background top icon
background center wave icon
background filled rhombus icon
background two lines icon
background stroke rhombus icon
header logo icon

Download "013. Школа менеджмента — Администрирование и эксплуатация в Яндексе. Вячеслав Агапов"

input logo icon
"videoThumbnail 013. Школа менеджмента — Администрирование и эксплуатация в Яндексе. Вячеслав Агапов
Similar videos from our catalog
|

Similar videos from our catalog

Video tags
|

Video tags

яндекс
yandex
Мобилизация
Школа менеджмента
Администрирование
эксплуатация
Вячеслав Агапов
Subtitles
|

Subtitles

subtitles menu arrow
  • ruRussian
Download
00:00:12
good evening everyone, my name is Vyacheslav,
00:00:15
today our lecture will be about
00:00:17
operation and we will go deeper into
00:00:20
operation, how it works in Yandex,
00:00:24
I work as a system administrator in
00:00:26
projects such as music, posters, I also
00:00:28
took part in projects such as photographs,
00:00:30
well, I talked to many people and
00:00:34
everyone there was an opinion that
00:00:37
exploitation is slavery not very
00:00:42
many people think that those who work in
00:00:44
exploitation are downright forced
00:00:45
laborers, but this is not so and we will talk about this a little
00:00:49
further, but now let’s
00:00:51
answer the question of what system administrators look like,
00:01:03
you are absolutely right, that is this is
00:01:07
such a stereotype that many people have and
00:01:09
everyone thinks that a beard and a sweater
00:01:11
means that the person is a system
00:01:12
administrator,
00:01:13
but unfortunately in our time or fortunately
00:01:16
it’s already difficult to say this is not so in our
00:01:21
time a person in a beard and a sweater is
00:01:25
understood as a hipster and now when
00:01:27
people see such a person I don’t immediately
00:01:28
think so hipster I don’t even think
00:01:30
system administration Lanter we
00:01:34
were joking let’s move on to what
00:01:37
we’re going to talk about today, we’ll
00:01:40
talk about who a wall
00:01:41
administrator is in general these days and
00:01:44
we’ll also see what they look like In fact,
00:01:45
everything that will be told is
00:01:47
structured and done this way in Yandex, but this
00:01:50
applies to all other companies, so
00:01:55
let's start what is exploitation in general
00:01:57
in our time in our time of
00:02:01
exploitation is the part of the company that is
00:02:04
responsible for these and what does it
00:02:06
mean that the people who
00:02:09
work they are engaged in supporting the
00:02:11
company's services, they are engaged in
00:02:14
optimal use,
00:02:16
searching for optimal use of
00:02:17
resources in any company
00:02:20
where there is an IT service, they all run on
00:02:22
hardware servers on computers, sometimes there are
00:02:25
too many computers and
00:02:27
in order to reduce them, the system
00:02:29
administrator is engaged in just
00:02:30
searching for how this can be done
00:02:34
among other things, wall
00:02:36
administrators are responsible for monitoring,
00:02:37
finding problems and eliminating them; as
00:02:42
monitoring, system administrators
00:02:43
are the first point of entry, that is,
00:02:46
if any problem occurs,
00:02:48
administrators are always notified about it; he
00:02:50
can already decide what to
00:02:52
do about it, or he can independently and and
00:02:54
completely subordinate or he is already calling the manager
00:02:56
to the developers and they make a
00:02:58
decision together now let's see who the
00:03:03
system administrators are in our
00:03:04
time, in fact, in fact, they are
00:03:06
ordinary people, they can be sitting among
00:03:08
you
00:03:10
and they look just like everyone else, that is
00:03:14
if you look left to right you
00:03:15
will see ordinary people, this person can
00:03:17
be a system administrator, well, he will become a
00:03:22
system administrator in the near
00:03:24
future, but time this used to have
00:03:27
stereotypes that system administrators are
00:03:29
only men,
00:03:30
but nowadays this is absolutely not
00:03:37
true, you and the narrator can also be them
00:03:39
and
00:03:40
but also girls in general, all system
00:03:52
administrators are inquisitive people,
00:03:53
they never go deep into one
00:03:56
specific problem, they always study
00:03:58
what is happening around them, they read all the
00:04:00
latest news, they follow the IT world in
00:04:02
general, the majority of them are system
00:04:06
administrators, extroverts, and
00:04:09
administrators very often have to
00:04:11
communicate with associates with managers with
00:04:13
developers with other colleagues if a
00:04:15
person is an introvert, he is very closed to himself, that’s
00:04:17
not where problems arise
00:04:19
after he sits buried in his
00:04:22
sandbox, he learns very little and he
00:04:25
is unable to exchange experiences among
00:04:27
everyone; they are also workaholics;
00:04:31
any system administrator who is
00:04:35
very he loves his job, he works
00:04:37
around the clock,
00:04:38
he is available 24/7, even when he
00:04:42
goes somewhere on vacation, he usually
00:04:43
takes a laptop with him, this is due not
00:04:46
only to the fact that he
00:04:49
absolutely loves his work with us, but also
00:04:51
because he is the first person
00:04:52
responsible for the service and problems
00:04:54
with the service it can happen at any
00:04:56
time when you are sleeping when you
00:04:58
are celebrating the New Year when we
00:05:00
hire employees, we usually
00:05:02
ask how a person generally feels
00:05:04
about night calls, that is, if a person is not
00:05:06
ready to answer at night at any time, then it is
00:05:10
very difficult with this person it’s starting
00:05:11
to work,
00:05:16
but now let’s talk about what in general we
00:05:19
want from managers as system
00:05:21
administrators and what it is generally needed. We
00:05:26
always want from managers to receive
00:05:27
information about major releases,
00:05:29
any change made, any
00:05:31
global change on a website in a mobile
00:05:33
application in anything, it raises a number of
00:05:36
problems that need to be solved
00:05:39
if we find out about these tasks already
00:05:42
during the launch, then problems usually happen,
00:05:45
we want to find out information about
00:05:50
new launches as a product, that
00:05:53
is, if we start something, come up with a
00:05:54
new mobile application, then it is advisable
00:05:59
for the manager to immediately come and
00:06:00
talk about it with the system
00:06:02
administrators even before he
00:06:04
communicated with the developers, since
00:06:06
system administrators have a very broad
00:06:08
outlook, they can immediately propose
00:06:11
some kind of architecture for this service,
00:06:13
and we also really want the system
00:06:18
administrator to have managers invite
00:06:20
us to all the lectures at all the meetings that
00:06:23
they hold. meetings with
00:06:25
developers with subcontractors with their
00:06:28
colleagues, managers, and others and
00:06:29
make some decisions during these
00:06:31
discussions, the other side can
00:06:33
offer technical solutions to
00:06:35
any problems and we begin to
00:06:38
say yes, of course, let us
00:06:39
do this in agreement, but then the manager comes to the
00:06:42
system administrator and tells
00:06:44
everything is said and the system administrator
00:06:45
says we can’t do this,
00:06:47
so we refuse and the manager does
00:06:50
n’t even think in advance that these are purely
00:06:52
technical ones that cannot be implemented,
00:06:55
this is mainly due to the fact that each
00:06:57
company has its own architecture
00:06:59
where all these applications are raised, this
00:07:01
can be purely it’s physically impossible
00:07:03
to do here, let’s talk a little about how to
00:07:09
plan a release in general, what’s good,
00:07:12
what’s bad, we’ll start with the bad,
00:07:16
very bad, this is a big release on Thursday
00:07:18
and any release on Friday, I don’t know who
00:07:21
has a suggestion why this is bad, yes, it’s
00:07:27
absolutely true if a big release is
00:07:30
rolled out on Thursday evening, a big
00:07:32
release can be understood as changes to the main
00:07:34
page of the site, which has the highest
00:07:36
traffic,
00:07:37
then if they change it, it will turn out
00:07:39
that from Friday everyone could
00:07:42
spend the entire weekend at work, this applies not
00:07:44
only to the system administrator, but
00:07:46
everyone will be involved, as well as the
00:07:48
developer manager system
00:07:49
administrators and heads of service
00:07:51
or service
00:07:54
to Friday this is the same and the
00:07:58
same thing is so terrible the worst is
00:08:01
anything on Friday evening, but what
00:08:04
good can managers do?
00:08:06
managers can warn in advance in
00:08:08
advance it all depends,
00:08:13
of course, on service, but it’s very good
00:08:14
when the manager has to be told
00:08:16
about a month in advance,
00:08:17
listen guys, we want to do this and
00:08:20
that, let’s think about how best to
00:08:22
do it, how to organize it, besides
00:08:25
this, it’s very best practice, it
00:08:29
rolls out to part of the audience those
00:08:30
changes that you want to see, well,
00:08:34
let’s also take an example is the main
00:08:35
page of any website, so there is not very
00:08:39
much traffic, and if you
00:08:42
immediately roll out big changes on the main page of the site,
00:08:44
they have changed everything in both functionality and
00:08:47
appearance, and if there is an error,
00:08:50
then one hundred percent of the audience will immediately see
00:08:53
this, it is
00:08:54
very undesirable to allow this, and therefore
00:08:57
the technique is used here that we roll out
00:08:59
only to part of the audience the
00:09:01
trick process can be divided over the whole day, that
00:09:03
is, managers with administrators or
00:09:06
with developers agree that we
00:09:08
start at 10 a.m.
00:09:09
10 a.m. everyone comes to work talking about
00:09:12
let’s show the new functionality to
00:09:13
ten percent of users, okay
00:09:15
turn it on after that we start
00:09:17
turn it on throughout the day and so on up to
00:09:19
100 percent, even if we see that
00:09:21
when turned on at 50 percent on 6
00:09:24
servers, the hardware cannot withstand it or
00:09:26
there is an error,
00:09:27
we can immediately roll it back to a smaller
00:09:29
number of audiences or even turn it off.
00:09:33
Another point is turning on by button
00:09:36
for managers. very relevant
00:09:39
because the manager usually
00:09:40
negotiates with all the partners
00:09:41
that the service has,
00:09:43
this could be an advertising partner, the
00:09:46
manager agreed with the partner that on
00:09:51
Saturday afternoon it is necessary to turn on a
00:09:53
commercial and since the system
00:09:56
administrator they really like to relax
00:09:58
on weekends they work, then of course they are
00:10:00
reluctant to do anything roll out if there is
00:10:04
such functionality as turning on by button,
00:10:05
then the manager can do this
00:10:06
on his own if the service is thought out
00:10:09
correctly and it allows it there anger
00:10:11
can wake up on Saturday afternoon and
00:10:13
turn on this ad block and then there
00:10:16
will be no problems,
00:10:20
let's also see what is needed to
00:10:23
launch how Do you think you can
00:10:25
calculate this formula?
00:10:34
No, of course, using this formula, we do
00:10:36
n’t estimate the size of the audience.
00:10:39
Let’s look at the example of
00:10:42
broadcasting concerts, how can you generally
00:10:44
estimate the audience? Any manager
00:10:47
should be able to do this in a good way, that is,
00:10:49
when a new service is launched,
00:10:51
it is necessary to know how many users
00:10:53
can maybe this may not be a single-digit
00:10:57
number, but I would like to have an idea
00:10:59
so that the manager has an idea of
00:11:01
what the order of users will be,
00:11:03
based on what the order of
00:11:05
the user will be, we can calculate the
00:11:06
load in these, let’s take the broadcast
00:11:10
of a concert, that is, we have a website on
00:11:13
which we know that we that all
00:11:16
users of the site are 50,000
00:11:17
people, and at the same time, we sent out a newsletter saying
00:11:20
that we will have such a concert,
00:11:22
well, we saw that yeah, we have 2,000
00:11:25
people signed up for this and I think
00:11:28
well, that is, the minimum number
00:11:29
expected by the manager is two thousand
00:11:31
people if the manager comes to us with
00:11:34
this information, we can calculate
00:11:35
what the load will be,
00:11:38
that is, it comes to us, they say in
00:11:40
such and such a site, such a broadcast,
00:11:42
we communicate with the developers, we find out
00:11:44
what the launch is,
00:11:45
what when the user presses the
00:11:47
play button to our server and only two
00:11:50
requests come Based on these two requests, we
00:11:53
can calculate the maximum load at a
00:11:55
single point in time, that is, imagine
00:11:57
that all fifty thousand people
00:11:59
went to the site at the same time because I don’t
00:12:01
know where YouTube was located, but they also pressed the
00:12:04
play button, while we know that 100 thousand requests will come to us
00:12:08
Based on this information, we will already be able to
00:12:10
prepare our servers in advance so
00:12:12
that such a flow of users
00:12:14
simply does not put it on us, and we
00:12:16
can calmly show this concert to everyone and everyone will be
00:12:18
satisfied and happy. The second
00:12:21
thing we need is a manager, a true
00:12:25
illustrator, we always need to negotiate,
00:12:26
asked Isla, this is possible To call it a
00:12:29
formal agreement on the availability of a
00:12:31
site
00:12:33
without this information is very difficult to evaluate at all,
00:12:35
but if there are any problems with
00:12:37
the site or
00:12:39
there are no problems at all, for example, let's look at
00:12:45
all these points that are
00:12:47
necessary first 1 of us in this one it says in
00:12:50
general what services should
00:12:52
work for us Let’s take a
00:12:55
mobile application or the main
00:12:58
page of a website should work.
00:12:59
What quality is quality? Here we
00:13:02
mean different metrics. The
00:13:05
main metrics that are important for the
00:13:07
system administrator
00:13:08
are the number of errors per site on the site
00:13:10
or in the application and also the time for
00:13:14
which this service is responsible, if we talk about
00:13:17
this in advance come to an agreement with the managers,
00:13:19
then it will be generally clear what is
00:13:21
considered good and what is considered bad; it is
00:13:25
also necessary to discuss the deadlines; the
00:13:28
deadlines here mean the deadlines for
00:13:29
fixing the problem; for example, if we
00:13:32
see that our site is starting
00:13:34
to respond or the application is located 300
00:13:37
milliseconds
00:13:38
one minute, it means that something has gone wrong
00:13:40
not so, accordingly, this needs to be dealt with right away
00:13:44
and if this is not the case, then the system
00:13:47
administrator can look at the schedule,
00:13:49
monitoring will call him, he will say yeah, he
00:13:51
answers in minutes, but the manager and I didn’t
00:13:52
agree on anything,
00:13:53
let me go and have some tea, well, here’s the manager
00:13:57
and of course he won’t like it very much all this
00:13:59
time to eliminate the problem, it may
00:14:01
change depending on what the
00:14:03
conditions are, that is, having agreed
00:14:06
on this for the first time in a month, this can be
00:14:09
discussed if it is clear
00:14:10
that they are too easy or too
00:14:12
complex and impossible to implement and, of course, who
00:14:15
is provided with this service, well, the main
00:14:17
consumers from of course, the user,
00:14:19
so everything is agreed on by the
00:14:22
users, what else is needed is
00:14:26
the ability to roll back if everything is
00:14:29
bad, even when we roll out a new
00:14:33
release, a new product to ten percent of
00:14:35
the audience, then we should always have
00:14:38
the opportunity to return this ten percent
00:14:40
back to 0 if we roll out at all
00:14:43
one hundred percent of the audience, then without
00:14:45
the possibility of a rollback, you can hang yourself immediately with
00:14:49
any even minor change that
00:14:51
is rolled out and brings an error with it,
00:14:54
if it cannot be rolled back, then it
00:14:56
can simply result in
00:14:58
man-hours; it can take a week to make;
00:15:00
it can take two weeks to make; so here
00:15:03
everything depends on the error itself if there is such an
00:15:05
opportunity, then it’s just
00:15:07
great, then everyone was calm,
00:15:09
managers and administrators and
00:15:10
developers, and the most important thing is that any
00:15:14
manager should be ready for the words there is no
00:15:16
system administrator of
00:15:18
any manager in their head they already have their own plans, they’re
00:15:21
just making plans for you, yeah, we’ll roll out
00:15:23
this functionality now, we’ll get a lot money and
00:15:25
everyone will be happy and healthy, but at the
00:15:28
same time they don’t even think about the technical
00:15:29
side of the issue and don’t calculate
00:15:31
ahead; in general, it’s possible to do this; it ca
00:15:34
n’t be done; well, it’s understandable for managers; such a
00:15:36
goal is to roll it all out and quickly, but
00:15:40
when the admins say the words no, the
00:15:43
managers get offended and go aside and
00:15:45
then don’t understand why they were
00:15:47
told no at all,
00:15:50
then we’ll look at what
00:15:52
managers will get from us if they
00:15:55
provide us with all this information in a timely manner,
00:15:56
the main one will not receive a
00:15:59
working service, that is, they came to
00:16:01
us and told us what they need, what they
00:16:03
want in what They were looking for deadlines for him, yes,
00:16:06
everything is up and running, they got
00:16:08
working service, everyone is happy, it works
00:16:11
24/7, it works even if
00:16:15
2000 people came to us, 50 thousand
00:16:18
people came, well, all users see the
00:16:23
answer in 300 milliseconds, or 500
00:16:26
managers are satisfied with this and everyone is happy
00:16:28
Well, the main thing is that the
00:16:30
manager will receive graphs, I don’t know
00:16:33
why, but many managers love
00:16:36
graphs, if managers have graphs, if
00:16:39
administrators don’t provide graphics for walls,
00:16:41
then they go to statistics, to analysts, and they
00:16:44
ask for graphs.
00:16:45
Well, on the one hand, it’s convenient, it’s
00:16:47
simple took and opened the monitor, there the laptop
00:16:50
turned on the monitor, looked at all the graphs
00:16:51
at once, everything is clear that the service in general
00:16:53
is happening, the service is working uninterruptedly,
00:16:59
uninterruptedly, it only works if, as
00:17:02
we said, there are no displays on Fridays and
00:17:04
Thursday evenings, well, accordingly,
00:17:06
everyone will have a good rest, everyone will be
00:17:08
happy, and at least there will be not only
00:17:09
administrators, but also managers and
00:17:11
developers, no one will
00:17:13
touch them either, all big releases, major ones such
00:17:19
as redesign of the necessary ones, not
00:17:21
only nodes at 10 percent each, well, it’s
00:17:24
also good to be able to roll them out in small
00:17:26
pieces, small pieces here
00:17:29
can mean different features,
00:17:31
this can be a change banners
00:17:33
changes buttons when such changes are
00:17:36
made minor, that is, the
00:17:38
designer came up with a new button
00:17:39
after that they placed it on the site, rolled out
00:17:42
only one button, everyone understands that this
00:17:44
change will not
00:17:46
cause any errors, but at the same time,
00:17:49
if along with the button there are also 10
00:17:51
changes in code of the application itself of the site of the
00:17:54
mobile version, it is quite likely that
00:17:57
this will have to be rolled back and then
00:17:59
designers and managers will not even see this
00:18:01
button that they have been waiting for so long when
00:18:07
launching a new service, it is always good
00:18:09
to plan for the resources that
00:18:12
will need to be spent on this; resources
00:18:15
here mean
00:18:16
hardware computers that is, here we would
00:18:19
here we think the system
00:18:21
administration side understands that any
00:18:23
service lives on computers and we can
00:18:26
say to the manager, yeah, okay, we need
00:18:28
one computer, but if they
00:18:30
provided us with all the necessary information,
00:18:31
then we can calculate in general how many
00:18:33
computers are needed for
00:18:35
our the service has withstood 50-100 thousand requests
00:18:38
or more, in order to do this we
00:18:42
carry out load testing
00:18:45
in Yandex, there is a development
00:18:48
called index tank, this development is
00:18:50
open, it is available to everyone,
00:18:52
with the help of it we can generally calculate how
00:18:55
many requests
00:18:56
our site will withstand
00:19:02
launching one of the services that
00:19:04
allows you to download music before
00:19:07
we launched it,
00:19:09
we didn’t even know at all how much this
00:19:11
service could withstand such a load,
00:19:13
here the index tank just came to our aid,
00:19:15
that is, we were able to estimate
00:19:18
how much what maximum number of
00:19:19
people can we upload music and
00:19:23
after that we were able to calculate
00:19:26
by shooting how many servers we need in general
00:19:28
for this and even if
00:19:30
our load increases it might look like
00:19:35
this here it turns out on the x axis we have
00:19:39
time on the y axis we have a load if
00:19:42
the service of good ones is well planned then it
00:19:44
can withstand a 4-fold increase in load
00:19:46
because such a situation can
00:19:48
happen at any time,
00:19:51
there were cases that there was some kind of
00:19:56
holiday and new information blocks appeared on the sites
00:19:59
which led to
00:20:01
more users coming to us
00:20:02
and any service should be
00:20:05
ready to such a load,
00:20:10
in order for everything to be good and everything
00:20:13
to work, you must always try to
00:20:15
find bottlenecks and optimize them so
00:20:18
that these bottlenecks can be
00:20:21
eliminated in the end, a good service
00:20:24
should be fault-tolerant; if there is a
00:20:26
bottleneck, then the service will live very poorly
00:20:29
in production;
00:20:31
one of the ways is this canopy search method
00:20:34
what it means is that the favorite
00:20:38
place of the system
00:20:39
administrator
00:20:40
is in the console, as you can see, here is a black
00:20:43
screen with white letters, someone
00:20:45
likes green letters,
00:20:46
and system administrators
00:20:49
just spend hours on end in this console, from
00:20:52
this console they get a lot of useful information
00:20:53
information, for example, a canopy can
00:20:55
determine why we generally
00:20:58
respond very slowly, that is, in this case
00:21:00
it is clear that here a database request is shown
00:21:03
and its execution time is three
00:21:04
seconds for a site that wants to
00:21:08
respond quickly three seconds is a lot and
00:21:11
we all find such places and try to
00:21:13
fix them the second tool is graphs
00:21:18
when the administrator has graphs set up at the walls,
00:21:22
he has to sit in
00:21:24
consoles less because he doesn’t
00:21:25
need it, he can also, as a manager,
00:21:27
open the monitor and look at all these
00:21:29
graphs, so when I said that
00:21:31
managers love graphs, I probably forgot
00:21:34
to say that the system administrators
00:21:35
also love them so that you do
00:21:37
n’t have to sit in the console for hours on end, just
00:21:39
make a few graphs on which
00:21:42
it will be clear what’s going on with
00:21:43
the service in general, it works, it doesn’t work after
00:21:48
we found the bottleneck, we saw
00:21:50
that requests are taking a long time to complete,
00:21:52
that there is a problem on the servers, we are
00:21:55
looking into them optimization as
00:22:00
optimization, there are several things that
00:22:04
we carry out, the first one is just an
00:22:07
adjustment to the database, that is, we
00:22:09
take queries in general and study what it
00:22:11
does; developers sometimes when they
00:22:15
write code they simply do not notice how to
00:22:18
work correctly with the database in general
00:22:20
they need to get all the information about
00:22:22
users who have information
00:22:24
birthdays information about yes circulation
00:22:27
birthdays information about em phone
00:22:31
number last name that is a lot of information and
00:22:34
if the Beretta developer just talks
00:22:37
about give me information give me only the
00:22:39
phone number for this user he can
00:22:40
accidentally take and get
00:22:42
all the data from the database and get all the data in general is
00:22:45
very expensive then we’re just with the
00:22:47
sea we’re talking
00:22:48
guys let’s fix it here so it’s
00:22:50
good the second is optimizing the
00:22:54
application itself in general if the manager and
00:22:59
not even themselves have no idea what
00:23:01
exactly will happen what
00:23:03
the service itself will look like, he can come and
00:23:05
assign a task to the developers to write
00:23:07
this and that developer will take it and say it’s
00:23:09
good and in the evening he will write you a
00:23:11
service that will be ready by the
00:23:13
next morning the manager will come
00:23:15
happy and say it’s okay
00:23:16
to launch, but at the same time when it’s the
00:23:21
application itself 10 thousand requests will come to the site itself,
00:23:24
then the manager, system administrator and
00:23:28
developer will understand that we have only one
00:23:30
computer and it can’t handle it, let’s
00:23:31
add 2
00:23:32
but at the same time disassembled the barrels, they will say, but we
00:23:34
can’t do this now because the
00:23:37
developer simply didn’t think about what will happen
00:23:39
in the future it is necessary to scale it up,
00:23:41
but then you will need to take and
00:23:44
write the entire application from scratch. Well, here’s another
00:23:49
optimization point like this: replacing
00:23:52
all server equipment in general; over the
00:23:55
past 10 years,
00:23:56
the leap in server development has been very
00:23:59
large.
00:24:00
When I came to Yandex, it was about
00:24:02
six years ago.
00:24:06
contained 16 gigabytes of memory it was very
00:24:09
cool all the time the system administrator and were
00:24:11
just happy to see such a server
00:24:13
server nowadays servers can already
00:24:16
contain 256 or 500 16 gigabytes of
00:24:19
memory that is, an increase of up to more than 8 times in
00:24:23
memory, this also applies to others
00:24:25
resources, if previously the service
00:24:28
worked well on two computers
00:24:30
that were purchased 10 years ago, then
00:24:33
if these services are launched on
00:24:34
computers that were purchased
00:24:37
today, they can simply
00:24:39
start working twice as fast on their own and
00:24:41
then no one will need to do anything at all,
00:24:43
neither a manager nor and for
00:24:44
developers who are not administrators, from the
00:24:50
graphs that we build, we immediately
00:24:53
notice such optimizations for services
00:24:57
that struggle for speed are very
00:25:00
relevant, even 56 gains of 50 100
00:25:03
milliseconds, and when we just see on the
00:25:06
graphs that such a gain occurred due
00:25:08
to the fact that we optimized the code and
00:25:10
replaced the servers, everything remains very
00:25:13
happy, here are examples of such graphs,
00:25:17
just the one you see; managers love using
00:25:21
this graph; managers can generally
00:25:23
estimate what kind of load on the service generally
00:25:25
comes; there are problems in general; when it
00:25:29
is displayed with an alley for the service, it is
00:25:32
usually also marked on the graphs;
00:25:34
which was visible without knowledge of the console; generally
00:25:37
exceeded whether we are this ass or not and
00:25:40
whether it is worth responding to the problem in
00:25:42
this case, the graph in red
00:25:43
refers to the excess of the
00:25:45
slug from one of the agreements with
00:25:48
managers,
00:25:52
but now let’s go deeper into
00:25:54
what administrators and
00:25:57
technical specialists do in general; they are divided into 2
00:26:00
types: there are system administrators are
00:26:02
developers, and in the past, in
00:26:06
some companies there is always such a
00:26:08
division into such specialists, and by
00:26:11
this division it is understood that
00:26:12
system administrators are the ones who
00:26:14
support the service, that is, they ensure
00:26:16
its functionality in answering calls
00:26:18
and monitoring, but that is, the developers are
00:26:21
these developers, they write the code itself and everyone is
00:26:25
no longer responsible for anything,
00:26:27
as they said here about devops of yes,
00:26:30
but there is such a concept and it is now
00:26:33
developing in many companies.
00:26:36
Nowadays, such concepts as devops
00:26:38
and raw materials have appeared about devops, they said that this is the
00:26:41
one who still deals with the heat, it’s
00:26:44
correct and professor, does anyone know
00:26:46
anything what it is, but one thing is Sergey,
00:26:54
he is also a technical specialist,
00:26:56
these are service website accessibility engineers,
00:27:00
but
00:27:02
this is also the concept that they went for, they slightly
00:27:04
combined administrators and
00:27:06
developers, that
00:27:08
is, by devops, you can understand the
00:27:10
person who writes the code and
00:27:12
installs it on the server itself, you don’t
00:27:14
need an administrator here,
00:27:16
but for this you need a good architecture, a
00:27:19
good service so that a developer can do this,
00:27:22
in this case serge appears,
00:27:24
that is, the son is already starting to write
00:27:27
code, write code inside which
00:27:29
helps developers launch their
00:27:31
applications after it appears
00:27:35
such a union then devops
00:27:37
from Irgi, the two of them become the first
00:27:39
responsible, that is, the boundary between
00:27:42
developers and administrators is
00:27:45
blurred, that is, if before it
00:27:47
was possible, if there is an administrator and
00:27:49
developers, you can come to the
00:27:51
administrator as the first entry point
00:27:53
that the service is not working and it would be bad for him but
00:27:56
when devops appear, the
00:27:57
administrators are already absent in fact,
00:28:00
and then the manager immediately comes directly to all the questions
00:28:02
where, in general, let’s
00:28:07
see what an
00:28:09
application looks like on a computer, an application,
00:28:13
it could be a website, it could be a
00:28:14
mobile application, that is, here it’s
00:28:18
not very important, well, applications
00:28:22
run on a server,
00:28:23
usually there are several such applications on one server,
00:28:27
and in order to generally understand what is
00:28:30
happening with the application, you need to
00:28:32
understand what resources the server has. In
00:28:38
this case, the server has
00:28:40
resources such as disk memory, network, each of these
00:28:44
resources is very important if we crash
00:28:48
when building a disk, then accordingly the computer
00:28:50
will stop working altogether, and if
00:28:52
applications are running in just one of their
00:28:55
samplers on one computer, then this
00:28:56
will lead to a complete collapse, in addition to the fact that there
00:29:01
is always one computer, there are several
00:29:04
computers in the brain for one computer service, the
00:29:08
number of computers can reach up to
00:29:10
hundreds,
00:29:11
that is, if the application is very highly
00:29:15
loaded, it is already optimized, how
00:29:17
can you ruin the most expensive good
00:29:20
servers, then scaling is
00:29:22
only possible by increasing the number of computers, all these
00:29:26
computers are installed in a data center, a
00:29:29
data center, this is great and great, a barn
00:29:33
on and a room where it’s just rida minus
00:29:35
these stamped ones computers work
00:29:38
around the clock, there is constant
00:29:41
cooling and all interaction
00:29:45
takes place inside,
00:29:46
but when we have only one data center,
00:29:50
various problems can happen.
00:29:52
I think that many have even heard that
00:29:55
some
00:29:57
data center stopped working due to the fact that a faucet
00:29:59
cut a wire. broke, but in many
00:30:04
places there is simply no mark that there
00:30:07
is a wire underground that supplies electricity
00:30:09
or the Internet to this data center, if
00:30:12
any construction work is being carried out,
00:30:14
then very easily a crane can simply
00:30:16
cut this wire, well, the time to
00:30:19
eliminate such a problem is
00:30:20
significant and therefore that you need to
00:30:23
first find the place where it happened, then
00:30:25
connect it all, so for good
00:30:27
architecture they always use several
00:30:29
data centers where the computer
00:30:31
interacts with each other and the loss of
00:30:33
one data center is no longer relevant,
00:30:36
we have already lost one that
00:30:38
centers have lost we have a second we
00:30:40
works well therefore, when
00:30:43
designing the architecture of any
00:30:46
service, it is necessary to take into account that
00:30:48
if we use two data centers,
00:30:51
we must be able to withstand the entire load with
00:30:52
one data center because accidents
00:30:55
happen quite often,
00:30:59
in addition to the fact that we roll out the code, we
00:31:03
also monitor the relevance of
00:31:04
security in full faith in many
00:31:07
companies there are separate services that
00:31:09
deal with this, an information
00:31:11
security service that looks at whether everything is
00:31:13
fine in the global sense of the word, but in
00:31:16
each service
00:31:17
there are components that the
00:31:19
administrator can choose
00:31:20
independently, since he chooses it
00:31:22
independently, then
00:31:24
the responsibility for the fact that this product falls on his shoulders
00:31:26
it is safe
00:31:29
to install any application here a year ago, it
00:31:32
may happen that it
00:31:36
has never been updated, but if a
00:31:37
new version appears, it is
00:31:39
necessary to update it, that’s why
00:31:42
administrators
00:31:45
are also responsible for ensuring that all programs are up to
00:31:47
date, well, they are
00:31:51
monitoring the vulnerability if this do not
00:31:54
do that in those companies where there is an
00:31:55
information security service, they can
00:32:00
notice any problem before system administrators,
00:32:02
come to you, but this is not always
00:32:04
pleasant, like you cancel the service, it’s
00:32:06
yours, everything is fine, but someone else generally
00:32:08
finds the problems that you are talking about I should have been the
00:32:10
first to know when the
00:32:16
concept of how and raw materials appeared, it was very, very
00:32:18
clear that automation of processes came to the forefront,
00:32:22
all the admins are
00:32:25
very lazy, and
00:32:26
that is, if a task needs to be performed
00:32:29
more than once, it already becomes
00:32:31
tedious,
00:32:32
any new task is directly interesting, the
00:32:34
eyes are burning, the eyes want doing it is all
00:32:37
great, but as soon as this task
00:32:39
is repeated 2 1 3 1 4 1 they just get discouraged
00:32:43
and they all don’t want to do this,
00:32:45
and here they immediately start
00:32:47
thinking about how we can even
00:32:49
automate this; this applies to
00:32:52
server maintenance, which is the same
00:32:55
example with a bad disk, if there is a bad disk on the server,
00:32:58
then you need to somehow
00:32:59
find it because usually there are
00:33:02
more than one of them, but to do this manually is
00:33:04
to determine which disk is broken,
00:33:07
why it broke, it’s very long and
00:33:09
tedious, so after doing this operation
00:33:11
5 times, the administrator says no I don’t
00:33:14
want to do this by hand anymore, let’s
00:33:16
do something and he’s just
00:33:17
starting to write code, he’s writing code that
00:33:20
will help him in the future not have to deal with this at all
00:33:22
when working with databases,
00:33:28
any database requires
00:33:30
maintenance,
00:33:31
roughly speaking, if the database there was a
00:33:33
problem with the data and it is necessary to fill it completely with Perry,
00:33:35
that is, fill it with new data.
00:33:37
This can happen due to problems with
00:33:40
data centers very often, that is, our
00:33:43
data center broke down
00:33:44
or the computer, after that everything
00:33:46
was restored and we need to somehow
00:33:47
restore all the programs that
00:33:49
they worked there, do it manually, it’s just
00:33:52
about when you have 100 computers in a
00:33:55
data center, the data center came out, I’ll arrange for you
00:33:57
these 100 to 100 computers, you need to lift them
00:33:59
in another place with the hands of one person,
00:34:02
this of course can be done, well, a person
00:34:03
shouldn’t sleep at all, that is, he should
00:34:06
work around the clock,
00:34:07
well, that’s why these places are all
00:34:10
automated, automatic layouts,
00:34:14
they help here a little by the
00:34:17
administrator himself, as much as by the manager, if
00:34:21
earlier it was necessary to upload
00:34:22
new code to the server, help from the
00:34:24
administrator, then when an
00:34:27
automatic layout appears, which is
00:34:28
configured correctly, he
00:34:30
is able to ensure that after the layout there are
00:34:32
no errors that everything is fine if
00:34:35
monitoring is configured, then this allows
00:34:37
administrators to get away from this,
00:34:39
when they get away from this, the
00:34:41
manager himself says, yeah, here we have new
00:34:43
functionality, let's go, we're rolling it out, but
00:34:47
this moment is very dangerous for
00:34:49
any manager, manager, developer,
00:34:52
system administrator,
00:34:54
this can be attributed to he can
00:34:57
say to everyone, yeah Friday evening there is a chance to
00:34:59
roll out something automatically, let me
00:35:01
do it without informing anyone,
00:35:04
so it’s very good when the teams
00:35:08
communicate with each other constantly, when the manager, the
00:35:12
administrator’s developers don’t know how to
00:35:14
agree on what everyone likes, what both
00:35:16
the manager and the manager like the system
00:35:18
administrator,
00:35:19
if this harmony is found, then there are
00:35:22
no problems with this at all, then the
00:35:24
manager understands that it is impossible to roll out in the evening,
00:35:26
but we can roll out tomorrow
00:35:28
morning and everything will be fine, but just
00:35:33
in addition to the
00:35:34
automation stage, there are other pieces of
00:35:37
infrastructure that administrators and
00:35:39
raw materials will call it so they write
00:35:42
on their own, here we talked
00:35:47
in general about what an administrator does globally,
00:35:49
let’s talk a little
00:35:50
about how data centers are structured in
00:35:52
Yandex, but every company that
00:35:56
has a highly loaded service
00:35:57
uses not one, but several
00:35:58
data centers. The picture shows how the
00:36:02
data centers are structured. the centers inside them are
00:36:04
just in these corridors, it’s very cold and
00:36:06
that’s why the system administrators wear
00:36:08
sweaters,
00:36:10
but if you walk there without a sweater, well, I don’t
00:36:13
know, you can get sick in one day,
00:36:16
probably, well, it’s not clear why the
00:36:19
administrators had a beard, it doesn’t seem to
00:36:22
warm the grounding,
00:36:25
but that’s also an option Yandex
00:36:31
currently has 6 data centers, but we realized that 6
00:36:35
to the centers is a lot and now we are
00:36:38
moving towards the fact that we will have only 3 data
00:36:40
centers, but these three to the center will be our own,
00:36:43
well, here is one of the data centers that are
00:36:46
built in the index it is located
00:36:50
abroad and our architects, we can
00:36:55
say so, did not agree so that our
00:36:56
data center will heat the village
00:36:58
next to which it is located, any
00:37:02
heat needs to be removed to the center and so they
00:37:04
will supply this energy to the village,
00:37:06
that is, the immediate advantages are, well, in
00:37:08
another data center in Russia
00:37:11
that
00:37:13
our architects built, they would also be able to
00:37:16
prove that there is nothing to burn inside where the servers are located,
00:37:19
so there is no
00:37:22
fire safety there, well, that’s all the experience, if
00:37:26
not the experience, and when in data centers in
00:37:29
queues there
00:37:30
really were fires like these mini that
00:37:33
is, the power supplies and
00:37:36
motherboards were on fire, but due to
00:37:39
this we were able to prove something I need a
00:37:40
fire extinguishing system in Yandex now there are
00:37:46
more than 100 servers, the
00:37:48
servers of this are large plates where
00:37:50
all the necessary components for work are left.
00:37:54
In addition to servers, data centers also have
00:37:57
disk shelves disk shelves are
00:38:00
large boxes that can
00:38:02
contain up to 70 disks,
00:38:05
if you think about it this way, the
00:38:08
weight can reach 300-400 kilograms. The
00:38:12
rule is that these disk
00:38:14
shelves are always placed below, closer to the
00:38:17
ground, so that they can be easily pulled out
00:38:19
and the transponder transported, but sometimes
00:38:23
I I don’t know why it’s possible that engineers are having so much
00:38:26
fun with
00:38:27
these disk shelves with 70 disks they are placed
00:38:30
at the very top and when you have to go and get the
00:38:32
disk shelves out, it’s simply not
00:38:35
realistic to do this, but
00:38:37
five or six people gather there and the six of them
00:38:39
try to lower it down, but some
00:38:43
shelves are left in place and they climb to the
00:38:44
very top to replace all these
00:38:46
disks,
00:38:48
but you also, in addition to the servers themselves and the
00:38:51
canopy, have all the network equipment in
00:38:54
our data centers and each of our data
00:38:57
centers has two channels
00:38:59
providing the Internet due to the fact
00:39:02
that the crane and excavators work well everywhere,
00:39:04
but very often one wire is
00:39:07
constantly interrupted then we work
00:39:09
on the second wire everything is becoming good
00:39:13
also all over the world there is now a trend
00:39:16
to switch to ipv6
00:39:19
we have heard about it no but what is the connection
00:39:23
you understand
00:39:28
yes it is absolutely true that the capacity is all
00:39:31
gone and it turned out that
00:39:33
now the
00:39:35
authorities that control they
00:39:37
simply stopped issuing them at all, so
00:39:40
this information has already arrived well, it has spread
00:39:43
around the world a very long time ago, then all of our
00:39:45
data centers they and our network
00:39:48
engineers are designed only for
00:39:50
use in the beer 6 industry environment,
00:39:53
that is, there are no IP addresses in the fourth addresses
00:39:55
at all, if so count all the thousands of
00:39:59
computers they work on 6 0 and in
00:40:04
each of our centers there are
00:40:06
more than 5000 units of network equipment installed,
00:40:12
let’s now talk about the
00:40:15
service itself, standard developers, they
00:40:20
more or less understand what
00:40:22
a service can look like when they start
00:40:25
making it from scratch service, they don’t have a
00:40:28
completely complete picture of it, here’s
00:40:32
how it can always look for the user,
00:40:36
and well, let’s just say there
00:40:40
are different developers, there are frontend developers,
00:40:43
there are backend developers, and there are full
00:40:45
stack developers, but if a person
00:40:48
is a frontend developer, then he
00:40:51
sees only the
00:40:52
entire external view, that is, he
00:40:54
develops a website page,
00:40:56
but either a page of a mobile
00:41:00
application and the backend developer
00:41:02
gets everything else, all those subsoil and
00:41:06
which are all that is buried there in the subsoil,
00:41:10
but in the general case, this can divide and
00:41:13
highlight such entities as front and
00:41:16
back and and backend already interact
00:41:19
with all databases with storages
00:41:21
and with all third-party services, that
00:41:23
is, the frontend doesn’t do this for
00:41:27
him, it doesn’t matter to him, the main thing is to
00:41:28
go somewhere to get the data, then draw it all
00:41:30
beautifully, if this is done,
00:41:34
why is
00:41:36
this done, in order to unload the
00:41:38
components
00:41:39
if do everything together front-end and
00:41:41
back-end everything is in a heap then such an application
00:41:44
must live on one computer
00:41:46
one computer it has limited
00:41:48
performance, so if
00:41:51
we have different pieces of code then we can
00:41:54
place these different parts on different
00:41:55
computers and then not think about it about how
00:41:57
we can scale this for the frontend in
00:42:04
Yandex, well, in terms of Yandex, such decisions have been made
00:42:07
that it is necessary to use but
00:42:09
Jess,
00:42:10
this decision was made due to the fact that
00:42:12
but Jess scales very well and it’s
00:42:14
enough to just start writing on it, that is, you don’t
00:42:17
have to study tons
00:42:20
literature for brands, we
00:42:25
use C plus plus
00:42:29
python java, many have probably heard
00:42:31
such words, but one thing if you want
00:42:34
in detail, it’s only on the Internet,
00:42:37
so we won’t go deeper, otherwise you can
00:42:40
just dig into these languages ​​and they were
00:42:42
also chosen due to the fact that they are
00:42:45
easily scalable
00:42:47
The service written in these languages
00:42:49
scales very well; they can be run
00:42:52
on one computer or on hundreds of
00:42:54
computers; they will continue
00:42:55
to work.
00:43:12
data, it can be pictures, it can
00:43:15
be music, that is, you
00:43:17
upload it once and then use it as much as you want
00:43:19
without touching the database at all.
00:43:22
Databases are smaller
00:43:26
entities, that is, they can
00:43:29
also be located on the same shelves
00:43:31
that contain just bytes of data, but
00:43:33
usually they are small in size. they
00:43:36
are made small in size in order
00:43:38
to quickly get data from there; this
00:43:40
data is ordinary changeable, let’s say it is
00:43:42
extremely possible to store information about a
00:43:44
person about a client,
00:43:46
and at the same time, if you made
00:43:48
any mistake in his name or
00:43:50
date of birth when saving, then this can easily be
00:43:53
corrected with this there
00:43:55
will be no problems, but this is a standard service,
00:44:01
it is assembled from such
00:44:03
pieces as front and back and databases and
00:44:06
storage, well, the usual from any
00:44:09
administrator, he selects only those
00:44:11
technologies that are familiar to him; any
00:44:14
administrator
00:44:15
loves new things, but this new is necessary
00:44:18
if he wants to launch something new, it
00:44:20
needs to be studied and implemented for a very
00:44:22
long time, and it turns out that the solution
00:44:24
that he can find on the Internet, they
00:44:26
simply may not be suitable,
00:44:28
there have been cases when a person came, he
00:44:31
says, here we want this is a new
00:44:33
solution, fashionable youth modern,
00:44:35
let's give it to us Let’s do it, well, the developers
00:44:38
thought, the managers thought, well, let’s, but
00:44:40
then a month passed and everyone understood this
00:44:43
decision, it’s good, but when you have
00:44:45
ten people on the site,
00:44:46
and when you have 100,000 people on the site,
00:44:49
then this solution simply
00:44:50
doesn’t work, well, the connections with this are all new and
00:44:54
fashionable any administrator likes the technologies,
00:44:56
but
00:44:58
launching them into production is very scary,
00:45:00
so all the new services in companies
00:45:04
that have already paved the way for
00:45:06
themselves, they have already released 23 applications, they
00:45:09
launch on the same technologies
00:45:11
because they are already familiar, it’s good to
00:45:13
operate, develop, monitor
00:45:15
them and there is such a question in general
00:45:20
when to come to the admins who can
00:45:24
answer
00:45:29
here another such formula is
00:45:34
necessary to come in advance
00:45:36
but this word in advance it is very
00:45:39
vague but all managers are
00:45:44
all developers administrators they
00:45:46
want to receive well,
00:45:47
any person wants to find out in advance
00:45:49
about any change, let’s say if
00:45:51
it starts a new service, then you
00:45:53
need to come a month in advance, two months in
00:45:56
half a year, that is, the manager found out x the new
00:45:59
service, let's launch it and everything is fine, it
00:46:01
immediately comes to the developers with the
00:46:04
administrators,
00:46:05
but tell all the wonderful things at this
00:46:08
moment the administrator can already start
00:46:10
thinking about the architecture,
00:46:12
but if the manager comes for two days before
00:46:15
the launch and he will say, we
00:46:17
have already written a service here with the developers and tomorrow we
00:46:19
want to launch it, well, if they persuade the administrators from the manager, they
00:46:22
will launch it,
00:46:24
but in a day it will all fall down, well, that’s why it’s
00:46:27
better to come here in advance,
00:46:29
even before the developers have even talked to the
00:46:33
developers really, everyone is
00:46:36
cool, but the developers go deep into
00:46:39
their own area, for example, you have a
00:46:42
division between front-end and back-end,
00:46:44
then the developer looks at the service of the
00:46:46
front-end country, he is great at this, he
00:46:49
generally understands it, you know, but he doesn’t
00:46:51
see what’s going on behind the front end, the same
00:46:54
thing backend that is backend he can
00:46:57
write perfectly everything will work
00:46:58
quickly great but he doesn’t see how the
00:47:02
database works or how the
00:47:04
storage works and he can’t estimate what will happen in a
00:47:08
year in this case if he
00:47:12
comes to the administrator in advance the
00:47:14
administrator usually sees all this
00:47:16
from above he generally sees the service from all
00:47:19
sides, that is, he knows which frontend
00:47:21
is used, which backend, which
00:47:23
databases and he knows the principle of their operation,
00:47:26
based on this, the system administrator
00:47:28
can just look at it and say,
00:47:30
yeah, here’s a new service, well, let’s launch everything, it
00:47:34
will withstand and then there will be no there
00:47:35
won’t be any problems,
00:47:37
we talked about it at all, but I told them
00:47:42
how
00:47:45
our time looks on the administrator and what stereotypes
00:47:47
people have about them, what information do we
00:47:50
generally want from managers and why is it that after
00:47:52
receiving this information we
00:47:56
provide something in response as it
00:47:59
would be stupid if the manager came to us
00:48:01
in advance, he told us all this, he
00:48:03
gave it all, the illustrator would look at it, he would say,
00:48:06
okay, you tell us this further,
00:48:08
well, that’s it, we also talked
00:48:13
about the essence of what the
00:48:15
administrator does and what to do, so-
00:48:18
called energy, that’s about and
00:48:22
Nowadays, there are a lot of raw materials and devops, but there is a
00:48:24
lot of talk,
00:48:25
but no one can give an exact definition,
00:48:27
that is, this is such a vague
00:48:30
concept that in every company for half a year
00:48:33
Sergei devops, it is understood differently,
00:48:35
but we talked a little about what
00:48:38
data centers are in general, what they are arranged in
00:48:41
Yandex, well, we also dismantled the standard
00:48:45
service,
00:48:46
that is, the standard service is the one
00:48:49
we looked at, it is accepted, well, I
00:48:52
probably don’t know half of the places in Yandex,
00:48:55
but of course this should not be the case for everyone,
00:48:57
every developer, every administrator,
00:49:00
he is free to choose himself, the same
00:49:02
manager can see new words
00:49:05
come and say guys, let’s
00:49:07
do this, as you found out, we read somewhere, we
00:49:09
looked at these, we thought, well, let’s
00:49:11
try, that is, a dialogue between
00:49:15
managers, developers,
00:49:17
administrators is one of the main
00:49:19
things that should be in general, if
00:49:21
this dialogue is not there, then the work will be
00:49:25
idle, the manager will inform
00:49:28
in advance administrators will not
00:49:29
do this because everyone has their own
00:49:33
character and behavior and if there is no
00:49:36
communication then all the work will stop but
00:49:42
that’s all thanks to the legacy
00:49:44
Boston
00:49:45
[applause]

Description:

Если системный администратор отлично знает свою систему и правильно всё настроил, ему остаётся только сидеть и наблюдать, как работает подопечный сервис. Рассказ о том, как устроена эксплуатация в Яндексе.

Preparing download options

popular icon
Popular
hd icon
HD video
audio icon
Only sound
total icon
All
* — If the video is playing in a new tab, go to it, then right-click on the video and select "Save video as..."
** — Link intended for online playback in specialized players

Questions about downloading video

mobile menu iconHow can I download "013. Школа менеджмента — Администрирование и эксплуатация в Яндексе. Вячеслав Агапов" video?mobile menu icon

  • http://unidownloader.com/ website is the best way to download a video or a separate audio track if you want to do without installing programs and extensions.

  • The UDL Helper extension is a convenient button that is seamlessly integrated into YouTube, Instagram and OK.ru sites for fast content download.

  • UDL Client program (for Windows) is the most powerful solution that supports more than 900 websites, social networks and video hosting sites, as well as any video quality that is available in the source.

  • UDL Lite is a really convenient way to access a website from your mobile device. With its help, you can easily download videos directly to your smartphone.

mobile menu iconWhich format of "013. Школа менеджмента — Администрирование и эксплуатация в Яндексе. Вячеслав Агапов" video should I choose?mobile menu icon

  • The best quality formats are FullHD (1080p), 2K (1440p), 4K (2160p) and 8K (4320p). The higher the resolution of your screen, the higher the video quality should be. However, there are other factors to consider: download speed, amount of free space, and device performance during playback.

mobile menu iconWhy does my computer freeze when loading a "013. Школа менеджмента — Администрирование и эксплуатация в Яндексе. Вячеслав Агапов" video?mobile menu icon

  • The browser/computer should not freeze completely! If this happens, please report it with a link to the video. Sometimes videos cannot be downloaded directly in a suitable format, so we have added the ability to convert the file to the desired format. In some cases, this process may actively use computer resources.

mobile menu iconHow can I download "013. Школа менеджмента — Администрирование и эксплуатация в Яндексе. Вячеслав Агапов" video to my phone?mobile menu icon

  • You can download a video to your smartphone using the website or the PWA application UDL Lite. It is also possible to send a download link via QR code using the UDL Helper extension.

mobile menu iconHow can I download an audio track (music) to MP3 "013. Школа менеджмента — Администрирование и эксплуатация в Яндексе. Вячеслав Агапов"?mobile menu icon

  • The most convenient way is to use the UDL Client program, which supports converting video to MP3 format. In some cases, MP3 can also be downloaded through the UDL Helper extension.

mobile menu iconHow can I save a frame from a video "013. Школа менеджмента — Администрирование и эксплуатация в Яндексе. Вячеслав Агапов"?mobile menu icon

  • This feature is available in the UDL Helper extension. Make sure that "Show the video snapshot button" is checked in the settings. A camera icon should appear in the lower right corner of the player to the left of the "Settings" icon. When you click on it, the current frame from the video will be saved to your computer in JPEG format.

mobile menu iconWhat's the price of all this stuff?mobile menu icon

  • It costs nothing. Our services are absolutely free for all users. There are no PRO subscriptions, no restrictions on the number or maximum length of downloaded videos.