism
for
large
data
changes
that
will
allow
us
to
monitor
production
database
clusters
for
various
signals,
leading
indicators
and
react
in
real
time
to
problems
by
throttling
or
even
pausing
the
execution
of
the
updates.
A
We
consider
this
initiative
as
a
critical
update
for
preventing
high
severity
severity
incidents
on
production,
databases
and
production
systems
in
general.
Our
final
priority
is
the
peculiar
issue
we
have
found
with
the
database
load
balancer
gitla,
so
we
have
found
that
sometimes,
on
a
very
specific
edge
case
scenarios,
we
may
have
queries
leaking
across
transaction
boundaries.
A
What
does
that
mean?
So
I
assume
that
you
have
a
transaction.
The
transaction
have
has
a
lot
of
rights
or
updates
on
the
database
and
at
some
point
you
have
an
error
due
to
a
needling
transaction
timeout,
that's
a
very
special
case.
That
means
that,
for
whatever
reason,
you're
executing
updates
and
at
some
point,
the
transaction
stays
id
for
15
seconds,
let's
say
by
default
on
most
systems.
A
So
in
this
case,
for
example,
we
can
see
that
we
do
a
ride,
then
we
sleep
and
we
have
said
set
for
this
demo
dating
transaction
timeout
one.
Second,
at
that
point,
you
have
an
error.
The
transaction
action
is
rolled
back,
but
because
this
error
is
a
needling
transaction,
our
load
balancer
catches
that
error
and
retries
the
rest
of
the
transaction,
and
the
most
important
thing
is
that
it
drives
it
on
a
new
connection.
A
Rights
that
have
been
drawn
back
are
not
anymore
there,
so,
but
so
leaving
the
database
at
an
inconsistent
state.
This
is
a
very
rare
rare
scenario.
It
happens
very
few
times,
even
on
gitlab.com,
I'm
not
sure,
even
if
it
happens
on
other
systems.
So
in
15.1
we
have
added
login
for
those
types
of
events,
and
you
can
see
that
in
a
system
like
gitlab.com,
we
have
very
few
occurrence
of
this
problem,
so
in
15.2
we
want
to
validate
that.
A
Our
assumption
is
correct,
analyze
the
data
we
have
covered
for
the
past
few
weeks
and
then
apply
the
fix
that
we
assume
we
indicate
this
problem
and
keep
on
monitoring,
make
sure
that
this
hap
does
not
happen
anymore.
That's
it
for
15.2.
Thank
you
for
watching
and
talk
to
you
next
milestone.