1
00:00:00,480 --> 00:00:03,480
foreign

2
00:00:08,220 --> 00:00:13,740
had a lovely lunch

3
00:00:11,219 --> 00:00:15,480
uh here we're about to hear from huon

4
00:00:13,740 --> 00:00:16,859
who has done work for some very big

5
00:00:15,480 --> 00:00:20,100
names including

6
00:00:16,859 --> 00:00:22,320
Apple Cyro and so on he's going to give

7
00:00:20,100 --> 00:00:24,320
us a talk about

8
00:00:22,320 --> 00:00:27,119
packaging with

9
00:00:24,320 --> 00:00:30,119
serverless for serverless because that

10
00:00:27,119 --> 00:00:32,279
must be easy right we've all done that

11
00:00:30,119 --> 00:00:33,430
all right if you're ready yeah take it

12
00:00:32,279 --> 00:00:36,479
away

13
00:00:33,430 --> 00:00:36,479
[Music]

14
00:00:39,540 --> 00:00:42,480
thank you

15
00:00:40,920 --> 00:00:44,760
you know that nightmare where you're

16
00:00:42,480 --> 00:00:46,440
standing in public speaking in public

17
00:00:44,760 --> 00:00:49,680
and you realize you're not wearing any

18
00:00:46,440 --> 00:00:51,180
clothes at EXO flare we were living in a

19
00:00:49,680 --> 00:00:53,579
bit of a nightmare it was taking

20
00:00:51,180 --> 00:00:56,039
embarrassingly in frustratingly long to

21
00:00:53,579 --> 00:00:57,600
build our artifacts for our AWS

22
00:00:56,039 --> 00:01:00,000
standards

23
00:00:57,600 --> 00:01:02,579
but we woke up and we put our pants on

24
00:01:00,000 --> 00:01:06,420
by adopting the pants build system we

25
00:01:02,579 --> 00:01:06,420
made that process effortless

26
00:01:06,960 --> 00:01:11,520
I'm hewan Wilson and this is Sterling

27
00:01:09,240 --> 00:01:13,740
our Corgi

28
00:01:11,520 --> 00:01:16,260
at the moment I'm working for exoflare

29
00:01:13,740 --> 00:01:18,840
which is a software company trying to

30
00:01:16,260 --> 00:01:21,960
make biosecurity easy so we're reducing

31
00:01:18,840 --> 00:01:24,780
the risk and prosperity of animal

32
00:01:21,960 --> 00:01:26,580
diseases that's what we're trying to do

33
00:01:24,780 --> 00:01:28,560
I'm mostly writing python as a back-end

34
00:01:26,580 --> 00:01:31,619
engineer but I'm also doing front end

35
00:01:28,560 --> 00:01:33,780
I'm doing uh data science and I'm doing

36
00:01:31,619 --> 00:01:36,060
devops

37
00:01:33,780 --> 00:01:37,799
that's a bit of a disclaimer I'm now a

38
00:01:36,060 --> 00:01:40,680
pants maintainer so this talk will be

39
00:01:37,799 --> 00:01:43,020
about this pants build system tool but

40
00:01:40,680 --> 00:01:45,060
when we were evaluating paths to work

41
00:01:43,020 --> 00:01:46,799
out what tools to use I had no

42
00:01:45,060 --> 00:01:49,439
connection to it as part of that

43
00:01:46,799 --> 00:01:51,420
adoption process I found enough bugs

44
00:01:49,439 --> 00:01:54,780
contributed enough fixes that I was

45
00:01:51,420 --> 00:01:57,299
nominated to become a maintainer as well

46
00:01:54,780 --> 00:01:59,700
for exoflare I've worked on compilers

47
00:01:57,299 --> 00:02:04,040
and developer tools at Mozilla and apple

48
00:01:59,700 --> 00:02:04,040
and machine learning at csiro

49
00:02:04,920 --> 00:02:09,899
over the next half an hour we'll remind

50
00:02:06,899 --> 00:02:12,239
ourselves what AWS Lambda and similar

51
00:02:09,899 --> 00:02:14,220
products are are

52
00:02:12,239 --> 00:02:16,200
and we'll work out way Python and

53
00:02:14,220 --> 00:02:17,640
interpreted language needs a build

54
00:02:16,200 --> 00:02:19,500
system

55
00:02:17,640 --> 00:02:21,780
then we'll look at the conventional

56
00:02:19,500 --> 00:02:23,459
wisdom for doing that build system which

57
00:02:21,780 --> 00:02:26,099
is using pip install and we'll see why

58
00:02:23,459 --> 00:02:29,340
it's slow and fiddly

59
00:02:26,099 --> 00:02:32,879
and then I'll talk about the solution we

60
00:02:29,340 --> 00:02:36,060
we really enjoy which is using pads

61
00:02:32,879 --> 00:02:38,099
sound good let's get into it

62
00:02:36,060 --> 00:02:39,660
functions as a service is a bit of a

63
00:02:38,099 --> 00:02:40,739
silly name but it seems to be the

64
00:02:39,660 --> 00:02:43,319
accepted

65
00:02:40,739 --> 00:02:45,540
category name for tools like AWS Lambda

66
00:02:43,319 --> 00:02:47,160
Google Cloud functions

67
00:02:45,540 --> 00:02:50,040
um whatever the other Cloud providers

68
00:02:47,160 --> 00:02:52,800
offer in this space

69
00:02:50,040 --> 00:02:54,120
at exoflare we mostly use AWS so I'm

70
00:02:52,800 --> 00:02:55,739
going to say Lambda throughout the talk

71
00:02:54,120 --> 00:02:57,239
but I think most of the lessons will

72
00:02:55,739 --> 00:02:59,640
apply to the different products I just

73
00:02:57,239 --> 00:03:01,680
can't speak to them personally

74
00:02:59,640 --> 00:03:03,780
so Lambda is effectively a snippet of

75
00:03:01,680 --> 00:03:06,180
code it could be a dozen lines it could

76
00:03:03,780 --> 00:03:07,379
be a dozen megabytes that you upload to

77
00:03:06,180 --> 00:03:09,060
the cloud provider and then they'll

78
00:03:07,379 --> 00:03:10,620
manage sort of invoking it for you and

79
00:03:09,060 --> 00:03:11,700
setting up the interpreters and all that

80
00:03:10,620 --> 00:03:13,920
stuff

81
00:03:11,700 --> 00:03:16,440
we use it for things like cron jobs so

82
00:03:13,920 --> 00:03:18,420
some cleanup that happens every hour

83
00:03:16,440 --> 00:03:19,800
we use it for some event processing and

84
00:03:18,420 --> 00:03:22,620
we use it for integrating with the cloud

85
00:03:19,800 --> 00:03:24,360
provider so when we do our deploy with a

86
00:03:22,620 --> 00:03:26,940
crowd formation we have some custom

87
00:03:24,360 --> 00:03:30,300
resources and AWS makes it easy to back

88
00:03:26,940 --> 00:03:32,760
those custom Resources with lambdas

89
00:03:30,300 --> 00:03:34,620
this talk is about serverless but it's

90
00:03:32,760 --> 00:03:36,300
not saying microservices save the world

91
00:03:34,620 --> 00:03:37,019
or that you should use serverless for

92
00:03:36,300 --> 00:03:39,480
everything

93
00:03:37,019 --> 00:03:41,819
we use Docker images for other things we

94
00:03:39,480 --> 00:03:45,599
only have a few Landers for different

95
00:03:41,819 --> 00:03:48,840
um different uh problems we solve

96
00:03:45,599 --> 00:03:51,180
this talk is about if you are using

97
00:03:48,840 --> 00:03:53,159
Lambda or other serverless offerings

98
00:03:51,180 --> 00:03:54,540
maybe you can learn from the suffering

99
00:03:53,159 --> 00:03:56,720
we went through and the solution we

100
00:03:54,540 --> 00:03:56,720
found

101
00:03:57,540 --> 00:04:00,959
let's make this concrete so here's an

102
00:03:59,459 --> 00:04:03,360
example of a Lambda

103
00:04:00,959 --> 00:04:06,180
the details don't matter too much but it

104
00:04:03,360 --> 00:04:07,860
takes in a bit of Json message then some

105
00:04:06,180 --> 00:04:11,340
string and then we'll return that string

106
00:04:07,860 --> 00:04:13,860
so message hello returns hello

107
00:04:11,340 --> 00:04:15,480
I personally don't like passing Jason

108
00:04:13,860 --> 00:04:17,820
myself I don't want to like check that

109
00:04:15,480 --> 00:04:19,680
the message key exists I don't want to

110
00:04:17,820 --> 00:04:22,680
check that the value is the string

111
00:04:19,680 --> 00:04:24,540
so we Define use one of these libraries

112
00:04:22,680 --> 00:04:27,120
a bit like pedantic that was referenced

113
00:04:24,540 --> 00:04:28,400
in earlier talks that lets us Define the

114
00:04:27,120 --> 00:04:31,440
structures

115
00:04:28,400 --> 00:04:32,880
more declaratively

116
00:04:31,440 --> 00:04:34,680
the last piece this is a Handler

117
00:04:32,880 --> 00:04:36,180
function so this is what AWS will invoke

118
00:04:34,680 --> 00:04:37,380
to actually get our Lander to do its

119
00:04:36,180 --> 00:04:39,000
thing

120
00:04:37,380 --> 00:04:41,639
the details of that don't matter too

121
00:04:39,000 --> 00:04:43,800
much but what does matter is that this

122
00:04:41,639 --> 00:04:45,360
message spec Library so the the thing

123
00:04:43,800 --> 00:04:47,880
I'm using for manipulating the Json

124
00:04:45,360 --> 00:04:49,139
that's an external Library it's from Pi

125
00:04:47,880 --> 00:04:50,759
Pi

126
00:04:49,139 --> 00:04:53,100
when we have an external Library we need

127
00:04:50,759 --> 00:04:55,440
to specify our dependency somehow so we

128
00:04:53,100 --> 00:04:57,840
could use pipeproject.tomel we could use

129
00:04:55,440 --> 00:04:58,919
a pip file but we'll keep it simple for

130
00:04:57,840 --> 00:05:01,139
these examples and just use

131
00:04:58,919 --> 00:05:03,139
requirements.txt

132
00:05:01,139 --> 00:05:05,220
so this defines our Lambda we've got

133
00:05:03,139 --> 00:05:06,419
requirements.txt with our message spec

134
00:05:05,220 --> 00:05:09,860
dependency

135
00:05:06,419 --> 00:05:09,860
and we've got our actual source code

136
00:05:10,259 --> 00:05:13,860
sounds pretty good and we can

137
00:05:11,940 --> 00:05:15,600
hypothetically get this into AWS somehow

138
00:05:13,860 --> 00:05:19,100
and invoke it and it would give the

139
00:05:15,600 --> 00:05:19,100
right behavior in theory

140
00:05:19,680 --> 00:05:24,840
but you'll notice that the the way to

141
00:05:22,500 --> 00:05:27,479
get into AWS is either a Docker image

142
00:05:24,840 --> 00:05:29,160
which has some downsides or zip files

143
00:05:27,479 --> 00:05:31,440
we use zip files so that's what I'm

144
00:05:29,160 --> 00:05:33,120
going to talk about in this talk

145
00:05:31,440 --> 00:05:35,940
the zip file has to have a particular

146
00:05:33,120 --> 00:05:37,620
format so the format has to be

147
00:05:35,940 --> 00:05:39,419
of course our source code has to be in

148
00:05:37,620 --> 00:05:42,900
there somehow so we've got Echo dot Pi

149
00:05:39,419 --> 00:05:44,400
in our echo.zip for our Echo Lambda

150
00:05:42,900 --> 00:05:46,620
and

151
00:05:44,400 --> 00:05:48,600
we have to include all of the source

152
00:05:46,620 --> 00:05:50,820
code of the dependencies so our

153
00:05:48,600 --> 00:05:51,960
echolanda uses the message spec

154
00:05:50,820 --> 00:05:53,699
dependency

155
00:05:51,960 --> 00:05:55,500
and thus we need to include its source

156
00:05:53,699 --> 00:05:57,120
code

157
00:05:55,500 --> 00:05:59,220
this is why we need a build system for

158
00:05:57,120 --> 00:06:01,340
python so what we had in our repository

159
00:05:59,220 --> 00:06:03,780
that was Echo dot pi and

160
00:06:01,340 --> 00:06:05,759
requirements.txt that's different to

161
00:06:03,780 --> 00:06:08,340
what goes in the zip file so we need

162
00:06:05,759 --> 00:06:12,360
some sort of a system for building a zip

163
00:06:08,340 --> 00:06:14,460
file that has the right format

164
00:06:12,360 --> 00:06:17,100
suppose we have a system then we get a

165
00:06:14,460 --> 00:06:19,440
zip file and we can upload that to AWS

166
00:06:17,100 --> 00:06:20,699
somehow so we could use the console we

167
00:06:19,440 --> 00:06:23,100
could use the command line interface we

168
00:06:20,699 --> 00:06:24,479
could use terraform we could use cdk

169
00:06:23,100 --> 00:06:27,000
there's all sorts of tools that will let

170
00:06:24,479 --> 00:06:29,039
us take code on our disk and put it into

171
00:06:27,000 --> 00:06:30,720
the into the cloud but that's not the

172
00:06:29,039 --> 00:06:32,699
focus of this talk the focus of this

173
00:06:30,720 --> 00:06:35,100
talk is turning our source code in our

174
00:06:32,699 --> 00:06:38,960
repository into a zip file ready to

175
00:06:35,100 --> 00:06:38,960
upload into the cloud provider

176
00:06:40,039 --> 00:06:45,360
if we look through the AWS documentation

177
00:06:42,660 --> 00:06:47,759
we can work out a system for doing this

178
00:06:45,360 --> 00:06:50,039
so this is a script up to still down

179
00:06:47,759 --> 00:06:53,100
first we go pip install our requirements

180
00:06:50,039 --> 00:06:54,720
into the the package folder then we can

181
00:06:53,100 --> 00:06:57,300
copy in our own source code so copy

182
00:06:54,720 --> 00:06:59,280
echo.pi into that folder and then at the

183
00:06:57,300 --> 00:07:01,020
end we zip it all up this gives us

184
00:06:59,280 --> 00:07:02,880
echo.zip

185
00:07:01,020 --> 00:07:05,100
if we were to look into this ZIP file

186
00:07:02,880 --> 00:07:08,520
just like with our normal computer

187
00:07:05,100 --> 00:07:10,979
investigation skills we could

188
00:07:08,520 --> 00:07:12,419
notice that it has basically the format

189
00:07:10,979 --> 00:07:14,220
that we were just discussing so we'll

190
00:07:12,419 --> 00:07:15,660
have our echo.pi in it and it will have

191
00:07:14,220 --> 00:07:18,600
the source code of the message spec

192
00:07:15,660 --> 00:07:21,300
dependency in it too

193
00:07:18,600 --> 00:07:23,819
that's uploaded to AWS let's invoke it

194
00:07:21,300 --> 00:07:25,860
and oh we get an error

195
00:07:23,819 --> 00:07:27,120
no module named massive respect dot

196
00:07:25,860 --> 00:07:28,740
underscore core

197
00:07:27,120 --> 00:07:30,479
that's a bit weird because when we look

198
00:07:28,740 --> 00:07:33,300
in the zip file there is a file that

199
00:07:30,479 --> 00:07:35,220
looks pretty related message back flash

200
00:07:33,300 --> 00:07:37,860
underscore core

201
00:07:35,220 --> 00:07:39,360
that that sounds like it matches up

202
00:07:37,860 --> 00:07:41,639
but a bit further in the file name it

203
00:07:39,360 --> 00:07:44,699
says Darwin Darwin's another name for

204
00:07:41,639 --> 00:07:46,440
the Mac OS operating system

205
00:07:44,699 --> 00:07:48,660
Lambda is running the Linux operating

206
00:07:46,440 --> 00:07:50,759
system so when I ran these commands this

207
00:07:48,660 --> 00:07:52,560
pip install the build script

208
00:07:50,759 --> 00:07:55,220
on my Mac machine

209
00:07:52,560 --> 00:07:57,660
it shows code that is specific to Mac

210
00:07:55,220 --> 00:07:59,759
and that won't work in Lambda

211
00:07:57,660 --> 00:08:01,560
understandably the Lambda python

212
00:07:59,759 --> 00:08:03,780
interpreter is ignoring this module

213
00:08:01,560 --> 00:08:07,520
doesn't want to try and run Mac specific

214
00:08:03,780 --> 00:08:07,520
code on the wrong operating system

215
00:08:08,160 --> 00:08:13,080
okay well we can fix this using Docker

216
00:08:11,099 --> 00:08:15,259
Docker for Mac makes it really easy to

217
00:08:13,080 --> 00:08:17,819
run Linux

218
00:08:15,259 --> 00:08:19,020
virtual machines commands on Linux while

219
00:08:17,819 --> 00:08:21,060
on Mac

220
00:08:19,020 --> 00:08:23,639
so we can go Docker run some arguments

221
00:08:21,060 --> 00:08:25,979
to get our source code in so choosing an

222
00:08:23,639 --> 00:08:29,520
image one provided by AWS seems pretty

223
00:08:25,979 --> 00:08:31,440
plausible use our script if we run this

224
00:08:29,520 --> 00:08:34,500
we get out another zip file we get our

225
00:08:31,440 --> 00:08:37,320
echo.zip version two I guess

226
00:08:34,500 --> 00:08:39,419
when we upload this to AWS and we invoke

227
00:08:37,320 --> 00:08:41,580
this we get

228
00:08:39,419 --> 00:08:43,140
the same error

229
00:08:41,580 --> 00:08:44,760
but we know where to look now message

230
00:08:43,140 --> 00:08:47,160
spec slash underscore call blah blah

231
00:08:44,760 --> 00:08:48,360
Linux so we've got the right operating

232
00:08:47,160 --> 00:08:51,660
system

233
00:08:48,360 --> 00:08:55,140
but it says Ah 64. that's a reference to

234
00:08:51,660 --> 00:08:57,200
the arm CPU architecture

235
00:08:55,140 --> 00:09:00,899
lambdas at least by default will run

236
00:08:57,200 --> 00:09:02,880
x8664 or Intel CPU so we've fixed our

237
00:09:00,899 --> 00:09:05,940
operating system mismatch

238
00:09:02,880 --> 00:09:08,279
but we still have a CPU mismatch

239
00:09:05,940 --> 00:09:10,440
the fix is easy enough we had yet

240
00:09:08,279 --> 00:09:13,200
another argument to our Docker command

241
00:09:10,440 --> 00:09:15,060
We Run The build we upload to AWS we

242
00:09:13,200 --> 00:09:16,860
invoke it and we're Victorious we

243
00:09:15,060 --> 00:09:19,019
finally have a working Lambda so we

244
00:09:16,860 --> 00:09:20,399
invoke it with message hello we get back

245
00:09:19,019 --> 00:09:22,980
hello

246
00:09:20,399 --> 00:09:24,600
very cool we now have somewhat of a

247
00:09:22,980 --> 00:09:26,040
build system we could probably make it a

248
00:09:24,600 --> 00:09:28,620
bit more General if we wanted to to

249
00:09:26,040 --> 00:09:30,240
actually be a system but we've got

250
00:09:28,620 --> 00:09:33,660
something that can produce functional

251
00:09:30,240 --> 00:09:35,399
zip files that work in Lambda

252
00:09:33,660 --> 00:09:36,600
unfortunately functional is just the

253
00:09:35,399 --> 00:09:39,120
minimum Bar for a build system

254
00:09:36,600 --> 00:09:42,300
performance is important too

255
00:09:39,120 --> 00:09:43,980
no one wants a slow build right

256
00:09:42,300 --> 00:09:46,260
this is the performance of the three

257
00:09:43,980 --> 00:09:48,060
different steps we ran so if you say

258
00:09:46,260 --> 00:09:50,760
that running natively on my Mac the very

259
00:09:48,060 --> 00:09:52,260
first thing we did I about half a second

260
00:09:50,760 --> 00:09:54,839
but it gave us a zip file that didn't

261
00:09:52,260 --> 00:09:57,420
work so it's just taunting us

262
00:09:54,839 --> 00:09:59,640
running on the arm docker took three

263
00:09:57,420 --> 00:10:01,560
times longer but again it's taunting us

264
00:09:59,640 --> 00:10:03,300
because the zip file doesn't work to

265
00:10:01,560 --> 00:10:04,740
actually get a working zip that gave us

266
00:10:03,300 --> 00:10:06,600
a working Lambda

267
00:10:04,740 --> 00:10:08,160
takes us five seconds

268
00:10:06,600 --> 00:10:10,620
this is because it's having to run a

269
00:10:08,160 --> 00:10:14,600
virtual machine that's emulating the CPU

270
00:10:10,620 --> 00:10:14,600
this comes with significant overhead

271
00:10:14,880 --> 00:10:20,839
five seconds would be tolerable if it

272
00:10:17,040 --> 00:10:20,839
was only five seconds but it gets worse

273
00:10:21,180 --> 00:10:24,720
imagine we're heading a second Lambda so

274
00:10:22,860 --> 00:10:26,700
a feature request has come in that we

275
00:10:24,720 --> 00:10:29,760
need to do some reporting that reporting

276
00:10:26,700 --> 00:10:31,620
is generating a plot so let's add we

277
00:10:29,760 --> 00:10:33,420
start with our echo.pi and we've got our

278
00:10:31,620 --> 00:10:36,060
requirement subtext let's add a new file

279
00:10:33,420 --> 00:10:37,980
plot.pi we're generating a plot so we'll

280
00:10:36,060 --> 00:10:40,200
import matplotlib

281
00:10:37,980 --> 00:10:44,100
another library from Pi Pi that is

282
00:10:40,200 --> 00:10:46,079
helpful for generating plots in the name

283
00:10:44,100 --> 00:10:47,640
it's from Pi Pi so we better add it as a

284
00:10:46,079 --> 00:10:48,720
requirement

285
00:10:47,640 --> 00:10:50,760
um

286
00:10:48,720 --> 00:10:52,860
so this is now our pair of lambdas we've

287
00:10:50,760 --> 00:10:55,440
got echo.pi for our Echo Lambda we've

288
00:10:52,860 --> 00:10:56,820
got plot.pi for our plot Lambda and

289
00:10:55,440 --> 00:10:58,500
we've got the requirements required for

290
00:10:56,820 --> 00:11:00,720
both of them

291
00:10:58,500 --> 00:11:03,180
we can tweak our build system to also be

292
00:11:00,720 --> 00:11:06,560
able to build the plot Lambda and look

293
00:11:03,180 --> 00:11:06,560
at its performance

294
00:11:06,660 --> 00:11:10,140
it's got far worse

295
00:11:08,339 --> 00:11:12,420
running natively on Macs that's as fast

296
00:11:10,140 --> 00:11:14,160
as we can go it takes about seven and a

297
00:11:12,420 --> 00:11:16,800
half seconds

298
00:11:14,160 --> 00:11:19,459
but to get a zip file that works because

299
00:11:16,800 --> 00:11:22,980
this plot Lambda has exactly the same

300
00:11:19,459 --> 00:11:24,959
platform dependent code issue

301
00:11:22,980 --> 00:11:27,240
to get a zip file that works it takes

302
00:11:24,959 --> 00:11:29,459
100 seconds

303
00:11:27,240 --> 00:11:32,040
that's a very long time and no one wants

304
00:11:29,459 --> 00:11:34,079
to wait for a build that long

305
00:11:32,040 --> 00:11:35,519
why is this one so much slower so our

306
00:11:34,079 --> 00:11:36,899
Echo Lander used to build in five

307
00:11:35,519 --> 00:11:40,200
seconds this one builds in 100 seconds

308
00:11:36,899 --> 00:11:42,420
well the plot Lambda is using matplotlib

309
00:11:40,200 --> 00:11:44,940
matplotlib refers to other labs

310
00:11:42,420 --> 00:11:46,740
libraries like numpy

311
00:11:44,940 --> 00:11:49,560
what this means is that the final zip

312
00:11:46,740 --> 00:11:51,899
package is almost 50 megabytes

313
00:11:49,560 --> 00:11:54,480
for the echolanda that we had before

314
00:11:51,899 --> 00:11:55,680
it was like 200 kilobytes that is a huge

315
00:11:54,480 --> 00:11:57,180
amount larger

316
00:11:55,680 --> 00:11:59,279
manipulating all that extra data is

317
00:11:57,180 --> 00:12:02,000
going to take extra time and we can see

318
00:11:59,279 --> 00:12:02,000
it on this graph

319
00:12:02,040 --> 00:12:05,220
this graph is nominally about the plot

320
00:12:03,959 --> 00:12:07,019
Lambda

321
00:12:05,220 --> 00:12:09,240
but it's actually

322
00:12:07,019 --> 00:12:11,279
by adding the new dependency to our

323
00:12:09,240 --> 00:12:13,980
requirements file we've made the

324
00:12:11,279 --> 00:12:16,260
echolanda really slow too

325
00:12:13,980 --> 00:12:18,300
when we build the echolanda we installed

326
00:12:16,260 --> 00:12:19,800
the requirements from this file so we

327
00:12:18,300 --> 00:12:21,240
installed message spec that's what the

328
00:12:19,800 --> 00:12:23,760
echolanda needs

329
00:12:21,240 --> 00:12:25,740
but we also install matplotlib

330
00:12:23,760 --> 00:12:27,720
even though the echo Lambda doesn't need

331
00:12:25,740 --> 00:12:30,180
map

332
00:12:27,720 --> 00:12:32,220
so we spend 100 seconds manipulating

333
00:12:30,180 --> 00:12:34,140
installing matplotlib

334
00:12:32,220 --> 00:12:37,200
for no reason

335
00:12:34,140 --> 00:12:39,480
not only that but the echolanda zip file

336
00:12:37,200 --> 00:12:41,279
went from 200 kilobytes to almost 50

337
00:12:39,480 --> 00:12:43,800
megabytes

338
00:12:41,279 --> 00:12:46,680
how large is it means slower uploads to

339
00:12:43,800 --> 00:12:48,420
AWS and it also means slower cold starts

340
00:12:46,680 --> 00:12:51,240
so when a Lambda starts for the first

341
00:12:48,420 --> 00:12:53,100
time it has to initialize their the

342
00:12:51,240 --> 00:12:56,339
system and that takes longer when you

343
00:12:53,100 --> 00:12:59,519
have a larger deployment package

344
00:12:56,339 --> 00:13:01,139
so preferably we're not we would prefer

345
00:12:59,519 --> 00:13:03,300
not to be spending 100 seconds building

346
00:13:01,139 --> 00:13:04,500
Echo pointlessly and we would probably

347
00:13:03,300 --> 00:13:07,500
also prefer not to be spending 100

348
00:13:04,500 --> 00:13:09,480
seconds building the plot Lambda

349
00:13:07,500 --> 00:13:11,639
we can solve one of those problems by

350
00:13:09,480 --> 00:13:14,339
splitting our requirements so on the

351
00:13:11,639 --> 00:13:17,339
left we've got our echolanda and we

352
00:13:14,339 --> 00:13:18,839
could have our Echo dot Pi file plus the

353
00:13:17,339 --> 00:13:20,579
requirements that that file made

354
00:13:18,839 --> 00:13:22,860
specifically

355
00:13:20,579 --> 00:13:24,360
and on the right we have our plot Lander

356
00:13:22,860 --> 00:13:25,920
and the requirements plot needs

357
00:13:24,360 --> 00:13:28,920
specifically

358
00:13:25,920 --> 00:13:30,839
if we do this now the echo Lambda will

359
00:13:28,920 --> 00:13:32,100
be back to building in five seconds

360
00:13:30,839 --> 00:13:33,839
the plot land will still take 100

361
00:13:32,100 --> 00:13:35,220
seconds to build but like at least with

362
00:13:33,839 --> 00:13:38,040
half the time that the total build

363
00:13:35,220 --> 00:13:40,200
process will take

364
00:13:38,040 --> 00:13:41,700
this can work

365
00:13:40,200 --> 00:13:43,560
but it gets harder as soon as we start

366
00:13:41,700 --> 00:13:45,959
trying to share code so imagine we

367
00:13:43,560 --> 00:13:47,579
wanted to do logging in the same way

368
00:13:45,959 --> 00:13:52,139
from both lambdas

369
00:13:47,579 --> 00:13:54,060
so we have some logs.pi utilities

370
00:13:52,139 --> 00:13:55,440
and in particular we want to do

371
00:13:54,060 --> 00:13:59,519
structured logging

372
00:13:55,440 --> 00:14:01,260
so log stop Pi Imports strap lock

373
00:13:59,519 --> 00:14:02,820
struck log is yet another library from

374
00:14:01,260 --> 00:14:04,380
Pi Pi

375
00:14:02,820 --> 00:14:05,459
so we better make sure it's installed in

376
00:14:04,380 --> 00:14:07,560
the lambdas

377
00:14:05,459 --> 00:14:09,839
if it's not included in the Lambda zip

378
00:14:07,560 --> 00:14:12,839
file then Echo dot Pi will import

379
00:14:09,839 --> 00:14:14,700
logs.pi logs.pi tries to import struck

380
00:14:12,839 --> 00:14:16,740
log struct log isn't available and the

381
00:14:14,700 --> 00:14:18,300
Lambda crashes

382
00:14:16,740 --> 00:14:21,180
one way to do this would be to remember

383
00:14:18,300 --> 00:14:23,220
which lambdas we need to which lambdas

384
00:14:21,180 --> 00:14:25,800
are using logs.pi and go and add it to

385
00:14:23,220 --> 00:14:28,079
all of those requirements

386
00:14:25,800 --> 00:14:30,779
but whenever we say remember and expect

387
00:14:28,079 --> 00:14:32,220
humans to do it that's always a losing

388
00:14:30,779 --> 00:14:33,600
proposition it's easy for us to forget

389
00:14:32,220 --> 00:14:35,639
to do something

390
00:14:33,600 --> 00:14:40,100
like if we hadn't added it to the echo

391
00:14:35,639 --> 00:14:40,100
requirements then Echo would crash

392
00:14:40,800 --> 00:14:44,519
this is fiddly this is annoying and it's

393
00:14:43,260 --> 00:14:46,199
still not solving the problem with the

394
00:14:44,519 --> 00:14:47,579
virtual machines

395
00:14:46,199 --> 00:14:49,199
there's other ways to solve this problem

396
00:14:47,579 --> 00:14:50,639
too like you could have sort of a shared

397
00:14:49,199 --> 00:14:52,860
set of requirements so you have your

398
00:14:50,639 --> 00:14:54,180
logs.pi with its own requirements and

399
00:14:52,860 --> 00:14:55,800
then when you build the echo land you

400
00:14:54,180 --> 00:14:58,800
install the shared requirements and the

401
00:14:55,800 --> 00:15:00,060
requirements specific to to Echo and

402
00:14:58,800 --> 00:15:02,760
that's basically what we were doing at

403
00:15:00,060 --> 00:15:06,380
exoflare but even that is fiddly and

404
00:15:02,760 --> 00:15:06,380
definitely doesn't spark Joy

405
00:15:06,779 --> 00:15:10,560
so we've looked through the conventional

406
00:15:08,519 --> 00:15:15,120
wisdom and we've looked at it in detail

407
00:15:10,560 --> 00:15:17,040
there's tools like AWS cdk offers sort

408
00:15:15,120 --> 00:15:19,320
of Packaging

409
00:15:17,040 --> 00:15:20,820
um for python where it can wrap up this

410
00:15:19,320 --> 00:15:22,380
process but under the hood it's

411
00:15:20,820 --> 00:15:23,880
basically doing the same thing last time

412
00:15:22,380 --> 00:15:25,920
I checked it was still using Docker to

413
00:15:23,880 --> 00:15:28,199
run pip install

414
00:15:25,920 --> 00:15:30,600
the impact of this is our cross-platform

415
00:15:28,199 --> 00:15:33,060
builds are really slow

416
00:15:30,600 --> 00:15:34,680
and when we have multiple lambdas with

417
00:15:33,060 --> 00:15:36,839
different dependencies

418
00:15:34,680 --> 00:15:39,000
we can either end up with unnecessarily

419
00:15:36,839 --> 00:15:40,740
slow builds and unnecessarily huge Zips

420
00:15:39,000 --> 00:15:42,000
if we include dependencies a Lambda

421
00:15:40,740 --> 00:15:44,339
doesn't need

422
00:15:42,000 --> 00:15:46,019
or we end up with a world of really

423
00:15:44,339 --> 00:15:48,000
fiddly dependency management and having

424
00:15:46,019 --> 00:15:49,680
to remember to update to updates 10

425
00:15:48,000 --> 00:15:51,720
different files whenever we make a

426
00:15:49,680 --> 00:15:53,519
change

427
00:15:51,720 --> 00:15:55,620
at EXO flight we were trying to optimize

428
00:15:53,519 --> 00:15:58,440
our build times with that fiddly

429
00:15:55,620 --> 00:15:59,519
approach but we still had so it took us

430
00:15:58,440 --> 00:16:02,339
three minutes to build our

431
00:15:59,519 --> 00:16:04,079
infrastructure before deploying and

432
00:16:02,339 --> 00:16:06,240
worth when we were trying to test it

433
00:16:04,079 --> 00:16:07,680
locally it'll take five minutes to test

434
00:16:06,240 --> 00:16:11,300
our infrastructure

435
00:16:07,680 --> 00:16:11,300
that is not very pleasant

436
00:16:11,579 --> 00:16:16,440
so we looked around for Solutions and we

437
00:16:14,100 --> 00:16:18,839
settled on the pants tool bills itself

438
00:16:16,440 --> 00:16:20,699
is the economic build system and that

439
00:16:18,839 --> 00:16:23,339
seems to be true so we use it for things

440
00:16:20,699 --> 00:16:25,740
like running tests running code linting

441
00:16:23,339 --> 00:16:28,380
running code formatting but most

442
00:16:25,740 --> 00:16:30,600
importantly building artifacts so pants

443
00:16:28,380 --> 00:16:33,120
is the build tool we use for turning our

444
00:16:30,600 --> 00:16:36,060
source code into zip files ready to

445
00:16:33,120 --> 00:16:38,160
upload to AWS

446
00:16:36,060 --> 00:16:39,360
as we're adopting pants and evaluating

447
00:16:38,160 --> 00:16:42,320
different tools we of course ask

448
00:16:39,360 --> 00:16:45,180
questions like what tools are available

449
00:16:42,320 --> 00:16:46,980
how do we use these tools and how do we

450
00:16:45,180 --> 00:16:48,959
adopt them incrementally to start seeing

451
00:16:46,980 --> 00:16:51,180
benefits earlier

452
00:16:48,959 --> 00:16:52,320
but we then asked Howard Sterling wear

453
00:16:51,180 --> 00:16:54,600
pants

454
00:16:52,320 --> 00:16:56,279
my wife thinks number three I reckon

455
00:16:54,600 --> 00:16:59,420
number seven but I'm told that those are

456
00:16:56,279 --> 00:16:59,420
overalls not the paths

457
00:17:04,319 --> 00:17:08,339
but more seriously we asked what

458
00:17:06,240 --> 00:17:10,199
performance Improvement would we see so

459
00:17:08,339 --> 00:17:11,819
this is the examples we had the example

460
00:17:10,199 --> 00:17:13,799
lambdas we had before

461
00:17:11,819 --> 00:17:15,720
it would take five seconds to build the

462
00:17:13,799 --> 00:17:18,000
echolanda using docker

463
00:17:15,720 --> 00:17:19,079
with pants takes two seconds that's a

464
00:17:18,000 --> 00:17:21,540
nice speed up

465
00:17:19,079 --> 00:17:23,400
but more significantly the plot Lambda

466
00:17:21,540 --> 00:17:24,480
will take 100 seconds to build with

467
00:17:23,400 --> 00:17:26,400
docker

468
00:17:24,480 --> 00:17:28,799
it now takes less than eight

469
00:17:26,400 --> 00:17:30,540
that's a huge speed up

470
00:17:28,799 --> 00:17:32,640
most of that hundred seconds is spent

471
00:17:30,540 --> 00:17:35,760
manipulating dependencies

472
00:17:32,640 --> 00:17:37,799
many playing Matlock modifying numpy

473
00:17:35,760 --> 00:17:39,299
pants is able to do this natively

474
00:17:37,799 --> 00:17:41,340
without having to run in a virtual

475
00:17:39,299 --> 00:17:44,220
machine and without having to run and

476
00:17:41,340 --> 00:17:46,020
without having to emulate the CPU

477
00:17:44,220 --> 00:17:47,700
so we're not having to we're able to run

478
00:17:46,020 --> 00:17:50,400
natively on Mac and see that really fast

479
00:17:47,700 --> 00:17:52,500
performance instead of having to um run

480
00:17:50,400 --> 00:17:54,240
virtual machines

481
00:17:52,500 --> 00:17:55,860
as I said this was mostly manipulating

482
00:17:54,240 --> 00:17:57,240
dependencies so let's look at that in a

483
00:17:55,860 --> 00:18:00,720
bit more detail

484
00:17:57,240 --> 00:18:03,660
here's the pipei page for message spec

485
00:18:00,720 --> 00:18:05,400
on the left there's a download files tab

486
00:18:03,660 --> 00:18:06,960
when you click on that you see a long

487
00:18:05,400 --> 00:18:08,580
list of files it starts with Source

488
00:18:06,960 --> 00:18:12,600
distribution and then there's a long

489
00:18:08,580 --> 00:18:15,000
list of Wheels dot WHL files

490
00:18:12,600 --> 00:18:17,640
when you run pip install these are the

491
00:18:15,000 --> 00:18:19,500
files that pip can choose from so when I

492
00:18:17,640 --> 00:18:21,419
ran a pip install on my Mac it would

493
00:18:19,500 --> 00:18:24,419
actually choose this wheel where it says

494
00:18:21,419 --> 00:18:26,460
something about Mac OS arm 64.

495
00:18:24,419 --> 00:18:27,720
this is

496
00:18:26,460 --> 00:18:29,760
this whale includes the

497
00:18:27,720 --> 00:18:31,740
platform-specific code that was causing

498
00:18:29,760 --> 00:18:33,179
problems at the very start so this is

499
00:18:31,740 --> 00:18:35,100
the this is where we're getting that

500
00:18:33,179 --> 00:18:37,799
Darwin file from

501
00:18:35,100 --> 00:18:39,600
the fix and what we need to be doing is

502
00:18:37,799 --> 00:18:42,000
actually choosing this wheel at the top

503
00:18:39,600 --> 00:18:45,120
so Lambda is running on Linux and

504
00:18:42,000 --> 00:18:47,940
running the x86 64 CPU

505
00:18:45,120 --> 00:18:49,679
so we need to choose the appropriate

506
00:18:47,940 --> 00:18:52,620
wheel the one that has the platform

507
00:18:49,679 --> 00:18:54,360
specific code for that platform

508
00:18:52,620 --> 00:18:56,820
as far as I know the only way to pick

509
00:18:54,360 --> 00:18:58,260
convince pip to choose a particular

510
00:18:56,820 --> 00:18:59,760
wheel is to be running on the matching

511
00:18:58,260 --> 00:19:01,080
platform that's why we had to have a

512
00:18:59,760 --> 00:19:02,880
virtual machine so that we can pretend

513
00:19:01,080 --> 00:19:07,260
to be Linux that's why we had to have

514
00:19:02,880 --> 00:19:09,960
the x8664 CPU emulation so that we can

515
00:19:07,260 --> 00:19:13,200
pretend to be matching that CPU

516
00:19:09,960 --> 00:19:15,539
the magic of pants via the related PEX

517
00:19:13,200 --> 00:19:16,799
project is that it understands the

518
00:19:15,539 --> 00:19:18,539
difference between the computer we're

519
00:19:16,799 --> 00:19:20,100
currently running on and the target

520
00:19:18,539 --> 00:19:22,679
environment

521
00:19:20,100 --> 00:19:25,559
so we're targeting Lambda that's Linux

522
00:19:22,679 --> 00:19:28,080
that's x8664.

523
00:19:25,559 --> 00:19:29,760
we can choose the appropriate zip and

524
00:19:28,080 --> 00:19:31,440
mash that up into the appropriate format

525
00:19:29,760 --> 00:19:34,140
even if we're not running on that

526
00:19:31,440 --> 00:19:36,299
platform so we can still run on Mac but

527
00:19:34,140 --> 00:19:38,100
we choose this ZIP and then

528
00:19:36,299 --> 00:19:39,660
turn it into the appropriate zip file

529
00:19:38,100 --> 00:19:41,520
for AWS

530
00:19:39,660 --> 00:19:43,200
this is why pants is able to run those

531
00:19:41,520 --> 00:19:44,400
builds so much faster because there's

532
00:19:43,200 --> 00:19:48,380
none of that overhead of virtual

533
00:19:44,400 --> 00:19:48,380
machines or CPU emulation

534
00:19:49,320 --> 00:19:53,940
sounds good how do we adopt pants let's

535
00:19:51,780 --> 00:19:55,440
go back to the the simplest version of

536
00:19:53,940 --> 00:19:57,419
the two lambdas that we had where we had

537
00:19:55,440 --> 00:19:59,460
our Echo dot pi and we had our plot.pi

538
00:19:57,419 --> 00:20:02,160
with our source code and then a single

539
00:19:59,460 --> 00:20:04,260
requirements file that contains all the

540
00:20:02,160 --> 00:20:06,600
requirements required

541
00:20:04,260 --> 00:20:08,820
to start using it we first add a pants

542
00:20:06,600 --> 00:20:10,200
dot tummel which specifies some Global

543
00:20:08,820 --> 00:20:11,700
configuration like with Japan's version

544
00:20:10,200 --> 00:20:13,980
we're using

545
00:20:11,700 --> 00:20:16,140
and we also add a build file

546
00:20:13,980 --> 00:20:18,120
the build file basically tells pants

547
00:20:16,140 --> 00:20:20,160
like here's my source code here's what

548
00:20:18,120 --> 00:20:21,780
you should be paying attention to

549
00:20:20,160 --> 00:20:23,700
so we need to say two things we say

550
00:20:21,780 --> 00:20:26,400
python requirements which will pick up

551
00:20:23,700 --> 00:20:28,500
our requirements.txt file and say python

552
00:20:26,400 --> 00:20:30,299
sources which will pick up the two dot

553
00:20:28,500 --> 00:20:31,799
Pi files

554
00:20:30,299 --> 00:20:33,720
once we've done this we've now told

555
00:20:31,799 --> 00:20:36,240
pants about our python code and in

556
00:20:33,720 --> 00:20:37,620
theory if we had any tests we could run

557
00:20:36,240 --> 00:20:39,240
those tests

558
00:20:37,620 --> 00:20:41,280
but we don't care about tests what we

559
00:20:39,240 --> 00:20:43,320
care about is building Landers

560
00:20:41,280 --> 00:20:45,120
to do that we activate the built-in

561
00:20:43,320 --> 00:20:47,160
Lambda back end

562
00:20:45,120 --> 00:20:48,780
and then we Define the two lambdas we

563
00:20:47,160 --> 00:20:52,679
want build

564
00:20:48,780 --> 00:20:54,000
so these the configuration is a little

565
00:20:52,679 --> 00:20:56,340
bit overwhelming here but like it's

566
00:20:54,000 --> 00:20:58,080
simple enough we say it's python AWS

567
00:20:56,340 --> 00:21:00,240
Lambda function we need to configure two

568
00:20:58,080 --> 00:21:02,220
things we need to give it a name so Echo

569
00:21:00,240 --> 00:21:04,620
for the echo Lambda and we also need to

570
00:21:02,220 --> 00:21:05,880
tell parents where to find the Handler

571
00:21:04,620 --> 00:21:07,799
function

572
00:21:05,880 --> 00:21:10,620
so to handle the function is in Echo dot

573
00:21:07,799 --> 00:21:12,299
pi and it's called Handler

574
00:21:10,620 --> 00:21:15,179
we need to do the same configuration for

575
00:21:12,299 --> 00:21:18,299
plot so we say call a plot find the

576
00:21:15,179 --> 00:21:20,700
Handler function in plot.pi please

577
00:21:18,299 --> 00:21:21,900
this is enough to Define our lambdas so

578
00:21:20,700 --> 00:21:23,039
we haven't had to write any shell

579
00:21:21,900 --> 00:21:26,100
scripts we haven't had to fiddle around

580
00:21:23,039 --> 00:21:27,600
with docker and this is indeed the Dozen

581
00:21:26,100 --> 00:21:29,760
lines of configuration that I was using

582
00:21:27,600 --> 00:21:31,679
to run those performance benchmarks we

583
00:21:29,760 --> 00:21:34,340
mentioned we saw before we're driven off

584
00:21:31,679 --> 00:21:34,340
this configuration

585
00:21:34,740 --> 00:21:39,240
to actually do the build we run one

586
00:21:36,720 --> 00:21:40,740
command to package colon colon

587
00:21:39,240 --> 00:21:42,059
when we do this it takes about eight

588
00:21:40,740 --> 00:21:43,140
seconds there's some streaming output

589
00:21:42,059 --> 00:21:45,120
I've trimmed it down to what's

590
00:21:43,140 --> 00:21:46,799
interesting let's walk through it step

591
00:21:45,120 --> 00:21:48,600
by step

592
00:21:46,799 --> 00:21:50,580
first we can say we make the minimum Bar

593
00:21:48,600 --> 00:21:52,440
for a build system it's functional it's

594
00:21:50,580 --> 00:21:54,659
produced our two zip files so we've got

595
00:21:52,440 --> 00:21:57,000
echo.zip got plot.zip we can upload

596
00:21:54,659 --> 00:22:00,000
those those to AWS and

597
00:21:57,000 --> 00:22:01,760
we've got working lambdas

598
00:22:00,000 --> 00:22:03,659
foreign

599
00:22:01,760 --> 00:22:05,340
there's something that was missing from

600
00:22:03,659 --> 00:22:06,900
the configuration on the previous slide

601
00:22:05,340 --> 00:22:08,340
we didn't have to say anything about

602
00:22:06,900 --> 00:22:10,559
message spec we didn't have to say that

603
00:22:08,340 --> 00:22:13,100
Echo uses message spec we didn't have to

604
00:22:10,559 --> 00:22:15,720
say that the plot Lambda uses matplotlib

605
00:22:13,100 --> 00:22:18,120
what pants has done is it's inferred the

606
00:22:15,720 --> 00:22:20,820
dependencies so it actually looks at our

607
00:22:18,120 --> 00:22:24,059
echo.pi file and observes it as an

608
00:22:20,820 --> 00:22:25,559
import message spec lying in there it

609
00:22:24,059 --> 00:22:28,919
understands this is a reference to our

610
00:22:25,559 --> 00:22:32,220
requirements so it knows that that's it

611
00:22:28,919 --> 00:22:34,140
to run the echo dot Pi file you'll need

612
00:22:32,220 --> 00:22:36,240
to have message spec available and thus

613
00:22:34,140 --> 00:22:38,220
when we build the echo Lambda

614
00:22:36,240 --> 00:22:41,340
pants make sure to include the message

615
00:22:38,220 --> 00:22:42,960
spec library in that Lambda

616
00:22:41,340 --> 00:22:44,880
there's no other import statements in

617
00:22:42,960 --> 00:22:47,640
that file there's no input matplotlib so

618
00:22:44,880 --> 00:22:49,799
matplotlib doesn't need to be included

619
00:22:47,640 --> 00:22:51,360
buff dependency inference means that our

620
00:22:49,799 --> 00:22:53,640
Lambda is automatically including only

621
00:22:51,360 --> 00:22:54,960
the dependencies that it actually needs

622
00:22:53,640 --> 00:22:56,460
we're solving that problem of

623
00:22:54,960 --> 00:22:57,960
accidentally having larger builds than

624
00:22:56,460 --> 00:22:59,640
necessary

625
00:22:57,960 --> 00:23:01,620
the same process applies for the plot

626
00:22:59,640 --> 00:23:03,059
Lambda so pants has looked into the

627
00:23:01,620 --> 00:23:05,400
plot.pi file and observed that

628
00:23:03,059 --> 00:23:08,780
matplotlib is imported therefore it

629
00:23:05,400 --> 00:23:08,780
better be included in the Lambda

630
00:23:10,380 --> 00:23:14,220
let's talk about performance because

631
00:23:12,000 --> 00:23:16,559
performance is the the next bar or above

632
00:23:14,220 --> 00:23:18,659
functional for a build system

633
00:23:16,559 --> 00:23:21,240
the total build Tech took about eight

634
00:23:18,659 --> 00:23:23,520
seconds and eight seconds is less than

635
00:23:21,240 --> 00:23:25,440
it would take to First build echo which

636
00:23:23,520 --> 00:23:27,360
was two seconds and then build plot

637
00:23:25,440 --> 00:23:28,919
which was also about eight seconds

638
00:23:27,360 --> 00:23:30,240
what's going on here

639
00:23:28,919 --> 00:23:33,299
well instead of building them one by one

640
00:23:30,240 --> 00:23:34,919
it's not going Echo then plot

641
00:23:33,299 --> 00:23:36,780
pants is automatically building these in

642
00:23:34,919 --> 00:23:39,780
parallel so it's building Echo and pants

643
00:23:36,780 --> 00:23:40,740
at the same time Echo M plot at the same

644
00:23:39,780 --> 00:23:41,820
time

645
00:23:40,740 --> 00:23:44,039
um

646
00:23:41,820 --> 00:23:45,900
parallel is nice it doesn't make too

647
00:23:44,039 --> 00:23:50,100
much of a difference for these lambdas

648
00:23:45,900 --> 00:23:52,140
but um it uh

649
00:23:50,100 --> 00:23:53,640
but if you had more it would make a much

650
00:23:52,140 --> 00:23:55,320
bigger difference

651
00:23:53,640 --> 00:23:57,000
when we're talking about performance we

652
00:23:55,320 --> 00:23:58,559
better run things multiple times so if

653
00:23:57,000 --> 00:24:01,020
we run pants package again it gets so

654
00:23:58,559 --> 00:24:03,419
much faster what's going on there

655
00:24:01,020 --> 00:24:05,039
well pants has built-in caching

656
00:24:03,419 --> 00:24:07,020
the first time we ran the package

657
00:24:05,039 --> 00:24:09,659
command we had to run build so we

658
00:24:07,020 --> 00:24:12,000
started with echo.pi and plot.pi

659
00:24:09,659 --> 00:24:13,320
and we built our zip files that will

660
00:24:12,000 --> 00:24:14,940
take two seconds and eight seconds they

661
00:24:13,320 --> 00:24:17,220
run in parallel

662
00:24:14,940 --> 00:24:19,140
when we run it the second time we

663
00:24:17,220 --> 00:24:20,820
haven't changed code we haven't changed

664
00:24:19,140 --> 00:24:22,919
configuration

665
00:24:20,820 --> 00:24:25,140
so instead of having to build the things

666
00:24:22,919 --> 00:24:26,940
again pants understands the inputs are

667
00:24:25,140 --> 00:24:29,580
the same therefore the outputs will be

668
00:24:26,940 --> 00:24:31,320
the same this happens so much quicker

669
00:24:29,580 --> 00:24:32,940
it's just copying files around rather

670
00:24:31,320 --> 00:24:35,820
than having to build like actually do

671
00:24:32,940 --> 00:24:38,159
the expensive build process

672
00:24:35,820 --> 00:24:40,020
this happens in a fine-grained way so if

673
00:24:38,159 --> 00:24:42,120
we had changed echo.pi imagine we'd

674
00:24:40,020 --> 00:24:44,220
added a feature to our Echo Lambda but

675
00:24:42,120 --> 00:24:46,559
we've left plot as it was

676
00:24:44,220 --> 00:24:49,280
when we ran this when we would package

677
00:24:46,559 --> 00:24:49,280
this again

678
00:24:49,860 --> 00:24:53,100
the plot Lander would still be cached

679
00:24:51,480 --> 00:24:54,600
because we haven't changed any code that

680
00:24:53,100 --> 00:24:56,520
affects the plot map

681
00:24:54,600 --> 00:24:58,500
the echolanda has changed so we need to

682
00:24:56,520 --> 00:25:00,960
run that build but now the whole process

683
00:24:58,500 --> 00:25:03,600
takes two seconds and still gives us all

684
00:25:00,960 --> 00:25:05,400
of the zip files that we requested

685
00:25:03,600 --> 00:25:07,080
this caching applies to more than just

686
00:25:05,400 --> 00:25:09,059
building packages so we use it for

687
00:25:07,080 --> 00:25:12,360
testing we use it for linting and the

688
00:25:09,059 --> 00:25:14,039
caching applies to that as well

689
00:25:12,360 --> 00:25:16,440
caching is the first and Panthers

690
00:25:14,039 --> 00:25:18,360
performance Powers so caching means that

691
00:25:16,440 --> 00:25:19,559
pants never really does work that it's

692
00:25:18,360 --> 00:25:22,500
already done

693
00:25:19,559 --> 00:25:25,320
when you when Pants is running a process

694
00:25:22,500 --> 00:25:26,039
it's usually because it has to it's

695
00:25:25,320 --> 00:25:27,840
um

696
00:25:26,039 --> 00:25:29,279
it's new work

697
00:25:27,840 --> 00:25:31,500
the next performance power is

698
00:25:29,279 --> 00:25:33,360
parallelism so when work has to happen

699
00:25:31,500 --> 00:25:35,580
happens in parallel

700
00:25:33,360 --> 00:25:37,620
the third one surprisingly enough is

701
00:25:35,580 --> 00:25:39,059
dependency inference depends the

702
00:25:37,620 --> 00:25:41,039
inference is what ensured that we did

703
00:25:39,059 --> 00:25:42,960
the minimum amount of work possible our

704
00:25:41,039 --> 00:25:45,059
Echo Lambda only included the dependency

705
00:25:42,960 --> 00:25:46,740
it needed to include

706
00:25:45,059 --> 00:25:49,020
the final one which is the one we sold

707
00:25:46,740 --> 00:25:50,460
first is native execution we're not

708
00:25:49,020 --> 00:25:52,320
having to do virtual machines we're not

709
00:25:50,460 --> 00:25:55,020
having to emulate CPUs

710
00:25:52,320 --> 00:25:56,460
so the work that needs to run is run it

711
00:25:55,020 --> 00:25:57,720
runs in parallel

712
00:25:56,460 --> 00:25:59,520
it's the smallest amount of work

713
00:25:57,720 --> 00:26:02,940
possible and it runs as fast as possible

714
00:25:59,520 --> 00:26:05,159
that makes for a happy developer

715
00:26:02,940 --> 00:26:07,380
for our system this is the impact it had

716
00:26:05,159 --> 00:26:09,600
so I remember how I set our builds we're

717
00:26:07,380 --> 00:26:12,840
taking about three minutes before

718
00:26:09,600 --> 00:26:14,279
now they take about 90 seconds and this

719
00:26:12,840 --> 00:26:16,140
is actually happening already on Linux

720
00:26:14,279 --> 00:26:18,059
so this is all because of parallelism

721
00:26:16,140 --> 00:26:20,460
and it's all because of the dependency

722
00:26:18,059 --> 00:26:22,980
inference reducing how much work happens

723
00:26:20,460 --> 00:26:25,140
far more significantly though when we're

724
00:26:22,980 --> 00:26:26,820
testing our infrastructure locally

725
00:26:25,140 --> 00:26:28,500
previously it will take us five and a

726
00:26:26,820 --> 00:26:30,000
half minutes and that's mostly because

727
00:26:28,500 --> 00:26:32,159
we would be building and rebuilding

728
00:26:30,000 --> 00:26:33,360
lambdas pointlessly every time we run

729
00:26:32,159 --> 00:26:35,880
tests

730
00:26:33,360 --> 00:26:37,260
now Panthers caching kicks in so instead

731
00:26:35,880 --> 00:26:39,179
of having to rebuild the land that's

732
00:26:37,260 --> 00:26:41,400
already been built it just gets served

733
00:26:39,179 --> 00:26:43,020
out of the cache thus our tests take 10

734
00:26:41,400 --> 00:26:46,039
seconds instead

735
00:26:43,020 --> 00:26:46,039
that's much more enjoyable

736
00:26:46,679 --> 00:26:50,700
in summary this is about as much

737
00:26:48,480 --> 00:26:52,740
clothing as we could get on Sterling

738
00:26:50,700 --> 00:26:55,020
he's not going to wear pants he barely

739
00:26:52,740 --> 00:26:57,779
tolerates his harness

740
00:26:55,020 --> 00:26:59,940
but actually at exoflare we saw that

741
00:26:57,779 --> 00:27:01,860
when we adopted pants using it for

742
00:26:59,940 --> 00:27:03,659
building lambdas it dramatically reduced

743
00:27:01,860 --> 00:27:05,700
our build times and our test times for

744
00:27:03,659 --> 00:27:08,520
our infrastructure as well as keeping

745
00:27:05,700 --> 00:27:10,919
our configuration really simple

746
00:27:08,520 --> 00:27:12,480
EXO flare has a strong open source

747
00:27:10,919 --> 00:27:15,120
mindset and understands how important it

748
00:27:12,480 --> 00:27:17,159
is such as me contributing to

749
00:27:15,120 --> 00:27:18,480
um pants as well as paying for our

750
00:27:17,159 --> 00:27:21,059
dependencies

751
00:27:18,480 --> 00:27:23,279
and where please come and talk to me if

752
00:27:21,059 --> 00:27:25,380
you're interested in biosecurity if

753
00:27:23,279 --> 00:27:27,120
you're looking for a job

754
00:27:25,380 --> 00:27:28,320
um and just generally if you want to

755
00:27:27,120 --> 00:27:30,179
have a chat

756
00:27:28,320 --> 00:27:32,820
the code examples and some more links

757
00:27:30,179 --> 00:27:35,720
are available in this Repository

758
00:27:32,820 --> 00:27:35,720
thank you very much

759
00:27:40,080 --> 00:27:43,740
unfortunately we haven't got enough time

760
00:27:42,120 --> 00:27:45,480
for questions but I'm sure you can find

761
00:27:43,740 --> 00:27:46,919
you on out and about in the conference

762
00:27:45,480 --> 00:27:48,600
if you have any further questions on the

763
00:27:46,919 --> 00:27:50,159
topic yeah please chase me if you have

764
00:27:48,600 --> 00:27:51,380
questions uh thank you we get the

765
00:27:50,159 --> 00:27:53,760
speakers

766
00:27:51,380 --> 00:27:55,940
thank you thank you guys thank you very

767
00:27:53,760 --> 00:27:55,940
much