building polylane.com | prev @cloudflaredev, @baselimehq (acq), aerodynamicist, physicist
sf
- Tweets 14,836
- Following 2,386
- Followers 18,366
- Likes 59,706
it looks like aws is building a new JS runtime and it has over 10x faster cold starts and is up to 2x cheaper on lambda 🤔
49
128
1,552
443,642
anthropic just published a blog highlighting a pattern I've been using for weeks: a search tools tool
instead of loading all tools into the context window of the model, I load a "search tools" tool the model can use to browse a catalog of tools
just-in-time tool discovery
74
73
1,573
134,044
I don't really understand why we keep trying to "estimate" how much time building software would take
engineers know it's bollocks
tech leads know they are useless
PMs know it's finger the air estimations
yet so many teams still do this silly dance
133
114
934
172,472
software is not about writing code 🤦
Cost of writing a single line of code...
Software developer: $8.40
GitHub Copilot: $0.008
Assumptions:
Median dev salary = $105k
250 working days per yr
50 lines of code per day
Copilot = $100/yr
Realistically, a dev oversees Copilot & could ship more than 50 lines w/ 🤖 help.
33
59
780
185,659
maybe it's not up to you to decide wether changing to main changed something
"Changing master to main changed less than nothing. Because nothing was or is ever enough in this arena. As soon as this word battle was won, it was just on to the next and the next (and the next)." world.hey.com/dhh/forcing-ma…
24
32
756
176,995
I spoke with a team yesterday that is moving away from serverless in favor of managed k8s on aws
it wasn't an easy decision for them and they went through a thorough review of their architecture
the reasons they highlighted to me are:
46
95
651
the amazon prime video blog was right
lambda becomes prohibitively expensive really quickly
we've seen tremendous growth over the past 3 weeks, and it's not financially sound anymore to run on lambda
25
25
507
195,970
I wrote down the key lessons from my baseline journey
- be obsessed with finding pmf
- better, faster, cheaper is not enough anymore
- don't be stubborn
- do things that don't scale, really
- create hype, don't be shy
- get a side project
- exiting a company is hard
link below
15
43
494
80,284
10GB per durable object
opens up a whole new category of applications
Cloudflare delivers the toolkit for AI agents with new Agents SDK support for MCP (Model Context Protocol) clients, authentication/authorization/hibernation for MCP servers, and Durable Objects free tier. cfl.re/4i67tJ2 #DeveloperWeek
10
19
421
45,828
the question I get the most at every conference or meetup: "what even are durable objects?"
I spent the past 2 years building almost exclusively with durable objects, and today I'm compiling my notes and favourite patterns into a single blog post
boristane.com/blog/what-are-…
24
53
421
75,197
architecture diagrams before and after migrating the data ingestion pipeline of baselime from aws to cloudflare
I wrote a blog post about how why we migrated, how we did it, how we improved performace, and how much we saved
try to guess how much, you're not ready
17
46
383
54,614
we just shipped zero-intrumentation, otel-native tracing for @CloudflareDev workers
we worked directly in the runtime to add native instrumentation
- no code changes
- no bloated otel lib
- zero perf degradation
single command to opt in: wrangler deploy
it just works
👏Today is the day👏
Workers ✨automatic✨ tracing is now in open beta!
✅Enable in seconds – no code changes required
🔎View and query trace data directly in the Cloudflare dashboard
📦Export traces (and logs!) to any external destination with an OTel endpoint
22
20
334
61,316
5 months later, we shipped observability right in the cloudflare dashboard, starting with logs
all you got to do is add this to your wrangler file
[observability]
enabled = true
and we're just getting started - what do you want to see next?
40
25
324
72,074
I advocated for event-driven architectures for years
but I'm changing my mind about them
you don't need to glue together event bridge, sns, sqs, dynamodb streams, etc. to build a resilient system
I'm appreciating simplicity
22
15
301
66,757
truth developers don't want to accept:
- staging and production will never be identical
- production is unpredictable and all states cannot be captured by your testing suite
- problems that you really care about only happen in production
- testing in production works
27
42
293
39,365
wait, am I reading this right? and the comments agree?
I'm truly confused
so to delete an item you make a POST to /resource/id/delete ?
or do you make a POST to /resource/id with body { action: 'delete' }
what is this?
Unpopular opinion: Don't use http verbs PUT, PATCH, DELETE. Just use POST for everything. Reasons:
- <form> doesn't support the others. Frameworks that allow it do so through hacks.
- URLs are free, you don't gain anything by overloading them.
- Purity < practicality
126
8
287
158,841
my new favourite tool, I call it: "uncertainty tool"
every time the model is unable to respond to a user query, it reports it to me, and I can review
it's also the place where the model can self-reflect and ask clarifying questions to the user instead of making things up
23
7
304
17,663
I keep reading that durable objects are hard
let's change that narrative
here's how I'm building @basebrainhq entirely on @CloudflareDev
I create a new database for every user using durable objects
and everything is working with @DrizzleORM
boristane.com/blog/durable-o…
28
26
292
40,255
explain to me how a container that runs and incurs cost while waiting for requests is "serverless"
39
19
277
60,580
I realised I have a not so common way of doing event-driven architectures on AWS and I think it's worth sharing. opinions are welcome
I don't emit events directly from services, rather, I write them in a dedicated dynamodb table (one table per service)
25
33
271
sqs and aws lambda have a pretty peculiar integration, and more developers need to understand this integration to avoid scratching their heads over seemingly impossible bugs
4
53
258
2 days after I joined cloudflare, brendan casually tells me "oh yeah you can run containers on our network"
my jaw dropped
now it's your turn to be jaw dropped
blog.cloudflare.com/containe…
14
15
264
16,536
observability without action is just storage
too many people store terabytes of data and call it observability
10
29
249
24,744
agent memory works best when you combine vector search, full-text search, and graph traversal
but who wants to maintain a multi-tenant graph database?
not me
so I built a system to create unlimited on-demand graph databases with @cloudflaredev durable objects
19
28
259
36,813
with a whole freaking working example demo
github.com/cloudflare/orange
13
241
20,203
I'm hiring my first role at @cloudflaredev
if you want to come hack ai apps and tooling with @threepointone , @ritakozlov , @southpolesteve , @irvinebroque , @zebassembly , @elithrar , @dinasaur_404 and the rest of the team, here's THE opportunity
13
29
241
35,825
we upgraded from aws sdk v2 to v3 and our cold starts went from p90 of 640ms to ~1900ms 🤯
I added a trace picked at random for illustration
28
8
231
168,785
- build website
- buy domain
in that order.
I always fail at this; 3 domains bought this week alone.
30
11
224
sunday night project: sqlite databases running in durable objects with http interface, transactions, and migration support
create as many databases as you want with a single http request, and query them from anywhere
57ms latency
damn
11
5
226
19,672
turns out we're now backed by Google 🚀
we're joining the @GoogleStartupUK Black Founders Fund, it's a few exciting months ahead for the Baselime team!
30
15
181
11,020
cloudflare workflows and ctx.waitUntil have eliminated a whole category of software from my toolkit - made them completely obsolete
all the sqs + sns + eventbridge shenanigans, gone!
11
11
184
16,260
today I deleted the last remaining ec2 instances on Baselime
we can now say we're 100% serverless, including all our databases and our storage layers
17
4
171
31,821
here's a team at Amazon Prime getting over 90% cost savings by moving away from serverless and using a monolith on ECS
a lot can be said, but I want to know your opinion 🤔
26
19
173
49,715
it's cloudflare birthday week - lots of announcements coming
but ngl, y'all are not ready for what workers observability (formerly known as baselime) has been cooking
and we're just getting started
9
8
169
16,102
👋 talking to users/customers is all good but people will lie to you, a lot.
I've been doing this almost daily for the past 6 months and here's what I have learned
24
19
159
you thought we were done?
distributed tracing for @CloudflareDev with multi-cloud support
- automatic correlation between logs and traces
- support for workers, kv, durable objects, queues
- all powered by the @baselimehq query engine
and this design 😍✨
my dms are open if you want this
8
12
165
38,408
it's still settling in that you can call sqlite on cloudflare with no latency
like, you can make database operations like you do arithmetic in code
how is this legal?
14
6
165
78,474
you lot are not ready
distributed tracing for next.js applications with OpenTelemetry
- correlation between your logs and traces
- support for @trpcio, @prisma and all your favourite orms
- all the powers of the @baselimehq query engine
and those diagram and timeline views 💅
my dms are open if you want this
9
15
164
44,420
it's funny how in my mind $5 from my "side project" feels like much more than £500 from my day job
12
2
156
we now expose cpu time and wall time directly in workers logs
you can quickly filter requests where cpu time exceeds a given threshold for example
remember: on workers you're charged for cpu time, not wall time - if this isn't a bargain, I don't know what is
6
10
163
11,481
this is what a typical small serverless event-driven application looks like
a few lambdas, an API Gateway, some SNS, SQS and S3 action here and there
but unless you're intentional about it, understanding how exactly these components interact and the failure modes can be hard
2
28
151
20,633
we're building observability 2.0 directly within the workers platform
the stuff you usually only get with heavy instrumentation, and costing $$$$$
we're making it native to our cloud
7
10
160
12,214
using the "ABCD" framework when discussing product is significantly improving the quality of the feedback I get:
- what is Awesome ?
- what is Boring ?
- what is Confusing ?
- what Didn't you believe ?
thanks @MaryRobinette for this.
1
24
131
I just published a new blog post
contextual rag on @CloudflareDev using workers ai, vectorized, and d1
it explains everything you need from hybrid search to rerankers and reciprocal rank fusion, with clear code implementation
you can copy / paste this and build a saas around
10
22
154
11,492
if you're sick of your ECS and Fargate logs being a mess... well, I got you
we just shipped native support for ECS and Fargate services with no code changes or special config; just connect your aws account
- full-text search
- real-time error tracking
- live tail
- slice and dice your data
let me know if you want this; it's not even in the docs yet
13
14
145
43,067
yesterday my cloudflare kv calls p99 went from 50ms to 15ms
I didn't do anything, just underlying infra getting better 🤯
8
10
145
43,304
one week in the cloudflare office - and seriously I'm pumped!
the concentration of talent on the workers platform team is 🤯
5
2
137
10,838
I spent last week at @sequoia offices in SF for the closing week of the Arc programme; it was pretty surreal to be surrounded by legendary founders and builders
now let's make observability as code the new norm
11
6
142
at every serverless event, someone would ask about staying vendor agnostic
I don't explain anymore, please be vendor agnostic and let your competitors that leverage their cloud provider fully win when you're busy managing stuff
15
20
128
I'm building a gpt wrapper
I'm doing this because I can't remember anything
it's a llm-powered brain extension for tab hoarders like me - chat with everything you've saved
it's entirely built on @CloudflareDev workers and I'll share all the technical details
10
4
139
9,749
hono is my default framework
fully typesafe
so intuitive you never have to check the docs
out of the box openapi schemas and zod validation, for both request and response
middlewares that work the way you expect them to
just a pleasure to use
It's my first blog post for Cloudflare official blog. Much information about Hono!
blog.cloudflare.com/the-stor…
6
7
130
11,785
we put together an OpenTelemetry on Next.js starter kit and it's live on the @vercel templates marketplace
😎
2
8
131
22,174
if you're building a startup, this is your weekly reminder that there's no correlation between how complex what you build is and your chances of hitting pmf
9
5
130
7,825
today we launch our @vercel integration
we started building this 3 weeks ago and the response from our community has been extraordinary 🤯
- full-text logs search
- grouping logs by requests
- automatic error tracking
- dashboards
- page views and web vitals
- advanced queries
6
14
130
23,337
first step towards ramen profitability: reduce expenses
I just moved to a slightly cheaper apartment
12
122
"what do you mean you built the whole infrastructure? you don't have a platform team? and you do support too?"
10
6
117
11,779
👋 I'm working on something new
@baselimeHQ
An observability solution made for serverless architectures
12
13
113
this is the latency of our most invoked lambda function
dropped from 2000ms to 400ms
we traded 1600ms of wall time on lambda for 2.1ms of cpu time on workers
how? we moved some of the data processing and augmentation from lambda to workers
10
10
116
27,438
opentelemetry was supposed to be vendor agnostic
now on the otel website, every vendor has their own distro listed
and trust me, these libraries don't all work the same
15
10
113
15,256
@baselimehq + @CloudflareDev
imagine what we can build together!
take a second to think about it, we're about to give the cloud platform you always wanted: workers, kv, pipelines, durable objects, @partykit_io, etc., and best in class observability
one more thing 😉
congrats to the @baselimehq team and @boristane. very excited to being the next level of observability to the cloudflare workers platform 🎉🎉🎉
blog.cloudflare.com/cloudfla…
18
11
114
11,475
1 year ago we joined @CloudflareDev
6
2
116
6,821
I've been using exclusively cloudflare workers and the cloudflare dev platform for about a year now
in the same time I have stopped using the word "serverless"
I guess this is true serverless
6
8
112
10,217
when you hear someone say "we need to break up the monolith" what they mean is:
"we need boundaries in our codebase such that different teams can have ownership of different domains and each can be developed, tested, and deployed independently".
5
22
102
is it me or cloudflare is casually offering a serverless clickhouse db?
it's limited but it covers the vast majority of use cases
8
4
107
34,932
today we launch support for @CloudflareDev workers and pages on @baselimehq
from logs with logpush and our edge logger to distributed tracing with opentelemetry, we enable you to detect, diagnose and resolve issues in your cloudflare apps before they become problems
your favourite dev tool companies use @baselime in production with cloudflare workers: @partykit_io, @triggerdotdev, @onlineornot or @unkeydev
and it's just getting started
the reason they work with us is pretty simple, baselime is the easiest and fastest way to instrument your cloudflare workers and pages for both logging and distributed tracing with opentelemetry
and baselime surfaces everything to you: errors, exceptions, slow requests, etc...
you get the power to drill down into issues and questions but we bring answers to you instead of letting you fight with a query builder or whatever AI query generator
the best part is it's free up to 20 MILLION events at @baselimehq
7
15
105
24,363
I'm building something fun!
I need to process thousands of urls, grab their contents, including all the media, and store them whilst enabling fast retrieval
stack of choice is durable objects nested into cloudflare workflows and cf browser rendering
feels too easy to be true
11
1
106
8,831
I'm not sure if you realise what it means when workers charge for cpu time instead of wall time
add on that 0 cold starts
it's pretty wild
7
10
103
17,257
forgot to mention I'm speaking at re:invent 👀
Come learn from @kylegalbraith and @boristane how to overcome the challenges of observing serverless applications. For more information on Cloudflare’s activities check out this page: cfl.re/AWSreInvent2024 #AWSreInvent
6
3
103
6,732
I met an engineer today who told me he "only knows serverless"
he's self-taught, and has been working exclusively in serverless-first teams
there's an interesting future ahead with more serverless-first developers joining the industry
10
6
94
15,507
