YouTube Transcript:
Building Operable Software with TDD (but not the way you think) - Martin Thwaites - NDC London 2023

Skip watching entire videos - get the full transcript, search for keywords, and copy with one click.

Video Transcript

um this talk just in case you're in the

wrong room we're talking about building

operable software with tdd

um and the obviously a hook of not the

way you think so hopefully there'll be

some surprises in there

okay so my name is Martin Thwaites I go

by martin.net in everywhere

um I am first and foremost an

observability evangelist I'm also a

developer advocate for a company called honeycomb

honeycomb

um socials are there we have a booth

just outside the door come see us blah

blah blah they paid me to be here so

come see us

um awesome

so what we're going to talk about today

we're going to talk about

operable software

um we're going to talk about what my

definition of operable software is more

specifically because that's all that

matters we're going to talk about how we

test software as developers we're going

to look at what's something called

outside in testing

um what that looks like and we're going

to talk about how we can use

observability and how does that fit in

to that operable software life cycle

now before we get to that I want to talk

a little bit about the motivations for

this talk

before I started working with honeycomb

I spent a couple of years working for a

UK Bank building a new banking software

and one of the things about banking

software is it needs to be accurate and

correct apparently giving somebody the

wrong balance and approving things when

they shouldn't be approved is a bad thing

thing

um so basically being able to say that

my code is correct

on the bottom layer isn't good enough

because a system is bigger than your class

class

and we had to spend a lot of time

thinking about how we test software we

wanted to go full CI CD we wanted to

have really continuous release cycles

and a lot of that goes against the idea

of resilient software the idea of

software that doesn't change you know

banking industry is we'll do three

releases every Century

um and that was a bad thing in agile

software in building software that is

resistant to change all that kind of

stuff it just isn't a thing in banking

so we spent a lot of time building out

approaches and that kind of stuff so

what I'm going to talk about today is

not theoretical

it is something that we spent two years

building and is running in a production

environment well production-like

environment they're not actually taking

payments yet but it is running an

environment and it is running cycles and

it is something that works so hopefully I've

I've

I'm not going to say dumbed this down

but what I've done is I've tried to

simplify this into some of the component parts

parts

but do take away from this that it is

something that works

okay so what do we mean by operable

software or more specifically what do I

mean by operable software first and

foremost it's software that's built for production

production

now I hate to say this

I don't care whether it works on your machine

machine

it's not where customers are going to be

using it when we build software we build

it to be run in production that's the

thing that we're thinking about we're

not thinking about how we can run an

emulator locally and it doesn't matter

what matters is does it run in the

production environment when customers

are using it

so what does that mean

built to run in production it's built to

run on production servers it's built to

run in a production environment it's

built to run with multiple regions it's

built those are things that you're

it's also built with live operations in mind

mind

now some of you may come from the

environments in the past where you'd

write your code you'd throw it over and

somebody else would then have to deal

with it because that's not your job they

do with it unfortunately everybody's Ops

now everybody is Ops

um we are working in a devops culture

where developers and operations teams

come together

you need to build your software with the

people who are going to run it in mind

sometimes that's yourself

because

past Martin is let's be honest a little

bit of an because he put a lot

of problems my way and future mine we

can leave this off of him because you

know he's fine um

okay now the other thing is built for

fast recoverability there are things

called the Dora metrics which Google

built on the devops reports

um which it mentions something called

mttr it's falling a little bit out of

favor right now but it doesn't mean that

it's not something you should be working

to which is the idea of how do I build

things that recover fast from failures

and all of that leads to the idea that

my application needs to emit robust

Telemetry signals whether that's metrics

or traces or God forbid using logs but

that is something that you need to be

thinking about because in production you

can't attach a debugger has anybody

tried to attach a bugger a debugger to a

live running IIs instance

yeah did it go well

yes for those that don't know

um whenever you attach a debugger to a

live running process it stops all the

threads every request just holds it's

not a pleasant day okay

so when we talk about operable software

what do we mean by software

now the traditional definition is the

code that you write that does the thing

that the business wanted unfortunately

that isn't the case it's everything

it's the application you write it's the

test that are part of your code

it's the pipelines that you've built

that test

do security analysis and deploy your

code into production

it's also the infrastructure that's been

built for it that might not be you who's

built that infrastructure but it's still

part of the software

because the production environment is

part of your team that delivers value

into your customers

that is part of it

you can't host your application without infrastructure

infrastructure

I mean that's just shipping your own

machine and that never works so but the

other thing is it's your observability

tooling it's the live operations team

that need to know whether your system is

up whether it's functioning all of that

is part of the software that you're building

building

so when you're building software if

you're not considering these things

you're missing out

you're neglecting part of your software okay

okay

so let's talk a little bit about testing

let's talk a little bit specifically about

about tdd

so what do we mean by tdd um

tdd has been completed over the years I

mean it's it's very abnormal we don't

take terms in I.T and use them for

different things ever do we no that's

not we don't take a well-known term like

you know I don't know devops and distort

it into another thing

tdd was a very very basic concept

contract First Development develop your

contract develop your tests do the thing

it's also Red Green refactor

is the idea that you write a test that

fails you make the test pass you can

refactor your code and then it fails and

then there's a big cycle that goes through

through

but the main basis of tdd

is you write tests first and then you

write the implementation unfortunately

unfortunately

tdd has not meant that to a lot of

people and one of the things that we

found while writing this approach

was people didn't like tdd they said

they thought it didn't work it's it

takes too long

and when you dig into it it's because

what they're doing

is they're mocking all the things um

the.net developers among you will get

the reference there um

but they will mock everything

everything

they will have a class with 17

dependencies all of them with mock

and they will test individual methods

I've got a method that I've mocked out

all the dependencies for I will test

that method to death

now nobody cares about your 4 000 unit

tests I know you like them

I know you spent a lot of time on them

but nobody cares about the level of your

unit tests what they care about is your

so what does why do we do tdd

because if it has so many constraints

and it's so bad why is it that we're

doing it what are the benefits that

we're trying to get out of tdd

now tdd is known to improve the design

of contracts I put methods on there

might not be methods it might be API

contracts for your SDK it might be API

contracts for your HD http apis

but it's based on improving

the design of your method contracts

because you write a test and it passes

that test you should do the smallest

thing that makes it past that test which

means your interfaces evolve your

classes evolve

it also helps under you improve your

understanding of the code base because

you're using these methods

you're not writing a test

that does a very very small thing you're

writing something that's actually using

these apis in the way that other people

it also documents your code if you're

writing these tests well

your tests are going to actually be

documenting your code as you go

because you've wrote a test for a

business outcome or a particular class

that you wanted to do something if

you've wrote your test properly you've

actually got documentation already

which means you never need to go into

Confluence again

okay but what's the Bad and the bad here

comes from what I believe is the the way

that a lot of people run tdd in their teams

teams so

it's normally written these tests are

normally written in the absence of any

consumers of this application

you'll get a business requirement that

says let's do a transaction and what

you'll do is you'll write a transaction

class and then you'll write some methods

on your transaction class like add amount

amount

that's not what your consumers have

asked for

what you're doing is you're assuming

what that class needs to do and then

assuming what you're going to use that

in and then assuming assuming assuming okay

okay

and because of that

all of those tests become quite brittle

to change

I've added a new parameter I've changed

the parameter

well I'll just delete all those tests

that's bad because now all of a sudden

you don't know whether your system still

works because you've deleted all the

tests and stats on new ones which means

that you then end up with manual testing

from the outside in and all of that kind

of stuff okay

okay

let's talk a little bit about component

testing everybody's seen this meme I assume

assume

where what we end up with is lots and

lots of unit tests four thousand unit

tests on a single class and it's

bulletproof that class

absolutely that draw was probably the

smoothest opening draw that you've ever seen

seen

but that's where integration tests come

in now integration tests over the many

many years like with all the other

things we've been talking about have

been misconstrued and changed and they

mean 17 different things to 17 different people

people

um so what I like to call these that I'm

talking about is component tests

um they all fall under a banner of

something called developer tests which

is something that Ian Cooper's been

pushing for quite a while which is the

idea that these are tests that the

developers write

to ensure the completeness of their code

the completeness being correct Acura

does the right thing these are tests

that you write

so outside in testing is part of what I

consider the component testing workflow

and what that means is we're working

from the outside of our application

we're working from where consumers live

that could be your HTTP API it could be

your SDK that you've built but where are

your consumers the thing that you

deliver as a package as a thing how do

your consumers use it so we're working

right from the outside all the way to

the inside

now what what I found is the majority of

time you can actually stay 100 on the

outside which is great because you don't

need to be on the internals

what we're doing is we're testing the

inputs and outputs and I'm obviously

using the HTTP vernacular here because

I'm guessing that I would say over 50

here write web applications whether it's

apis all that kind of stuff

but that could be just using your apis

what it means is we're testing from our

consumers or users we're testing what

our users will do

they will add something via a post they

will go and get something fire get well

let's do the post and do the get then

but we're also testing specifically what

the businesses are asked us for our

users to be able to do

which means that we're really close to

only delivering what we need to do

now you might have delivered this this

class with 4 000 unit tests that has 15

methods on it and those methods are a

um and then you look at the actual usage

in your application there's only one of

them used and they use one property from

that object you've done too much

running from the outside in means that

you're only doing the things that you

need to actually build

so benefits

benefits

business requirements first

all of a sudden you've built you've

designed all your business requirements

you've satisfied all your business

requirements there's no need for that

further layer of testing

if you've got tests that prove each

individual step of your application

you end up with better readability

because you're not using your internal

business knowledge

you're not using a test that says

transaction service returns approved

approved

now what you're saying is the

transaction was approved

the transaction was cleared

authorization was failed

you're using the business terms inside

of your test which means that they are

more easy to read

a little bit more separated

so that we can start to act on

individual tests and we can just add

more tests

we got so many times people saying I

don't want to add more tests because

it's more work

you know I've got to copy 70 lines of

code well that's the wrong reason not to

write tests

what you want to do is make it easy to

write new tests what these approaches

give you is something really really easy

in a framework that somebody says

there's a bug in production

and what you've got is a Lego

set of different tests that they go oh

this is broken it's like right grab one

let's grab one oh yeah it's broken

oh yeah oh I can fix that great fixed

job done and now I've got a test for it

guess what you've done you've wrote a

test first

and it failed and then you fixed it

okay so

then we can go for persistent check so

now they've come back to us and said

well when I add the to do I need to be

able to get it back like great okay now

I need a database

um but up until this point I didn't

um like I say I love being pedantic

um so this is now right we need to add a

database to this

but again what we've done

is we've abstracted away context

now if you read what this test does it

should be self-evident

it says create a valid to do item

now you could say

add a valid to do item to the API

whatever makes context important to you

whatever makes that understandable to people

people

but we've gone and created a to-do in

the API

we've not gone to the database

we're not understanding the scheme of

the database and injecting something

into the database we're using the API here

here

we're essentially calling into the

previous tests

that create a valid to do item

and what that's going to do is give us a

valid to do item back

and then we're going to go and get it

from the API

now at no point here am I touching the

database I'm not reaching into the

bowels of my application and

understanding what columns exist in the

database I don't need to

I can run this against the API because

that's what my users are going to do if

you've got users that are going into

your database and inserting to do items

into your database you need to have a

real think about your applications and

your users

that isn't what they're doing so why is

that what you're doing in your tests

nobody cares when there's a bug in

production and you say well I've got a

test that inserts into the database and

make sure it comes back I just didn't

test that I post it through the API and

it comes back nobody's going to take

that as a valid excuse

okay

recap a little bit about what we've

tested there

was testing our Startup configuration

how many people have created a service

class in.net

and had their 4 000 unit tests against

their service class and then forgot to

add it to dependency injection before it

at least in Oslo they were about 50 of

truthful people

um but that's what happens people will

write these tests and their unit tests

pass great ship it and then it fails to

start in production and you're like you

need to surpassing

and what you've actually forgot to do is

just register that implementation

against your service collection

the other thing with testing is routing

we're testing that the API routes that

because that happens as well

you get people who forget to put a guide

identifier a good qualifier on some of

their names so the routing looks like it

should work but it doesn't because

we're also testing serialization and deserialization

deserialization

because when you push stuff into your

application via the API it's going to do

your Json serialization deserialization

now Pro tip on this one

is when you're using this approach

rewrite all of your classes in your test

assembly do not reuse the classes in

your main assembly

because what that gives you is if you

have to change the classes in your test

assembly because of something that's

breaking in your main assembly all of

your consumers are going to have to do

the same

and what you can actually do is put path

approval checks inside of your commits

that says if somebody's changed this

somebody else needs to look at it

so what you've actually got is contract

testing built in to your application

if somebody changes the serialization

method from camel case to Pascal case on

the Json objects all your tests are

but if you use the same serialization

classes then you've got a problem

we're also testing model validation

required parameters regexes

because we're going through the full

HTTP pipeline when we do it

we're testing things like database interaction

interaction

if you're using Entity framework which

this is an example that's up on GitHub

that you can go and have a look at it's

using Entity Framework

it's using SQL Lite don't use the

in-memory database because constraints

don't work

but if you're using sqlite you've got a

database that you can use

so you're actually testing that the

database interaction Works your foreign

we're testing selecting of columns from

the database properly

making sure that you're actually using

the right column names

in your tests

but we're also testing the customer apis

we're not testing our classes right down

below we're not testing those

we're testing our apis the things our

customers are going to use which means

we're testing the business requirements

which means that we're testing only what

needs to be tested

I don't care that your class returns an

object with 17 properties on it if I

only need one of them

and we're testing all of that

and like I say if you do that you end up

now

just before we move on to that one

one of the main objections that I get

for this is it takes too long to run the tests

tests

because I'm running up an application so

it's going to take too long um

normally that's people who haven't tried it

we have around two and a half thousand

tests in that framework that run in

eight seconds

so anybody who says this takes too long

has not tried it and done it properly it

takes time to build Frameworks around it yes

yes

it will work out the box a lot of the

time you can reuse the web application

from Factory you can use different

contexts for your database so that you

get clean tests yes it takes a little

bit of work it's not too slow though

if you're worried about your test taking

eight seconds I'd ask you how long you

deploy is taking

because I guarantee you it's not less

but what can't we test this way

you know I've sold you the dream I'm

sure you're all going to go away

tomorrow or Monday and do all of this

stuff which is great

but what can we test

so we can't test things like how the

back end changes but the front end doesn't

doesn't

to the user cashing being one of the big

examples of that if I'm doing a get

against my API how do I test that

caching works

because I get the same result no matter

what if I'm not getting the same result

then my Test's broken I should get the

we can't test things like internal State

changes like audit logs

things like that

we can't test performance

anybody who thinks they can test

performance locally

you're also not testing integration with

cloud services

so you're not testing integration with

Cosmos you're not into date the

integration with Dynamo things like that

you can do that with this approach

because you're working from the outside in

what you can do is you can switch out

the implementation with the live

implementation and rerun the tests

maybe if you've got branching structure

something like that you run that test on

your main branch when it's on when it's

actually merged in but you don't run it

on every feature Branch maybe or the

other way around

but because you're running from the

outside in it's literally a case of in

your service collection just replacing

your implementation for I don't know

the other thing you can't test is

configuration so you can't test that

your deployment pipelines have put the

right key Vault ID into the config so

and this is where we start to talk a

little bit about observability

so all those things yes you can't really

test them with the approach that we just

talked about

but we can augment it a little bit in

order to get some of those techniques in there

there

so I'm gonna have to talk a little bit

about what observability is

and this is the definition from our

co-founder who first coined the term

observability in the context of software

about six years ago

so it's about understanding and

debugging on no no knowns it's about the

ability to understand the inner state of

our system by asking questions from the outside

outside

that's the statement that I'm more

we're writing observability into our applications

applications

writing those Telemetry things so that

we can understand our internal State

when we're in production

if we're writing operable software in a

way that means that we can use it in

production well can we not do the same

thing with observability in our local environments

environments

and that's what I set out investigating

about a year ago

and we came up with this concept I hate

the fact that we use the term odd um

because what we actually turned it was

observability during development which

is just really confusing but the idea is

how do we use observability or more

specifically tracing I wanted to call it

tracing during development but

apparently tdd's taken and we're not

allowed to just augment names and use

them for something else um

um but

but

this whole Concepts don't call it odd

call it whatever you want but it's the

idea of how do we bring observability

into our applications and our local

development folks

observability is an output it's a side

effect of our application it's something

that our application does

so why aren't we testing it if it's

observability can detect things like parallelization

parallelization

you can know whether two things run in

parallel or run in series

that's important in some scenarios

I have a few anecdotes that I can share

offline but one that I can share online

is a friend of mine went into an organization

organization

and they had a problem that the startup

of their desktop application was taking

30 seconds

and he went in put some open Telemetry

and tracing data in there and said well

when you do this grpc call you're

initiating the connection for everyone

so it's initiating the connection 20 000

times it's like no no it opens the

connection once and then just sends all

the data over he's like it doesn't

that's not what it does it's like no no

it absolutely does and it's like but

the computer's telling me that it

doesn't and computers don't lie they

only tell you exactly what they're doing

um and then they went in and went oh oh

yeah it's not

um and turned 20 seconds into 200 milliseconds

milliseconds

um just by three lines of code of reuse

the connection

so observability can tell you these

things and they're important in a lot of applications

the other thing it can do is detect code

paths it can detect if you go through

say caching

if the caching code is called

you can tell if this was a cached

version or a not cache version

but the key is if it's important to you

locally for you to be able to write a

test on your class about it it's likely

going to be important when it gets onto

production and it goes wrong

it's likely going to be important that

you know that that's where it went wrong

in your application

what you really want is when somebody

sends you a book

what you want them to do is say it was

on this line here and it sent the wrong

information all right I'll change that

line there and commit it and push it

like that that's a really good QA who

can come to you and say it's this line

of code can you fix this line of code

observability done properly with tracing

causality all of those kind of things

can really show you internally in the

application what it does

so if that's important why aren't we

bringing that into our testing why

aren't we bringing that into our

so let's take a look at some caching code

for any future employers out there that

may look at hiring me I don't write

caching code like this this is an example

example

it's very basic I wouldn't write it like this

this

um so to give me an idea of what this is

doing for those that don't read c-sharp

which I'm hoping is a minority what

we're doing is we're setting a cache key

we're using a memory cache here to try

and get a list of items list of to-do

items using that cache key if that

didn't return anything then we're going

to go to the database and go and get the

to-do items and then we're going to

return the ones from the database instead

instead

the methodology there is pretty sound

that's how you do a lot of things in

applications you'd have sliding

expirations all that kind of stuff

but that's pretty much what a cash code

would do

now the thing is

both the line that says return to-do list

list

and the one that says return all to do's

they're both returning the same information

information

so external to the application I don't

know whether it's actually got the data

from the cache or I've got it from the database

so when I write my test

and write my test that says well the

first call needs to come from the database

database

I don't know how to write that because

whether it comes from the database or it

comes from the cache it's going to be

the same so how do I do it yes I could

write headers to it but those can be

messed up

and this is where something called open

Telemetry comes in

so open Telemetry is the number two

projects to come out of the cloud

compute native Cloud native compute Foundation

Foundation

behind this little small project they

did called kubernetes don't know whether

you've heard of it um

but it's the number two projects come

out there it's becoming the de facto

standard in How We Do Telemetry

signaling Telemetry signaling being at

it's completely open source

it's run by a lot of the vendors the

vendors have been building this as an

open source project

but one of the really interesting things

about it is something called the

in-memory exporter now what that means

is and especially in the.net world

your production application does not

need to depend on open telemetry

in order to get Telemetry out

because a lot of organizations will go

we're not we're not ready to take open

Telemetry into our production

application yeah there's some problems

but because

net uses something called activity

activity has been built into.net since

it's either done at 2 or 3.5 and it's

changed a little bit since but it's part

of the BCL

so you can instrument your code with activity

activity

without thinking about open telemetry

and then you can actually bring open

Telemetry into your test because largely

organizations don't care what Frameworks

you bring into tests they care about

what's deployed to production

and what we're doing here is we're

creating a tracer provider

inside of our web application Factory

not inside of our code

not touching the production code here

but in all we're doing here is saying

listen to this particular activity source

source

bringing the asp.net core

instrumentation which will give us spans

for requests

and any spans that are created any

activities that are emitted from your

application stick them in this list

so we've got a list now

got a list of individual spans that have

been created as part of our test

and then what we do

is we add it to our services

because when you're using the custom web

application Factory

you can add additional things to your

service collection that are only present

when your tests wrong

now because I've added this as a

signaltern like that

that means it's going to start listening

to all of the activities that are run

and created during my tests

so that's pretty interesting we can do

that another Pro tip when you're writing

your custom web application tests run

this clear providers

it'll save you a ton of time in your

tests because if you don't and you've

accidentally enabled console logging

your tests take a hell of a lot longer

because it outputs all of your logs to

the console which takes a lot more time

a bonus nothing to do with what we're

talking here

I don't charge extra for that just to be clear

clear

so what does our code look like now

well we wrap

our database call

inside of a span we get the added benefit

benefit

that we now get to know how long it

takes to get things from the database

that's something that's interesting in

our application

it's not something we're going to test

against because we can't test

performance locally it's pointless

because we're running against an

in-memory SQL and

SQL light will be a lot faster or slower

depending on what you're doing

we're going to wrap it in there and give

it a specific name

that's it

it's not particularly complicated but

now what we can do

is in our test

we can say

well find all of those collected spans

that have been created

and tell me did it contain one that had

get to do this from the DB well this

test is making sure that the first cold

calls the database

I actually wouldn't write this test

because that's pointless

because we know that the first test can

will only pass if it's got data

so kind of pointless test this one

the second test though is the inverse of that

that

well what we want to do is make sure

that the second cold does not touch the database

database

and what we can do there is we'll run

our first get

clear out all the spans

run our second get and make sure we

don't have

something called that

so now what we've got is a really

actually useful test that tells us

whether it's going to go to the cache or

not or is whether it's more specifically

whether it's going to call the database

or not to get that data

obviously very very simplistic test here

there's a lot of nuance that will be in

your application that you'll need to

throw into that

do you use I distributed cash and you

have to use the in-memory version of

distributed cache you have to make sure

those things are serializable

that's another thing that people miss

when they try and use I distributed cash

which needs serializable entities

whereas in-memory cache does not

and that fails when you put the actual

real reddish cache in place but it

okay

now the other ones that we did we

so now I can take that first test where

it calls the database

I can make sure well Collective spans

has span with name

I've got a really readable test

that tells me that I could even go

further and have a static list a

constants of all my spam names

because that might be important that

they remain consistent throughout

releases you can have them in shared

libraries for instance that you can use

and that could actually be referenced in

a constant there's lots of stuff that

you can do but the main thing is look at

your tests make sure they're readable

make sure they're testing the things

now the other thing I said is we could

talk about parallel processing we can

make sure that these things are run in parallel

parallel

that's something that you can test with

observability much much harder to test

when you're running actual code even

when you're running code inside of a

unit test where you're just test in a

class it's still hard to test parallelization

parallelization

but what we can do with this is say well

get me the spans with this name so start get

get

bad name but hey ho

and then get me all of the spans that

say get from DB so maybe in this example

what we've got is it's actually doing

seven or eight calls to the database but

we don't want to run them in series we

want to make sure that it runs all five

of them at the same time because that's quicker

quicker

and then we can say well make sure that

every single one of our processing spans

has a parent ID of our new span you

could do things like make sure that none

of the spans start after the other spans

there's lots of ways to do this

that allow you to be able to see into

your code

and see what it's doing

so a couple of do's and don'ts

use it for things you care about in production

use it to test things that are important

don't use it to test things that are not important

do make sure that you're thinking about

the data on your spans

that might be important to you take the

first two into account

but you can test against some decent

data in your spans

don't check the database was called

directly don't test your query don't

test that it hit the database with

select star from users

that's not what it's for that's abusing

the system don't abuse the system

don't check that an individual method

was called

don't create a span for every method and

then go in and say did it cull this

particular algorithm method that's not

important to your test that's how we

now the other thing that we can do once

we've got all of this in place

we're using open Telemetry in our tests

what we can then do is we can use

tracing tools to start to understand our

applications better

during our test Cycles

so what we can do is we can take all of

our test runs

and we can send them into something like

honeycomb it's not specific to Honeycomb

but you can

we can send all of those in and we can

say well how many spans were created

during that test how many spans were

created during the next test over that

whole run how long did the whole run

take what was the average time for the

test things that you used to get out of

the box

when you were using things like team

City and things like that when you move

into using say GitHub a lot of those

tools are lost

but you can use that now the other thing is

that because we're producing traces

inside of our application now

and we're using open telemetry we can

actually see the traces and visualize

the traces in your Trace tool

so this is an example of that does not

call database test

what I can do is I can look at that and

say well what is the shape of the trace

for that particular thing when I ran the test

test

now if you're using these

when you go to production

if you use the same tools

in production for tracing your

production application as you are using locally

locally

you can start to see what that should

look like in production and when you

look at it and go well no no no that's

that's not how it should work

you're getting more knowledge about how

some some interesting reading on stuff

the repository is there Martin JT to do odd

odd

a couple of interesting other talks to

follow on

tdd Revisited is a talk that in Cooper

did for the first virtual NDC back in 2021

2021

um which a lot of the concepts that I've

come up with about how we decompose tdd

into something that's useful come from

talking to Ian and that talk it's a

really really interesting talk about

what tdd is and is not

the other one is David Whitney talking

about writing tests that don't suck

which is a really interesting talk about

the refactoring and naming stuff that

we're talking about and extracting

context run from a JavaScript context

but it's exactly the same context me and

David worked on this for the last two

years so all of those concepts are very

very the same um

okay sales spiel

we have a free forever account with

honeycomb so the things that we were

talking about where you put open

Telemetry in the end of your application

you can actually send all of that to

Honeycomb for free for yourself locally

if you want to

we allow 20 million events to come

through US per month for free forever

if you're writing more than 20 million

spans in your tests per month let's talk

because I'd like to know what you're doing

doing

um we also have we wrote the book on

observability engineering it's free to

download which you can get on the

website which is not honeycomb at all

this is all about how observability

Works what observability is and there is

a full chapter on odd in there

but yeah that's me I'm happy to take questions

questions

um we have a ton of stickers

um we had a thing where people said oh

we'll do a sticker exchange and see

whose stickers are best and my response

was what's it like to lose

um all the best stickers are on our

booth out there come and take some

stickers or come talk to me you don't

have to talk to me just take the

stickers that's fine

um but yeah I'll take questions now but

come by the booth come say hello if you

want to demo if honeycomb well happy to

do it [Applause]

oh and also please do do the ratings

outside it really helps me if there's

specific feedback of stuff you wanted to

see on and didn't that kind of stuff do

tell me ping me on Twitter it all helps

other people who are seeing these

presentations to make them better do it

for me do it for all the other speakers

Click on any text or timestamp to jump to that moment in the video

Most transcripts ready in under 5 seconds

One-Click Copy125+ LanguagesSearch ContentJump to Timestamps

Paste YouTube URL

Enter any YouTube video link to get the full transcript

Most transcripts ready in under 5 seconds

Get Our Chrome Extension

Get transcripts instantly without leaving YouTube. Install our Chrome extension for one-click access to any video's transcript directly on the watch page.

Add to Chrome — Free

Works with YouTube, Coursera, Udemy and more educational platforms

Get Instant Transcripts: Just Edit the Domain in Your Address Bar!

YouTube

←

→

↻

https://www.youtube.com/watch?v=UF8uR6Z6KLc

YoutubeToText

←

→

↻

https://youtubetotext.net/watch?v=UF8uR6Z6KLc

YouTube TranscriptPreparing your results…

YouTube Transcript:Building Operable Software with TDD (but not the way you think) - Martin Thwaites - NDC London 2023

Video Transcript

Paste YouTube URL

Transcript Extraction Form

Get Our Chrome Extension

Get Instant Transcripts: Just Edit the Domain in Your Address Bar!

YouTube Transcript:
Building Operable Software with TDD (but not the way you think) - Martin Thwaites - NDC London 2023