Bestand wählen

The little bot that wasn't

Zitierlink des Filmsegments
Embed Code

Automatisierte Medienanalyse

Erkannte Entitäten
cool so K hiring thank you for coming to my talk little what that wasn't an operator's apology and that's the last slot of the day and it's Sunday so thanks to everyone who is still in this assumes that um right um so since then 3 people as as the last 1 what according to you what would you think and operate is somebody or an that we who you think and operate areas but many people in this room come along with the you site and feel that's a good point and so I see here and who do you think and operate those the problem you want figure this begin by operator i mean about each 1 of us is an operator yielding a mobile the wise and you're here so and get every 1 of those is an operator and I mean this song this sense of of anyone who interact so that the rise of computer is an operator so what's this apology about and by apology I mean a defense of or a justification of what I do as an operator or what we do is as operators 6 2007 being we live in an age of present shock and software is eating the world along with data so the Western need to apologize yes
because no does this like I do so about
myself and I'm an amateur under my favorite happening track is from the from sample lead to censor
cost I'm border and handbook and the distance comes about comes out to be plentiful minds but and but I still love that more than a marathon I call it an and with on a
nice securing and tried I also added regionalization dilettantes um I was 1 of the last stage of proofreaders for Doctor of professor what about it was book on data visualization called the truthful my name is an acknowledgment so at least I qualify as a deterrent and of course I'm my frost poem
so I've been an open-source altering contributed for the last 7 years and some 1 of the projects I would myself is all sport that's but the word to my talk is about and I read many Puppet modules and and I want so and collect it was 1 of my favorite pieces of software it still is I've written a bunch of integrations and more useful and collective as well the money is a monitoring system I would be edge this client for the money and I want sort order lost as looking for transport and and last year I was in war with about what and as a member of the Network Operations Center I and a lot of document documentation contributions have been probably projects these days I worked as a
systems engineer and I work at's model and and these are some
of the technologies that we will use of the hat we get by many of them using Amazon Web Services for all of our infrastructure and we learn a party in Apache Tomcat I'll we have used by the name of monitoring and metric system is based on graphite as listed above so
I want to express support portrait of an operator as a systems engineer and some but before I get on with that I wanna thank frost for giving me this opportunity to speak about and the and operator and so and
last year I did this article from the logging magazine about the death of system administration and that triggered some thoughts and me which and I wasn't sure how I deal with the what the article was trying to say
um so 1st I
wanna put forth a justification or a defense of being a system administrator no going back to when operator is 1 thing of about mentioning was that anybody at the supermarket or more specifically guess years at the supermarket and all of you attracted them pretty much every day
and the other ones who have been up your groceries and I been you so they also operators and what they do is not really interesting but the kind of operators that I really love other ones still do their jobs as if their life depended on it all they have just doing it I like the difference between cashiers would just take your groceries
from the Cumbria but Bennett and quickly get into the collection a as again like in 1 swoop out there are some who we for you to collect each of your groceries for them in your back and then you move I like the 2nd category of operators for what I'm trying to say is that system administration isn't exactly um something that rock stars would like for example but it's still the a good job and so I my argument is that system administration is not dead but some of the Treasury has moved so for example somebody still has to install graphite and I don't mean this in in in a in a in a way it to a friend the graphite community I really mean that installing fight is still something somebody has to look into and spend time on own but that's
change right now you like Puppet modules to install stuff so you light up a bit more due to install graphite somebody still has to be with so it's initialization scripts and not just in its trips now you have thousands of summaries super region super service um and links often so in its upstart system B and B 1 2 and of course who was handling the inner-city accompanying somebody is well thank him is your system administered somebody still has to have the operating system like Ron jobs don't forget them you get a call for these things up by the sea level and turns out it's just a cron job which misfired or deduction properly and it who wrote that and no but most of the best products well so so summary still has has to go through all the change logs of all the software that you're responsible for that you learning and somebody still has to check that these nodes find and use undocumented features and good through but trackers did have issues with request batches just to make sure your systems up and running so we're still has to subscribe to a mailing lists their thing and sometimes you get good responses I a sea of course but you still have to hold 1 through I every now and then and of course some reactants to restart the them so it's so it's
OK exhibits not I love this stuff I still love it and because all that and you know really asked for you the chance to will be tried and systematization is as a spectral a job as any other but there's more to it so back to the portrait of an operator as systems engineer let's get on to some other things that a system engineer to they supposed to do so monitoring for bread and cheese and
I like what the authors of belts and we'll have to say about this that
the systems that we have to operate today of so complex that not a single person not a single person cannot understand the whole system completely it's about the good to do that to try to understand how everything in your infrastructure works but you still need European and so if you got it all style you need your band to make good music and some so monitoring is 1 of those things like you need to interact with other members of your team of other genes to you to figure out what needs to be more I don't have how the monitoring is essential for learning the business and so on so it's more than just uptime monitoring can be really issues in software design and logic and some it get it's an information borderline but only if your collecting all the light metrics otherwise it's a cool online for example not on get is a web server for Java applications and it exposes metrics about threads and now most people would think about just the threats which are currently busy like these threads like that's what go once the log on a system on that was the scalability of your application but that's not true the geometric that best thing to be used are the trends in service state and those are the ones which actually um described the congruency of his system so small differences like these matters locked up for performance for capacity planning and to this general understanding of the system but in this regard I really like gmx and which is the 1 which is job monitoring extensions of EU don't get is another example is an example of a of an application which implements Geremek's very nicely with all the relevant metrics that you might need I come they had this itself the Python projects out there like I spent quite a few evenings just trying to find the right Python libraries to an instrument of least a sample piece of wood and I couldn't find too many of them all too many well-maintained Python libraries but I think it's changing so all we can fix said but monitoring is also creative real thinking and another example I came across in the downtown will cause an a pattern called cluster immune system and so close to the immune system of a of deployments means that that's your monitoring is integral part of the whole softer delivery pipeline so and it so let's say you do with in these of application and the latency is higher by 50 ms cluster immune system pattern of deployments that would be considered of a deviation and of you would be essentially all back um so that's 1
of the other thing is in cloud environments specifically there are many ways why you can look at monitoring and so it's a creative exercise to pick a good way of
doing it yeah and
cool and dashboards and to everyone who uses graphite anymore and monitoring system use dashboards and ideally like the next generation of best words which are simply books and and it can be they can be very useful for an quick to be really dating some of something that you anticipate or expect to work in a certain way like 1 service causing latency and others so as you can quickly create would for that sort of thing correlated a few metrics not the best way to do that before time-series data but it's still a good would for quick radiation ideally I agree with or I echo word to use cross-lingual talks about in his book The creations that operation decisions must be users of online and guidance modest offline ones which basically means that you can't look at it like OK um let's collect everything we know about a certain thing and let's spend a few days or weeks to and I think about let's say the monitor performance of the system or why that application was behaving a certain at the we expect so you have to be constantly thinking about a your platform units researcher your applications to make the right decisions at the right time especially in case of outages or incidents you don't have that luxury you need to act in direct now so a good monitoring and the is essential for learning small operations which um takes me to the point of alert fatigue like in all your enthusiasm you didn't and all the monitors or a lot for for all of your services and in 2 weeks your e-mail is very grids thousand or 2 thousand messages and you have been in a state of alert fatigue now the problem with that is it's easy to get an come for dealing with that after while when you perceive that nobody is I'm looking at it than who cares and this also on that is an example of studying static patterns of thought has anyone made from wood again as an and the out of motorcycle
maintenance of the metaphysics of what quality so he doubts about static was dynamic patterns of thinking and I think cloud environments are of a good example of it's a it's a good example where you need to have a dynamic patterns of thinking to make the right decisions and do the right thing and she did uh of course
without data you're just a person with an opinion so all the metrics that you have collected what do they mean how do you make sense or all that so you so you you don't need to different lenses to look at your data islands of purely for example for all your performance and capacity planning requirements that's 1 lens that you will need to use Statistix Data Science um there are start ups nowadays which aim to solve some of these issues but yeah even they can't answer all the questions without active involvement from operators so it's you still have to spend time to think about these things I'm not Monday it myself but I know enough to be dangerous a sometimes monitoring is funny and profitable of some just a question do you guys have multiple metrics so monitoring systems cool now sites of which ones you have them OK you so do you have like another monitoring system and a cool case so it cannot be nice set up we have 2 monitoring systems 1 is dated about the other is monitored interest matrix I mean 1 is dated on and the other 1 is definably fight no uh of a whole department and become mean we are the Engineering Deans used to think about Grove why we need to and I have the same response to it like here let's just use 1 of these and I I personally was biased towards data dog um but I that i wanna speak for what that so what happened was that some I was supposed to present our world the feasibility of it all the lowest which in 1 of our main applications which power a business platform and to find the dipping warrant of that probably in terms of the lord it puts on the system and I was I was specifically looking for a high CPU usage spike I was relying on the that all the time because it had worked for me pretty much all the time before that I had tested and I had to test such things and this time around this was 1 of the big this like after this McGonigal production with that feature in the application and I did this test about the diamonds and this affected business also it wasn't like you know this but it was completely foolproof so each time I was making I was making I was doing the best and the there was some stress associated with it in terms of well you know somebody from this CIA's Office might just jump in and say what were they doing so I was working in that pressure and I couldn't see that system CPU spike that I was expecting to see and this is where having a separated at another system helped me um I'm under then I wasn't really a fan of the final and nothing against the software but I think we were learning about the relation and I don't like the the follicle appellate I think it really makes it hard to understand what's going on lately save me that day and we collect a metrics in graphite at 1 2nd resolution for 10 minutes and so I could see that CPU spike and that saved the day for me so but no monitoring is funny right moving on to automation and the duct tape of complex systems um the I think automation if you look at job descriptions of futures look at any news source which talking about technology or what's changing or what's coming in the future automation is a very important or a common commonly heard phrase and everybody's thinking well the jobs are going conical they are how this is gonna affect the economy but I I couldn't would this better myself but David Mendel he has certain books is it in a book a specifically on this he talks about technologies are not directly but not not from the perspective of what
I've been talking about like systems engineering system winter administration but some but out of systems in aerospace also in deep-sea know things like that like how automation in such systems affect the way operators will could those systems but I think I piece is not that different and I would take the
example of infrastructure automation and does anyone of fewer use config management of of automation tools like CloudFormation out their form an interview as sort of OK so um the it's very easy to create stacks of associated services with a few lines of codes of using tools like formation therefrom puppet and it can make it really easy to bring up whole of business in a new regional in a few hours does this mean that you eliminate the need for having operators on you in 1 single person do all of it all right it doesn't but it does need
a change in I'm still link and which is I think mind-set if I'm not wrong in German and yeah if you look at
it In this way that's that's becomes reusable components of your whole platform you can do all of these is more often you can test new features more quickly and this is how we did it I 1 of our pool of pieces of the application in a data pipeline we tested a couple of features i thing in a matter of a few days by creating confirmation stacks for that each time we were testing of features so that was really fast and some updates of course have to be initiated and monitored by more operators yes you can automate a lot of monitoring itself but still somebody needs to initiate that and you doing need operators so coming back to what of the read Mendel actually said that
automation changes the type of human involvement and transforms it but it does not eliminate it right the
the next thing that our system engineers must think about this performance analysis and Peter Denning uh he was 1 of the seminal figures in he is 1 of the seminal figures in computer science and he was very instrumental in of making which memory what it is today or what it has been over the years and then he wrote about 1 paper on performance modeling where he said that performance modeling is experimental computer science and advanced at at its best and he was right I but performance analysis is it it's also experimental computer science at its worst based on how you're approaching it what data you have to walk with performance analysis in your set so this speak about that in the clouds and because that's what I have most experience with as the cloud makes it interesting that whole thing of of the it makes it interesting and the you still need as an understanding of capacity planning even with all the flexibility in the last 2 city that cloud providers promise and they also provide multiple pricing models and mediums to experiment with and you do need an understanding of performance analysis to make sense out of all those options you can also mix and match concurrency paradigms in the Cloud like you can use so as architectures along with conventional my descending and multiprocessing techniques of where application the cloud actually makes it some times possible it's not always possible to do the best performance analysis that you must to and
because your cloud provider did not exposed to light user-level user metrics and I like this happened to me on and I was looking for utilization metrics for all of us to wage war looms and after a lot of discussions and it we found out that some of the user facing metrics are not exactly the present paper we didn't get it all of the exact performance and the this yeah at this the mixability that's kind of of G-box linear looking at club providers not completely a black box but more like the gray box and this can affect the correctness of your performance models and uh that can affect your predictions for the resources that you might need in the future another thing to think about
is uh it'd application that if all these interesting options available to you so for example you concordancing Python if you're using the Monte setting what do know is limited uh is not real competency is limited by the global into a lot of data is a project which aims to get rid of it and uh but you just need to be aware that you can't expect uh food and concurrency out all from your application if you're using a particular module in Python even this concurrency is also hide a lot but these days has been for a few years and but even that is limited to using 1 single core the main ListNet said so that's also a limitation no programming and I'm sure everybody has there what's in OWL 1 of the most
common things that there what people don't go out is that you need to know and you need to know how to program as in those operated which is fine but this is not that I'm going to speak about programming in a different sense so
I think programming is basically to see now uh or it should be and I think it's a prerequisite for more in existence because someone's and someone else's avoidance and cooled is affecting our lives every day and getting gave the keynote on this subject a in a much with the farm or loss of extreme consequences in her case when is directly related to somebody's life so sigh well United and also as an administrator you want your systems to behave nice side of the application need not be a black box to you of some tools that you can use for better visibility and understanding board parts in your application and 1 of the tools is green grass a I quickly show how whole thing graph looks like I so that's a plane graph of Java application
and basically you see all the court parts that are being aligned walk from this application and you can see
which 1 is us sitting on CPU most uh is consuming most CPU which 1 of these uj our libraries which ones are system and the libraries and processes so the sort of tools exist today and and you can
use them to better understand what the applications of going from an administrative point of view of cool sometimes you must apply emergency patches but if you have to for example in the past I think 1 couple of occasions I had to batch of production public master and of yeah you know the administrative softer these but it is cold and your 1st a Python there nobody is probably not interested in fixing your Puppet Master and the life of it must so yeah you must program of the and lastly about security and I had actually I left out security completely rows of life was the offered the slides but that obviously did not feel right for me was what it
has always been that security must be part of the any operators part process and um it's not something that you look afterwards it's something that's been you built in your application or in your systems and obviously that's not the case but but not for me I never got a why somebody would not look at security from the beginning so for me it's it's just stranger than fiction and the reason alive
the the there's 1 thing I wanted to sort security that my own initiation with
computers was inspired a lot by fictional fiction fiction on security on networks computers computer network security and maybe that's why from the very beginning I was very conscious of having a secure having a secure environment for running applications in such so you now what about
the little boy that wasn't um no the I really like this anecdote from an enduring life that once a gene that he was a machine and when you woke up and I said I don't know whether I'm Q doing dreaming that machine or a machine being that and keep doing and somewhat similar thoughts I had when I wrote board about 5 years ago and I wrote it as a natural way for an operator to put off for an operator to and get a sense of negative as you walk in in in as you walking in an office in your office and what's the what's the most recent thing that happened what incidents went off what issues do you need to walk on so in that sense I draw something known as all sport which was a pluggable bought for operations teams with multiple sources of data and information like issue trackers monitoring systems in chat software with essentially this source adults
afforded became what it's now so I do that in the May 2012 and this graph stance and you but the interesting that don't share years this is from Google so yes even though jet is past its sell-by date according to me but I just wanna let it out there that it sometimes it's nice to have this obligation i'd like is and book authors die of about getting the recognition they deserve and they don't even know that OK now they have millions of readers but at least I can see from Google what's happening which adults now so I don't think I was the 1st 1 of course Iousy has had bought since whatever and determine reward their Hugh bought offer around this time I think you 6 months or a year before I released mine so but yes still use nice um so what about the board that wasn't this is all nice and not has anybody that's all those book the dangling men he talks about a the guy who tries to juggle
between is my interest for scholarly inquiry like literature and art was is what he has to do do on his bed and also eventually to work for the military so he dangles between these and of things and and I I think it's sort of similar for operators to like and of based on you know the number of jobs that do exist for programmers systems and system engineers for us I feel I get a sense of the dangling operator like um I was in a test for about 2 years of 1 half years back where I had the best opportunity for myself as a system administrator and this was for a gene company and I basically a field
that and that form was much harder than the full I had last week from an onboard in the street in Hamburg and it took me quite a while to recover from that so I do like this school remedies is that knowledge will experience can make you better knowledge and experience without knowledge can make a schizophrenic I think that's all of freely applies to operate as you know with the amount of technology these days so what you do then well you just keep operating you've been rant yourself but you have to keep operating easier said than that easier said than that of course but you must believe that what you're doing is really important our even if it's trying out a new wind plug-in to keep the excited about your favorite editor and you must do it gone stopped um what was the enemy of the enemy is the resistance year press read but that's is the the idea that resistance is and the life-threatening force that's that would prevent you from pursuing your creative the call and you know up all the lizard drain as sets Gordon cause it's like you know some as humans of brilliancy rewards for many years and 1 part of a
brain the lizard brain he called it it only wants to eat and you only want to you know survive and all resigning and you know competing against everyone else and that's an enemy that your enemy and the problem is people who don't feel awful to the food and realize that and other lizard events around you can also be your enemy and the death instinct which Thomas modern has talked about his a pure lodging and he calls it the death instinct when you don't wanna believe in in your faith and for operators your feet could be unique it could be any of the it could be and if you have any 1 of these things so don't stop believing what can you do that will you mean that if you have to put keep operating so again
this is easier said than done but but it's necessary for for example some 2 years back when I had that fall that I missed my gene company thought hard about what I've been doing for 5 years n the if it was meaningful at all and that's when I really have some thought about data visualization as a feed like a today you need the lens of data visualization to make any sense of what you're doing if you're a programmer ifyou're assistant the no matter what you do lization can be a very or should be an integral part of the toolkits to really understand what you're doing it's being used across many in this articles and I think we should be at the forefront of that and given that we are the ones who will with these things all the you can still make you can make it easy for yourself by having by uh situated the invention I borrow this term from situated cognition which states is that you think about certain things in its political social cultural or individual context and so in for us in computer science and the idea like Peter Denning Sunday principles of computing the bookie gives a whole framework of how to look at computer science and I think you can make use of a thing that
feeling inspires you and yourself in that domain or in that field right
that was that some I wanna say thanks to a lot of people this is especially to the nature of trees that you're listening and things were giving my American dream in Germany and the student also quality he introduced me to frost on some that should say frost con 2016 but yet introduced into fast gone last year and Dr. Neil Gunther for being of a patient mentor and looking forward to the data analysis and training in September follow my companion some she is
followed for 54 and I've
been of for writing and in collective and it's 1 of the it's 1 of the applications that really inspired me as a system administered early on to you actually think about programming and but still I think 1 of the best pieces of software that any system administrator can make use of I wanna thank all the operators everywhere in the supermarkets in everybody was developing everybody is an administrator but not lizards if you're listening and if don't operator stop being that and the idea of Hamburg Hamburg has been kind to me for the last 1 and a half years and yet I hope to be done some of that love back and of course what bins for the 1st computer in my small room in the cubbyholes of love which I won't be standing here and get the land but
it any questions and if although he no material like any other like no was the flame graphs
1 of the book we have some I actually I think I did not mention it come correctly
from so the book is called the
truth will out right sexually 1 of the really important books that came out I think last year have written by Professor Alberto guys or he teaches at the University of Miami Florida and yeah I think I just with this can go on the progress set it's my name is in Acknowledgements but yeah this book is very important he's an important figure in the whole field of data visualization and and this is the sequel to his book called the function a lot you can also go for that before going for this 1 and some yeah so when I spoke and I actually got it the 1st or the 2nd day I was in Hamburg and it was there was a I 1 of the best feelings ever from New right then
thanks a lot for thank you tools to the thanks to verse
Ungerichteter Graph
Service provider
Gruppe <Mathematik>
Statistische Analyse
Skript <Programm>
Generator <Informatik>
Einheit <Mathematik>
Physikalische Theorie
Ordnung <Mathematik>
Lesen <Datenverarbeitung>
Mathematische Logik
Virtuelle Maschine
Weg <Topologie>
Modul <Datentyp>
Fächer <Mathematik>
Endogene Variable
Installation <Informatik>
Architektur <Informatik>
Binder <Informatik>
Komplex <Algebra>
Wort <Informatik>
Prozess <Physik>
Natürliche Zahl
Statistische Hypothese
Fortsetzung <Mathematik>
Computerunterstütztes Verfahren
Lineares Funktional
Physikalischer Effekt
Web Site
Konfiguration <Informatik>
Arithmetisches Mittel
Projektive Ebene
Extreme programming
Web Site
Nichtlinearer Operator
Kontextbezogenes System
Physikalisches System
Inverser Limes
Operations Research
Leistung <Physik>
Physikalisches System
GRASS <Programm>
Offene Menge
Deskriptive Statistik
Negative Zahl
Elastische Deformation
Schreiben <Datenverarbeitung>
Kategorie <Mathematik>
Güte der Anpassung
Kontextbezogenes System
Intelligentes Netz
Ebener Graph
Web log
Rhombus <Mathematik>
Dienst <Informatik>
Rechter Winkel
Mathematische Logik
Automatische Handlungsplanung
Patch <Software>
Dienst <Informatik>
Service provider
Open Source
Lesezeichen <Internet>
Arithmetische Folge
Endlicher Graph
Skript <Programm>
Ideal <Mathematik>
Automatische Handlungsplanung
Komplexes System
Patch <Software>
Metrisches System
Gemeinsamer Speicher
Imaginäre Zahl
Kartesische Koordinaten
Arithmetischer Ausdruck
Einheit <Mathematik>
Prozess <Informatik>
Figurierte Zahl
Nichtlinearer Operator
Prozess <Informatik>
Installation <Informatik>
Twitter <Softwareplattform>
Virtuelle Maschine
Keller <Informatik>
Kombinatorische Gruppentheorie
Framework <Informatik>
Zusammenhängender Graph
Matching <Graphentheorie>
Diskretes System
Einfache Genauigkeit
Cloud Computing
Keller <Informatik>
Formale Sprache


Formale Metadaten

Titel The little bot that wasn't
Untertitel An operator's apology
Serientitel FrOSCon 2017
Autor Chawla, Mohit
Lizenz CC-Namensnennung 4.0 International:
Sie dürfen das Werk bzw. den Inhalt zu jedem legalen Zweck nutzen, verändern und in unveränderter oder veränderter Form vervielfältigen, verbreiten und öffentlich zugänglich machen, sofern Sie den Namen des Autors/Rechteinhabers in der von ihm festgelegten Weise nennen.
DOI 10.5446/32308
Herausgeber Free and Open Source software Conference (FrOSCon) e.V.
Erscheinungsjahr 2017
Sprache Englisch

Inhaltliche Metadaten

Fachgebiet Informatik
Abstract The talk is about an operator’s view of his profession, an apology rather, in the sense of being an apology that G.H. Hardy wrote about, in his book A Mathematician’s Apology.
Schlagwörter Cooexisting with Bots

Zugehöriges Material

Video wird in der folgenden Ressource zitiert

Ähnliche Filme