@jaredpalmer @github @georgebrock Will this also come with a fix for review comments getting lost in old revisions? Most people I know doing stacked diffs are editing commits in the stack rather than adding fixes on top.
English
David Bell
689 posts
Stacked Diffs on @GitHub will start rolling out to early design partners in an alpha next month. In the meantime, here's video of our progress so far:
(h/t for @georgebrock + team for their awesome work)
English
@TobiM @0xdabbad00 imo iam data makes more sense from a “things i can operate” definition of service.
English
@dastbe @0xdabbad00 Yes, sure. Just thinking which source for the service count is the most authorative…
English
It appears over 5% of AWS services are currently in a state of deprecation. 🪦
Notes: There are currently 373 services (as defined by end-point prefixes in the SDK) and 20 are currently in a state of deprecation in my list.
Scott Piper@0xdabbad00
AWS deprecating Lookout For Equipment, Lex V1, and DeepComposer. 🪦 #aws_breaking_changes github.com/SummitRoute/aw…
English
@TobiM @0xdabbad00 some services don’t have public endpoints, some endpoints service multiple services.
English
@0xdabbad00 Interesting, IAM has currently 421 services:
awsiamdata.com
Do you know why there’s a difference?
English
@mipsytipsy it’s also incredibly dumb (in a good way). text-based events that were batched and pushed to s3 on a minutely basis. this ends up being extremely cheap compared to other systems that lose a ton of precision.
English
It's a little-known fact that should be MUCH better known: that AWS has been using arbitrarily-wide structured logs internally to understand core services for AGES.
They don't use the term "canonical logs", so it's just about impossible to find them talking about it. But true.
Victor | priv/acc@vcastellm
@mipsytipsy I remember an AWS post from like 4 years ago about the exact same topic. For me it made total sense but there's a big stopper, we use DataD in my last 2 companies and metrics are super cheap but logs are super expensive so this approach is more difficult to adopt. Still the case
English
@__steele what a shame, though it wasn’t clear what its market was. i wish the internal journaldb (originally named qldb) were made available instead.
English
Amazon QLDB will shut down. All data will be deleted next year. They’re recommending a migration to Aurora Postgres.
English
@mikejulian any insights you can reveal on what’s keeping costs down?
English
There's one company we work with that's been at this for eight years through hypergrowth and we're all still looking at their bill and thinking, "that's it?!"
Bodes well for the future, I think.
English
Reminds me of something that surprised me:
most $50m/yr+ run-rate bills look like cloud circa 2012: mostly EC2 instaces, S3, EBS, RDS
aka they look like a datacenter
dax@thdxr
when you’re a small child you worry about serverless costs eventually you grow up and worry about inter-az egress costs
English
Moving in with my now wife at 25, double income & a dog, and living in this kind of density for the next 5 years, absolutely DESTROYED my bar for what is possible from a convenient urban environment.
Having to drive 4 minutes to a grocery store now feels like torture.
English
@nathankpeck @thdxr biggest difference is that for cdk the cdk binary and cloudformation are the actual runtime, so in your “cdk deploy” step extending the runtime is extremely difficult compared to pulumi.
English
the fundamental flaw with cdk is that it promises you a programming model but ultimately it’s just a yaml generator
that means the code you write in the order you write it has nothing to do with how it ultimately executes
every pain point in cdk can be traced back to this
English
@nickcoult > If you need the data in the EBS volume to be retained after the task exits, check Delete on termination.
this statement sounds inverted?
English
Big launch for the ECS team! We have been working on this for a while and are very happy to bring it to customers. aws.amazon.com/blogs/aws/amaz…
English
@kohidave I only ever use one partition key, it's named "DAVID-IS-THE-BEST/"
English
@Austen @pbreit @coffeelance the implication is that companies are using risk of long-term high inflation as a smoke screen to ramp prices way beyond cost increases from inflation. Now that inflation rate is going down, they have no excuse to keep these prices high.
English
@pbreit @coffeelance Inflation “going down” from 9% to 3% means prices are going up more slowly, not that they are dropping
English
Absolutely wild that a post like this goes through many formal layers of approval without getting shot down and goes out to millions of people
English
@secengjeff @rakyll thats a mean thing to say about aurora limitless database (in preview)
English
How did we go from continuously launching to launching once a year at the big conference?
English
English
Maybe this is convenient for non-critical stacks but anything business critical still needs a multi-stage pipeline.
They have support for staging and prod branches. Maybe that's enough for some?
aws.amazon.com/about-aws/what…
English
Reading @GergelyOrosz' The Software Engineer Guidebook on route to Vegas. Loving it so far. Just went through his insights on platform teams, and it's spot on. Feels like a literal description of my team.
English
@QuinnyPig if they put an apple and a zucchini in and put an arrow between them, it would actually be perfect
English
Someone in Amazon’s marketing department needs to fix the Whole Foods logo; it needs a big fukken smirk underneath it to bring it into brand compliance.
English
@BayesianNuance @amimevalepitito @notanastronomer i got a next day passport last week. you need to show that you have travel within a week. maybe spent 1 hour over the course of two days on it.
English
@amimevalepitito @notanastronomer Yes it's been a while, but I'm familiar with some variety of options on that front. I just didn't recall having one with that rapid of a turnaround. My memory is fuzzy though, and a lot of passport experiences might be blending together.
English
As an addendum, it's almost always better to use empty structs and fields rather than enums, since that empty struct can be easily extended with override configuration in the futre.
English
Instead, just create three structs for http, tcp, and grpc and put the configurable fields under each one
* end users know intuitively what fields they can set for a given protocol
* validation is just verifying only one protocol is specified, then validating that protocol
English
second: prefer a "topic-comment" style API with repetition than flat API where it's up to the service what fields can be configured together.
A good example is an API for defining http, tcp, or grpc ports
Randall Hunt@ranman
What are some random API design choices that you like? What are some things people don't often think about when making APIs? Yes, I am building an API. No, I can't be more specific at this time.
English