July 26, 2025 • 27 mins
Jon's gone fishing, Shay's gone camping, so the community took the time to optimize all the things :)
https://cupogo.dev/ for all the links and details! This show is supported by you, and there are many ways to support the show (directly and indirectly). Thanks for listening!
- Meetup roundup
- Proposal moved to Active: http3
- Datadog's blogpost: How Go 1.24's Swiss Tables saved us hundreds of gigabytes
- Turso's blogpost: We rewrote large parts of our API in Go using AI: we are now ready to handle one billion databases
- Daniel Harp's ^ throw operator
- Ahmet Alp Balkan's blogpost: Kubernetes List API performance and reliability
Mark as Played
Transcript
Episode Transcript
Available transcripts are automatically generated. Complete accuracy is not guaranteed.
Shay Nehmad (00:00):
Welcome to Cup o Go. This show is supported by
(00:03):
you. If you wanna learn how tosupport, find our Slack channel,
find our email, our swag store,or anything else, please visit
us at cup0go.dev. That iscup0go.dev. If you're a regular
listener, there's no ad breakanymore, at least for this
episode.
Let's kick it off. This is Cupof Go for 07/26/2025. Keep up to
(00:33):
date with important happeningsin the Go community in about
fifteen minutes per week. I'mShay Nehmad
Jonathan Hall (00:38):
And I'm Jonathan Hall.
Shay Nehmad (00:39):
Hey, John. What's up?
Jonathan Hall (00:40):
Hey. Caught a fish, kind of. I I did. My son
kind of caught a fish.
Shay Nehmad (00:46):
That's that's a good start. Email or or
physical? Yeah.
Jonathan Hall (00:51):
We went we went fishing, the the old kind of
fishing with, you know, a rodand a reel. And it was a
terrible day for fishing. Thefish were not biting. But our
neighbor on the dock caught one,headed on the line, let my son
reel it in. So my son got to sayhe caught a fish, little six
inch rainbow trout.
Shay Nehmad (01:05):
Awesome. That's a good reason to take a break. I
was camping out in LassenNational Park, and my daughter
played in snow for the firsttime in her life.
Jonathan Hall (01:14):
Nice.
Shay Nehmad (01:15):
And she was super excited and sinking like, let it
go, you know, let it go.
Jonathan Hall (01:21):
Yeah. She loved Of course, why not?
Shay Nehmad (01:23):
So we took a break. We took a week off. We
appreciate you. I don't know howyou managed to survive without
us, but there's a lot of newsnews to talk about. We wanted to
do a conference roundup, so kickus off, Jonathan.
Jonathan Hall (01:35):
Yes. We've got a bunch to talk about. First up,
in Manchester in The UK, JamieTanner, listener in front of the
show, is doing a hands ondependency workshop at Fatsuma.
That's on Wednesday, July 30.That's just next week.
So if you are in or want to bein Manchester, check that one
out. Link in the show notes.Next up, FINE Conf twenty twenty
five. We've talked about itbefore. We've talked about it
(01:57):
last year.
Andy Williams, also a goodfriend of the show, is
organizing this one for FINE,the GUI library and framework
for Go. They're still lookingfor some speakers. The
conference takes place September19. If you'd like to speak
though about FINE or somethingrelated, let Andy know. They're
just looking for speakers andthey'll help cover your cost of
(02:19):
travel if you need that.
So, yeah, if you'd like to be inEdinburgh, Scotland, either to
attend or to speak, reach out toAndy Williams. Link also in the
show notes.
Shay Nehmad (02:29):
So these are two conferences by friends of the
show, but there's also the Wiki.
Jonathan Hall (02:33):
There's also the Wiki enemies of the been going
on there. Theoretically.
Shay Nehmad (02:37):
Yes. Enemies of the show. Enemies of the show.
Jonathan Hall (02:42):
Rust conferences have no. No.
Shay Nehmad (02:44):
I don't think the Rust the Rust conferences are
enemies. I don't think so.Except like the Erlang, the
Clojure, the JavaScript, thePython. Don't know. Rust and Go
people are friends.
Jonathan Hall (02:58):
Alright. So anyway, the Go Wiki forever has
had a page about conferences. Iused to just check that every
week. It rarely gets updatedanymore, so this is not really a
Wiki, but it's been updatedrecently. So I'm gonna breeze
through these.
Next up, we talked about this onthe last episode. Go South
Africa is happening August. It'san online conference, so you
could attend even if you're notin South Africa. GopherCon UK is
(03:21):
happening August in London.GopherCon, just GopherCon.
I guess it's GopherCon US, butit's the original.
Shay Nehmad (03:28):
It's just like Prince. Yeah. It's happening in
New formerly known as GopherCon.
Jonathan Hall (03:34):
This one's happening in New York City.
That's in The United States, forthose of you who don't know
where New York City is. That'sAugust. August is a busy month
for Go conferences. GovrConIndia will be happening
September.
Is that Jaipur or how do youpronounce that that city? Do you
know?
Shay Nehmad (03:53):
I actually have someone who speaks Indian right
next to me. So let me ask onesecond. Okay. Parin, how do you
pronounce Jaipur?
Jonathan Hall (04:00):
Jaipur.
Shay Nehmad (04:01):
Jaipur. Jaipur. Okay. Thank you, Parin. You are
you are now on the forever onthe episode of Cup of Go for
this poor guy just sitting nextto me in the office.
Jonathan Hall (04:13):
Go for Con India will be coming up September 1415
in Jaipur, India. Go Lab will behappening in Italy on October '7
in Florence, Italy. Go West, wetalked about this a couple
episodes ago. We had theorganizers on the show. This is
happening.
Shay Nehmad (04:29):
So what is the Go West conference? Yeah. So, we
wanted to create a conferencethat's more accessible to, a lot
of people, because, a lot ofpeople can't really afford to go
spend, you know, $5.06, $700 toattend a conference. Right?
Jonathan Hall (04:45):
This is happening October 24 in Lehi, Utah. That's
also The United States. And thelast one for now, Go For Gun
Africa will be happening. Not tobe confused with Go South
Africa. This is Go For ConAfrica.
They're two differentconferences, August 2425 in
Lagos, Nigeria. I think I'mdone.
Shay Nehmad (05:05):
One thing I'm excited about in Go For Con
India is they are hosting otherthan the I think the the call
for papers is still open and Ithink they're also still looking
for sponsors. But the it's afirst that I see. They have a
language roast.
Jonathan Hall (05:21):
Uh-huh.
Shay Nehmad (05:21):
So they're gonna do a debate on like which
programming languages
Jonathan Hall (05:26):
is the
Shay Nehmad (05:26):
best or whatever. Just like lighthearted roasting
and and laughs.
Jonathan Hall (05:31):
That sounds fun.
Shay Nehmad (05:32):
I mean, likes doing it anyway. So
Jonathan Hall (05:35):
Yeah. Of course. Why not why not make it part of
the conference and make itofficial?
Shay Nehmad (05:39):
Yeah. That's actually a cool idea. Yeah.
August is busy. I would I'm,like, surprised because it's,
like, exactly the weeks wherekids don't have any, any more
camps or whatever, the lastweeks of August.
We highly recommend going tothese meetups and conferences if
you can, if you have the time.It's a great way to network and
meet all these people,especially if you submit a talk
(06:01):
and you actually show it. It'sgood for a lot of reasons.
Recently, we found out that oneof them is that you can talk
about the Cup O'Go podcast inthe meetup or the conference,
and then more people listen tothe show. Yeah.
But it's good. There's a long,like, archive here. I love how
there's, like if you want toknow, like, when the Go
conference in Tokyo in 2013 was,you can find the link. I don't
(06:23):
know. Maybe it's useful.
Jonathan Hall (06:24):
Alright.
Shay Nehmad (06:25):
But there are still a lot what I'm saying is there
are still a lot of call forpapers open. So if you have any
cool things to share or you needsome motivation to write the
thing, you could definitely usethat.
Jonathan Hall (06:35):
We will probably mention some of these again as
they become closer, but don'tcount on it. We'll we'll try to
remind you of the important onesso you don't forget, but we
don't wanna just be talkingabout conferences. Let's talk
about some actual Go news.
Shay Nehmad (06:48):
Just one before, if you do like our friends of the
show, Jamie and, Andy, if youhave a meetup coming up or
you're arranging a new group orwhatever, feel free to just
contact us at the Slack groupand we'll, like, we'll promote
it for sure. Yeah. No problem.
Jonathan Hall (07:04):
That's how two of the ones we mentioned today came
up from people just telling usfrom the
Shay Nehmad (07:08):
And the conference I set up in San Francisco was
basically it happened basicallybecause of the podcast.
Jonathan Hall (07:15):
So you
Shay Nehmad (07:15):
know what I mean? We're all for it. Right.
Jonathan Hall (07:18):
So I've been hearing a lot about this Web
three thing. Is that old newsnow? Maybe it is because I think
we're talking about AI thesedays, but AI is the cool news.
That's what this is. Right?
HTTP three. It's Web three. It'sWeb four. I'm so confused. What
is this proposal?
Shay Nehmad (07:34):
It's web forehead. It's where I place my hand
whenever someone no, no, it'snot web three. It's HTTP three.
Oh. That's what this proposal isabout.
So, Damien Neal, I think they'rea member of the Google team, at
least their email is at Googledot com, has a proposal to add
HTTP three client and serverimplementations. And it's sort
(07:59):
of like this progressive,implementation of a thing in the
standard library where there's aproposal to start with x slash
NAT slash internal slash HTTPthree. And then, like, once it's
mature enough to walk on itsown, you know, you move it to h
u p three for to x slash NATslash h u p three. And, you
(08:19):
know, from x to the standardlibrary is a much taller order,
but I consider X, we talkedabout this a few times on the
show, like things that areinternal inside the X library,
they're pretty stable normallyfor normal usage. If you want to
use HTTPthree, you could do itonce this proposal is done.
So this proposal has been opensince, late twenty twenty four,
(08:43):
almost half a year now. And twodays ago, it's been added to the
active column of the proposalsproject. Since the beginning of
the year, there's been a lot of,like, change lists implementing,
like, this thing in the internalh t p three library, and I think
this means that it's, like, it'sit's maybe ready to get
(09:04):
promoted, or at least they wannaconsider promoting it and see
what they need to do to clean itup. What is HTTP three, though?
Like, why would we even want it?
Jonathan Hall (09:12):
Yeah. I was gonna ask, what is it and why would it
not be included? I mean, if if
Shay Nehmad (09:16):
So I wanna explain it in terms of, HTTP one and
two. So if you don't know HTTPat all, this might be a bit
confusing, but we can do likeintro to networking 101 in a
fifteen minute show. So the bigchange is that HTTP one and HTTP
two use TCP, but HTTP three isbuilt on top of the QUIC
(09:38):
protocol, uses UDP. Actually,funnily enough, the best way to
understand it is to actuallyread the RFC, which is super
rare, right? Usually when youread an RFC, you're like, what
the hell is going What are allthese details?
Oh my God. But the intro to theRFC, to the relevant RFC, is
nine one one four, explains itthe best. It's like HTTP
(10:00):
semantics, things like GET andPOST and resource stuff, right?
And the status code, like 200 isokay and four zero four is not
found. They're pretty good.
HTTP one was a problem becauseit's like white space delimited
text fields, right, for the HTTPmessages. So it's really hard
that while you can read it,like, to ton of complexity when
(10:23):
parsing, which we mention allthe time on the show. Right? Oh,
security Exactly.Vulnerabilities in Yeah.
And also it doesn't havemultiplexing. Multiplexing
meaning you can have, like,multiple connections on the same
connection sort of thing.Because it doesn't have the h t
p one doesn't have that, youneed to open multiple TCP
connections. And TCP, theprotocol, beneath the HTTP in
the in the layers, if you openmultiple connections, they don't
(10:46):
share parameters between them.So things like congestion
control and things like, youknow, just network efficiency,
they don't actually getutilized.
TCP assumes that, you know, theapp above it will properly use
that connection between theclient and the server, but you
end up opening 15 connections tothe same API endpoint because
(11:06):
you wanna render 15 things onthe page. This was HTTP one.
HTTP two has binary framing, anda multiplexing layer, but
because it's on TCP, on top ofTCP, the multiplexing ended up
being, taking a big hit inreality because if one of the
streams in the multiplex, in themultiplex connection gets like a
(11:30):
TCP packet loss or aretransmission, they all wait
because they're all in the sameTCP connection. So it's sort of
like, it's better because it'sbinary and that's way better.
And it should improve latency,but you end up like all these
fast cars are waiting in thesame light.
You know what I mean? The sameoff ramp. So it tends to have a
(11:52):
stall. And if you have tons ofactive transactions, like, let's
say that it's a web page, 50 APIcalls to the same server at the
same time to, I don't know,render maybe a very complicated
dashboard in an SPA orsomething, h u p two tends to
perform worse. Right?
And it's completely opposite towhat you would think. It's like,
oh, but it has multiplexing.Yeah. But actually, it's better
(12:14):
to open 50 network connections,instead of having all 50 of them
go through one. Since then,there's a new, transport
protocol, called QUIC, q u I c,the technical details of which
are absolutely fascinating, butI can't explain them.
I need, like, three hours,honestly. But they have use UDP
(12:36):
on top. So you don't have allthis, like, retransmission, blah
blah blah. Just send it becauseyou prove prove like, provide
the reliability at the streamlevel and the congestion control
is across the entire connectionbecause UDP doesn't really have
the concept of connections.Right?
It improves the performance ofHTTP compared to like TCP map.
It's also more secure because ithas TLS 1.3 at the transport
(12:56):
layer. So it has a lot ofbenefits and practically it's
just faster. Like all of thesethings end up to it being more
efficient. You get like, youdon't have the setup latency of
a TCP because it like does TCPfast open.
Like always, when you talk aboutprotocols, the devil's in the
detail, right? There's a lot ofthings you need to take care of
(13:18):
and push promise and caching andgateways and go away and like, I
mean, there's so many detailshere and so many security
vulnerabilities just waiting toget discovered, you know? And
you have to, like frame typesand error codes and blah blah
blah. So there's a lot ofdetail. The proposal and all the
(13:38):
change lists attached to it arebasically let's implement it in
Go.
There is a package calledQuickGo, but it's not part of
the standard library. And thingslike HTTP implementations, think
it sort of intuitively makessense why it should be part of
the standard library. Right?That's where we're at right now.
There's a lot of change listsfor the internal,
(14:00):
implementation.
And if you actually open up thechange lists themselves and look
at the details, I mean, it's notsimple. You know what I mean?
It's not easy to to understandwhat's going on. And there's
like to dos in the code. Oh,defer, cancel, read.
I gotta parse it. I need todetect panics, like consider
tracking the never index statusof headers with the n bit set in
(14:20):
their queue pack encoding. Imean, it's a very, very detailed
sort of work to implement theseprotocols. If you're interested
in that sort of stuff and maybeyou wanna read this and read the
spec and make sure they did itcorrectly, like these changes
are open now and you can findthe latest ones in the proposal.
And I'm excited to see what willhappen now that it's moved to
the active column.
Jonathan Hall (14:41):
Yeah. Is is there any pushback on this proposal?
Like, there anybody saying, no,no, we shouldn't put this in Go
or not yet? Or is it more justhashing out the details about
how to do it?
Shay Nehmad (14:51):
So actually, it's just Neil like, it's sort of an
umbrella proposal where Neilsaid, we're gonna do this, and
then, Gopher bot is like, hereare all the change lists. Now,
there's a there's a open issuefor, like, implementing
HTTPthree in the standardlibrary. In 2019, it was a
(15:11):
draft. Oh, this issue was openedin 2019, right, by Johan
Branderhorst Satzkorn. Yeah.
Actually from Canada, so I don'tknow. But it's like, there there
is an implementation of a quickgo, like existing one in pure
go, and the discussion startedlike 2019, and then you scroll
(15:33):
down to the most recent thing,it's like 2024, and the quick
implementation is already not ininternal, so it's just very
progressively going forward. Theactually, there's no pushback.
The only thing was like, hey, in2023, someone's like, all the
major browsers already support ht p three, why don't we have
this already? So, you know, it'slike, I mean, there's been a lot
(15:56):
of discussion, especially whenit got started.
But the thing was like, youknow, we don't wanna put it in
the standard library before thespec dies down and everybody
implements it and all the smokeclears up. I think now we're at
that at that point. Like, backin 2020, it wasn't even
finalized by the IETF, soputting it then wouldn't be a
(16:17):
smart idea. You know, Gostandard library is not quick to
adopt all these newtechnologies. Now it's not a new
technology anymore.
Jonathan Hall (16:24):
It's not quick, but I'm ching. Ah, nice. So I
guess this will be in the nextversion of Ghost since all they
have to do is tell Claude thatthey want HTTP three and it'll
just spit it out. Right?
Shay Nehmad (16:37):
Don't even don't even wanna, you know? Actually,
the the scariest part is how howmuch of the spec has been
written with AI, right? The codeis the easy part, but if someone
was like, I don't I don't wannaworry too much about the details
of TLS and QUIC, Somehow, I feellike the people working on this
(16:57):
proposal take their jobsslightly more seriously than
that.
Jonathan Hall (16:59):
I hope so.
Shay Nehmad (17:01):
All building the Internet.
Jonathan Hall (17:03):
We've got a couple more items to talk about
a little bit quicker,
Shay Nehmad (17:06):
I think.
Jonathan Hall (17:07):
The next one is
Shay Nehmad (17:08):
Wake up.
Jonathan Hall (17:09):
Uh-huh. Okay. So the next one is from Datadog.
Datadog had a blog post, how Goone point two four Swiss tables
saved us hundreds of gigabytes.Now when I first read the
headline, I thought, oh, Iunderstand this story.
They do a Swiss tables thing,which we talked about in the
past briefly, the way they dothe new implementation for maps,
(17:31):
how hash mapping works in Go,saved a bunch of memory, that's
cool. So I started reading itand they start by saying, We had
some terrible memory regressionsin Go 124. Like what? This is
the opposite of what I wasexpecting to read. So they
upgraded from 123 to 124 andtheir memory usage got worse and
they started investigating,working with the Go team and
they did find the regressionsand solve that.
(17:51):
But in the meantime, once theysolve the regressions, they
noticed that some of theirservices were using much less
memory than before. And it turnsout that's because of these this
new Swiss tables implementation.So the article goes into some
pretty low level detail like bitmapping of memory registers and
stuff. So if you're into thatsort of thing to want to
(18:12):
understand the nitty gritty ofwhy, I'm not going to read that
to you. You have to go read ityourself.
But I thought just the TLDR wasinteresting. They went after a
regression, which was using muchmore memory process and ended up
saving a ton of memory as wellin some cases. That's kind of a
fun story if you like techsleuth stories.
Shay Nehmad (18:30):
It is, it looks detailed and super well written.
Like there's charts explainingthe old table and the new table
and the hashes and the addressesand the groups. Very, very cool.
So they upgraded and it made itworse, but then they fixed the
part where it got worse and gotbetter. So overall Yeah.
Jonathan Hall (18:48):
Overall, a net win.
Shay Nehmad (18:50):
Yeah. Net win for the Swiss tables. Right? Yes.
Yes.
One other thing they mentionedis that they use the Go
community a lot. Like, theycollaborate a lot with other
people in the community, whichis pretty cool from from,
Datadog. They've, reached out,and now they're sharing their
results as well. That's the wayto go when you're a big company
(19:11):
like Datadog to make sure thatthe ecosystem you use is
actually you know what I mean?It's like they they give back
sorta.
That's what I I
Jonathan Hall (19:21):
It's one
Shay Nehmad (19:21):
of the
Jonathan Hall (19:22):
nice things about open source is, everybody
benefits from from changes likethis.
Shay Nehmad (19:27):
And they have a few tricks as well, like setting the
go memory limit and looking atobviously, being Datadog,
they're like, oh my god. Runtimemetrics and and profiling were
critical to our investigation,which you can do with Datadog.
Baa, ba, ba. But the bottom lineis they reduce 70% in map memory
(19:48):
usage, basically withoutchanging the logic.
Jonathan Hall (19:52):
Yeah. Yeah. That's pretty amazing.
Shay Nehmad (19:55):
I also have a company who did, like, a
optimization blog post. Again,we're not gonna go super deep
into the detail, but Terso Tech,they do, like, I don't know if
you heard of it, but it's likeSQLite for every they spin up,
like, a tiny SQLite instance forevery AI agent in your thing.
(20:18):
Okay. And they reimplementedSQLite in Rust. Okay.
Yeah. Kinda kinda cool. Theyhave some some very techy things
going on. I haven't actuallyused it, but they had a headline
that, like, caught my eye. I waslike, We rewrote large parts of
API in Go using AI.
I was like, Oh, cool. Did theywrite like a language generator
(20:38):
or a tool or whatever? No, itturns out the developer is
called the Avinash. So it'sAvinash Intelligence. Okay.
They got the clickbait, but theylike had to fully rewrite their
API, just because of performanceissues. They have a thing where
they have a database per agentand then people started using a
(20:59):
ton more agents and more AI andwhatever, so they had a lot more
tiny databases to set up. Andthey used a trick we shared in
the show called interning.
Jonathan Hall (21:08):
If you
Shay Nehmad (21:09):
remember, we mentioned it with a pretty
recent thing. The blog post, Ithink was 2024, maybe it's not
that recent, almost a year ago,Jesus. But we actually took our
sweet time reporting that. Butthey used unique. Handle because
you know, it's the same usertalking to a different, DB.
(21:32):
So it it was it made sense tointern, like, the app, and that
saved them a ton of memory. Andthe second thing they did as a
SQLite company is, okay, wewill, like, do a cache and we
will store the, you know, thethe second layer of that cache,
we'll store it in SQLite. Andthe cache miss is, like, very
(21:53):
fast because they use like fastdisks instead of keeping
everything in memory becausethey want to scale to a billion
databases. So just a these areway I think way more normal and
useful tricks to deal with likememory issues in Go. Datadog
obviously have like real hugecompany problems.
(22:14):
Yeah, yeah. This is slightlymore basic. So if you're earlier
in your like scaling outjourney, maybe this blog post is
useful for you as well. Allabout managing memory. People
don't have enough memory.
Why don't they just downloadmore RAM?
Jonathan Hall (22:27):
I can't remember why I didn't do that last time.
Went fishing. Went fishing andforgot everything I knew.
Shay Nehmad (22:34):
Alright. No ad break. Let's jump to the
lightning round. Let's do it.Lightning round.
Jonathan Hall (22:45):
First up in the lightning round, you probably
remember a few weeks ago, wesaid we're never gonna talk
about error handling again inGo. I'm breaking my word. We
found something by Daniel Harp.He posted on LinkedIn a few
weeks ago, two weeks ago, thatthe Go team has given up on
error handling and that kind ofupset him a little bit. So he
(23:05):
decided to just write his ownversion of error handling.
I'm not sure what problem thatreally solves, but it's kind of
neat to see someone try this. Heimplemented a throw operator. So
anybody who follows Go on socialmedia probably already knows
this has happened, but the Goteam, specifically Robert
Grisemere, has published a postabout this error handling in Go.
(23:26):
TLDR, they've we've been talkingabout improving error handling
for years with dozens orhundreds, possibly thousands of
proposals directly or indirectlyrelated to it and cannot get
anywhere close to consensus. SoGo team has decided for the
foreseeable future, we will stoppursuing syntactic language
changes for error handling.
We'll also close all open andincoming proposals that concern
(23:50):
themselves primarily with thesyntax for error handling
without further investigation.It's not what you think. It's
not like try catch or whatever.It's a little bit different than
that. You don't even have towrite the word throw, use the
caret and basically use thecaret in place of the error
value in an assignment.
Whenever that gets assigned to,it immediately returns. So
(24:13):
there's a link in the show notesto to his code. It's working
code. It's a fork of Go. You caninstall it if you want to.
It's just kind of a cool thingto look at. I don't I don't
imagine anybody's going to usethis for production stuff since
it's an unofficial fork of Go,but it's kind of a neat thing to
look at. So that's that's mine.
Shay Nehmad (24:30):
The I I love the comments. This is, so LinkedIn
coded. People are just like, Ithink a language like Java will
be more suitable for you.There's nothing wrong with Go's
verbose error check. I Iactually think LinkedIn is a
SciOp parody site, and they'rejust waiting to let
Jonathan Hall (24:51):
us know.
Shay Nehmad (24:52):
I like it.
Jonathan Hall (24:54):
It's cool. It's it's
Shay Nehmad (24:56):
a cool, thought experiment.
Jonathan Hall (24:58):
Yes. Exactly.
Shay Nehmad (25:00):
My Lightning Round, bone to throw is it's the reason
it's Lightning Round and not amain thing, it's because not
actually about Go, it's aboutKubernetes. But I assume a lot
of our listeners are actuallylike sort of platform engineers,
DevOps kind of people justbecause a lot of the people
working Go tend to be that sortof person. So there's a pretty
(25:21):
well written blog post by AhmedAl Balkan, I hope I'm saying
that correctly, about theKubernetes list API performance
and reliability. They runhundreds of thousands of bare
metal servers over hundreds ofKubernetes clusters. Jesus.
And just the API call to listthe resources has some pretty
(25:42):
funky defaults with, like,pagination and watch cache,
which make it when you combinethem both, it makes it, like,
super slow. And then it's apretty cool blog post because it
lists, like, some upstreamimprovements in, like, some
various versions and lists like,points you towards the actual
GitHub discussions. And finally,it's, the recommendation is
(26:04):
upgrade your clusters, enableaudit logs, and, like, you know,
use these privileged RBACs,experiment with a GoGC
environment variable to tunegarbage collection and cube API
server processes. And they werelike, we found that setting it
to 200 is better than notsetting it at all. I mean, a
useful trick.
If you end up managing so manyKubernetes resources, look back
(26:28):
at your life, see what you didwrong, and choose another
profession. But as long as youhave to deal with it, I think
this blog post could be usefulfor you. Also, like, pretty well
written. Alright.
Jonathan Hall (26:38):
Well, I think that wraps it up. No, break
today. Enjoy your Following
Shay Nehmad (26:42):
your feedback.
Jonathan Hall (26:43):
Yeah. See you next time.
Shay Nehmad (26:45):
Program exited.
Welcome to Cup o Go. This show is supported by
(00:03):
you. If you wanna learn how tosupport, find our Slack channel,
find our email, our swag store,or anything else, please visit
us at cup0go.dev. That iscup0go.dev. If you're a regular
listener, there's no ad breakanymore, at least for this
episode.
Let's kick it off. This is Cupof Go for 07/26/2025. Keep up to
(00:33):
date with important happeningsin the Go community in about
fifteen minutes per week. I'mShay Nehmad
Jonathan Hall (00:38):
And I'm Jonathan Hall.
Shay Nehmad (00:39):
Hey, John. What's up?
Jonathan Hall (00:40):
Hey. Caught a fish, kind of. I I did. My son
kind of caught a fish.
Shay Nehmad (00:46):
That's that's a good start. Email or or
physical? Yeah.
Jonathan Hall (00:51):
We went we went fishing, the the old kind of
fishing with, you know, a rodand a reel. And it was a
terrible day for fishing. Thefish were not biting. But our
neighbor on the dock caught one,headed on the line, let my son
reel it in. So my son got to sayhe caught a fish, little six
inch rainbow trout.
Shay Nehmad (01:05):
Awesome. That's a good reason to take a break. I
was camping out in LassenNational Park, and my daughter
played in snow for the firsttime in her life.
Jonathan Hall (01:14):
Nice.
Shay Nehmad (01:15):
And she was super excited and sinking like, let it
go, you know, let it go.
Jonathan Hall (01:21):
Yeah. She loved Of course, why not?
Shay Nehmad (01:23):
So we took a break. We took a week off. We
appreciate you. I don't know howyou managed to survive without
us, but there's a lot of newsnews to talk about. We wanted to
do a conference roundup, so kickus off, Jonathan.
Jonathan Hall (01:35):
Yes. We've got a bunch to talk about. First up,
in Manchester in The UK, JamieTanner, listener in front of the
show, is doing a hands ondependency workshop at Fatsuma.
That's on Wednesday, July 30.That's just next week.
So if you are in or want to bein Manchester, check that one
out. Link in the show notes.Next up, FINE Conf twenty twenty
five. We've talked about itbefore. We've talked about it
(01:57):
last year.
Andy Williams, also a goodfriend of the show, is
organizing this one for FINE,the GUI library and framework
for Go. They're still lookingfor some speakers. The
conference takes place September19. If you'd like to speak
though about FINE or somethingrelated, let Andy know. They're
just looking for speakers andthey'll help cover your cost of
(02:19):
travel if you need that.
So, yeah, if you'd like to be inEdinburgh, Scotland, either to
attend or to speak, reach out toAndy Williams. Link also in the
show notes.
Shay Nehmad (02:29):
So these are two conferences by friends of the
show, but there's also the Wiki.
Jonathan Hall (02:33):
There's also the Wiki enemies of the been going
on there. Theoretically.
Shay Nehmad (02:37):
Yes. Enemies of the show. Enemies of the show.
Jonathan Hall (02:42):
Rust conferences have no. No.
Shay Nehmad (02:44):
I don't think the Rust the Rust conferences are
enemies. I don't think so.Except like the Erlang, the
Clojure, the JavaScript, thePython. Don't know. Rust and Go
people are friends.
Jonathan Hall (02:58):
Alright. So anyway, the Go Wiki forever has
had a page about conferences. Iused to just check that every
week. It rarely gets updatedanymore, so this is not really a
Wiki, but it's been updatedrecently. So I'm gonna breeze
through these.
Next up, we talked about this onthe last episode. Go South
Africa is happening August. It'san online conference, so you
could attend even if you're notin South Africa. GopherCon UK is
(03:21):
happening August in London.GopherCon, just GopherCon.
I guess it's GopherCon US, butit's the original.
Shay Nehmad (03:28):
It's just like Prince. Yeah. It's happening in
New formerly known as GopherCon.
Jonathan Hall (03:34):
This one's happening in New York City.
That's in The United States, forthose of you who don't know
where New York City is. That'sAugust. August is a busy month
for Go conferences. GovrConIndia will be happening
September.
Is that Jaipur or how do youpronounce that that city? Do you
know?
Shay Nehmad (03:53):
I actually have someone who speaks Indian right
next to me. So let me ask onesecond. Okay. Parin, how do you
pronounce Jaipur?
Jonathan Hall (04:00):
Jaipur.
Shay Nehmad (04:01):
Jaipur. Jaipur. Okay. Thank you, Parin. You are
you are now on the forever onthe episode of Cup of Go for
this poor guy just sitting nextto me in the office.
Jonathan Hall (04:13):
Go for Con India will be coming up September 1415
in Jaipur, India. Go Lab will behappening in Italy on October '7
in Florence, Italy. Go West, wetalked about this a couple
episodes ago. We had theorganizers on the show. This is
happening.
Shay Nehmad (04:29):
So what is the Go West conference? Yeah. So, we
wanted to create a conferencethat's more accessible to, a lot
of people, because, a lot ofpeople can't really afford to go
spend, you know, $5.06, $700 toattend a conference. Right?
Jonathan Hall (04:45):
This is happening October 24 in Lehi, Utah. That's
also The United States. And thelast one for now, Go For Gun
Africa will be happening. Not tobe confused with Go South
Africa. This is Go For ConAfrica.
They're two differentconferences, August 2425 in
Lagos, Nigeria. I think I'mdone.
Shay Nehmad (05:05):
One thing I'm excited about in Go For Con
India is they are hosting otherthan the I think the the call
for papers is still open and Ithink they're also still looking
for sponsors. But the it's afirst that I see. They have a
language roast.
Jonathan Hall (05:21):
Uh-huh.
Shay Nehmad (05:21):
So they're gonna do a debate on like which
programming languages
Jonathan Hall (05:26):
is the
Shay Nehmad (05:26):
best or whatever. Just like lighthearted roasting
and and laughs.
Jonathan Hall (05:31):
That sounds fun.
Shay Nehmad (05:32):
I mean, likes doing it anyway. So
Jonathan Hall (05:35):
Yeah. Of course. Why not why not make it part of
the conference and make itofficial?
Shay Nehmad (05:39):
Yeah. That's actually a cool idea. Yeah.
August is busy. I would I'm,like, surprised because it's,
like, exactly the weeks wherekids don't have any, any more
camps or whatever, the lastweeks of August.
We highly recommend going tothese meetups and conferences if
you can, if you have the time.It's a great way to network and
meet all these people,especially if you submit a talk
(06:01):
and you actually show it. It'sgood for a lot of reasons.
Recently, we found out that oneof them is that you can talk
about the Cup O'Go podcast inthe meetup or the conference,
and then more people listen tothe show. Yeah.
But it's good. There's a long,like, archive here. I love how
there's, like if you want toknow, like, when the Go
conference in Tokyo in 2013 was,you can find the link. I don't
(06:23):
know. Maybe it's useful.
Jonathan Hall (06:24):
Alright.
Shay Nehmad (06:25):
But there are still a lot what I'm saying is there
are still a lot of call forpapers open. So if you have any
cool things to share or you needsome motivation to write the
thing, you could definitely usethat.
Jonathan Hall (06:35):
We will probably mention some of these again as
they become closer, but don'tcount on it. We'll we'll try to
remind you of the important onesso you don't forget, but we
don't wanna just be talkingabout conferences. Let's talk
about some actual Go news.
Shay Nehmad (06:48):
Just one before, if you do like our friends of the
show, Jamie and, Andy, if youhave a meetup coming up or
you're arranging a new group orwhatever, feel free to just
contact us at the Slack groupand we'll, like, we'll promote
it for sure. Yeah. No problem.
Jonathan Hall (07:04):
That's how two of the ones we mentioned today came
up from people just telling usfrom the
Shay Nehmad (07:08):
And the conference I set up in San Francisco was
basically it happened basicallybecause of the podcast.
Jonathan Hall (07:15):
So you
Shay Nehmad (07:15):
know what I mean? We're all for it. Right.
Jonathan Hall (07:18):
So I've been hearing a lot about this Web
three thing. Is that old newsnow? Maybe it is because I think
we're talking about AI thesedays, but AI is the cool news.
That's what this is. Right?
HTTP three. It's Web three. It'sWeb four. I'm so confused. What
is this proposal?
Shay Nehmad (07:34):
It's web forehead. It's where I place my hand
whenever someone no, no, it'snot web three. It's HTTP three.
Oh. That's what this proposal isabout.
So, Damien Neal, I think they'rea member of the Google team, at
least their email is at Googledot com, has a proposal to add
HTTP three client and serverimplementations. And it's sort
(07:59):
of like this progressive,implementation of a thing in the
standard library where there's aproposal to start with x slash
NAT slash internal slash HTTPthree. And then, like, once it's
mature enough to walk on itsown, you know, you move it to h
u p three for to x slash NATslash h u p three. And, you
(08:19):
know, from x to the standardlibrary is a much taller order,
but I consider X, we talkedabout this a few times on the
show, like things that areinternal inside the X library,
they're pretty stable normallyfor normal usage. If you want to
use HTTPthree, you could do itonce this proposal is done.
So this proposal has been opensince, late twenty twenty four,
(08:43):
almost half a year now. And twodays ago, it's been added to the
active column of the proposalsproject. Since the beginning of
the year, there's been a lot of,like, change lists implementing,
like, this thing in the internalh t p three library, and I think
this means that it's, like, it'sit's maybe ready to get
(09:04):
promoted, or at least they wannaconsider promoting it and see
what they need to do to clean itup. What is HTTP three, though?
Like, why would we even want it?
Jonathan Hall (09:12):
Yeah. I was gonna ask, what is it and why would it
not be included? I mean, if if
Shay Nehmad (09:16):
So I wanna explain it in terms of, HTTP one and
two. So if you don't know HTTPat all, this might be a bit
confusing, but we can do likeintro to networking 101 in a
fifteen minute show. So the bigchange is that HTTP one and HTTP
two use TCP, but HTTP three isbuilt on top of the QUIC
(09:38):
protocol, uses UDP. Actually,funnily enough, the best way to
understand it is to actuallyread the RFC, which is super
rare, right? Usually when youread an RFC, you're like, what
the hell is going What are allthese details?
Oh my God. But the intro to theRFC, to the relevant RFC, is
nine one one four, explains itthe best. It's like HTTP
(10:00):
semantics, things like GET andPOST and resource stuff, right?
And the status code, like 200 isokay and four zero four is not
found. They're pretty good.
HTTP one was a problem becauseit's like white space delimited
text fields, right, for the HTTPmessages. So it's really hard
that while you can read it,like, to ton of complexity when
(10:23):
parsing, which we mention allthe time on the show. Right? Oh,
security Exactly.Vulnerabilities in Yeah.
And also it doesn't havemultiplexing. Multiplexing
meaning you can have, like,multiple connections on the same
connection sort of thing.Because it doesn't have the h t
p one doesn't have that, youneed to open multiple TCP
connections. And TCP, theprotocol, beneath the HTTP in
the in the layers, if you openmultiple connections, they don't
(10:46):
share parameters between them.So things like congestion
control and things like, youknow, just network efficiency,
they don't actually getutilized.
TCP assumes that, you know, theapp above it will properly use
that connection between theclient and the server, but you
end up opening 15 connections tothe same API endpoint because
(11:06):
you wanna render 15 things onthe page. This was HTTP one.
HTTP two has binary framing, anda multiplexing layer, but
because it's on TCP, on top ofTCP, the multiplexing ended up
being, taking a big hit inreality because if one of the
streams in the multiplex, in themultiplex connection gets like a
(11:30):
TCP packet loss or aretransmission, they all wait
because they're all in the sameTCP connection. So it's sort of
like, it's better because it'sbinary and that's way better.
And it should improve latency,but you end up like all these
fast cars are waiting in thesame light.
You know what I mean? The sameoff ramp. So it tends to have a
(11:52):
stall. And if you have tons ofactive transactions, like, let's
say that it's a web page, 50 APIcalls to the same server at the
same time to, I don't know,render maybe a very complicated
dashboard in an SPA orsomething, h u p two tends to
perform worse. Right?
And it's completely opposite towhat you would think. It's like,
oh, but it has multiplexing.Yeah. But actually, it's better
(12:14):
to open 50 network connections,instead of having all 50 of them
go through one. Since then,there's a new, transport
protocol, called QUIC, q u I c,the technical details of which
are absolutely fascinating, butI can't explain them.
I need, like, three hours,honestly. But they have use UDP
(12:36):
on top. So you don't have allthis, like, retransmission, blah
blah blah. Just send it becauseyou prove prove like, provide
the reliability at the streamlevel and the congestion control
is across the entire connectionbecause UDP doesn't really have
the concept of connections.Right?
It improves the performance ofHTTP compared to like TCP map.
It's also more secure because ithas TLS 1.3 at the transport
(12:56):
layer. So it has a lot ofbenefits and practically it's
just faster. Like all of thesethings end up to it being more
efficient. You get like, youdon't have the setup latency of
a TCP because it like does TCPfast open.
Like always, when you talk aboutprotocols, the devil's in the
detail, right? There's a lot ofthings you need to take care of
(13:18):
and push promise and caching andgateways and go away and like, I
mean, there's so many detailshere and so many security
vulnerabilities just waiting toget discovered, you know? And
you have to, like frame typesand error codes and blah blah
blah. So there's a lot ofdetail. The proposal and all the
(13:38):
change lists attached to it arebasically let's implement it in
Go.
There is a package calledQuickGo, but it's not part of
the standard library. And thingslike HTTP implementations, think
it sort of intuitively makessense why it should be part of
the standard library. Right?That's where we're at right now.
There's a lot of change listsfor the internal,
(14:00):
implementation.
And if you actually open up thechange lists themselves and look
at the details, I mean, it's notsimple. You know what I mean?
It's not easy to to understandwhat's going on. And there's
like to dos in the code. Oh,defer, cancel, read.
I gotta parse it. I need todetect panics, like consider
tracking the never index statusof headers with the n bit set in
(14:20):
their queue pack encoding. Imean, it's a very, very detailed
sort of work to implement theseprotocols. If you're interested
in that sort of stuff and maybeyou wanna read this and read the
spec and make sure they did itcorrectly, like these changes
are open now and you can findthe latest ones in the proposal.
And I'm excited to see what willhappen now that it's moved to
the active column.
Jonathan Hall (14:41):
Yeah. Is is there any pushback on this proposal?
Like, there anybody saying, no,no, we shouldn't put this in Go
or not yet? Or is it more justhashing out the details about
how to do it?
Shay Nehmad (14:51):
So actually, it's just Neil like, it's sort of an
umbrella proposal where Neilsaid, we're gonna do this, and
then, Gopher bot is like, hereare all the change lists. Now,
there's a there's a open issuefor, like, implementing
HTTPthree in the standardlibrary. In 2019, it was a
(15:11):
draft. Oh, this issue was openedin 2019, right, by Johan
Branderhorst Satzkorn. Yeah.
Actually from Canada, so I don'tknow. But it's like, there there
is an implementation of a quickgo, like existing one in pure
go, and the discussion startedlike 2019, and then you scroll
(15:33):
down to the most recent thing,it's like 2024, and the quick
implementation is already not ininternal, so it's just very
progressively going forward. Theactually, there's no pushback.
The only thing was like, hey, in2023, someone's like, all the
major browsers already support ht p three, why don't we have
this already? So, you know, it'slike, I mean, there's been a lot
(15:56):
of discussion, especially whenit got started.
But the thing was like, youknow, we don't wanna put it in
the standard library before thespec dies down and everybody
implements it and all the smokeclears up. I think now we're at
that at that point. Like, backin 2020, it wasn't even
finalized by the IETF, soputting it then wouldn't be a
(16:17):
smart idea. You know, Gostandard library is not quick to
adopt all these newtechnologies. Now it's not a new
technology anymore.
Jonathan Hall (16:24):
It's not quick, but I'm ching. Ah, nice. So I
guess this will be in the nextversion of Ghost since all they
have to do is tell Claude thatthey want HTTP three and it'll
just spit it out. Right?
Shay Nehmad (16:37):
Don't even don't even wanna, you know? Actually,
the the scariest part is how howmuch of the spec has been
written with AI, right? The codeis the easy part, but if someone
was like, I don't I don't wannaworry too much about the details
of TLS and QUIC, Somehow, I feellike the people working on this
(16:57):
proposal take their jobsslightly more seriously than
that.
Jonathan Hall (16:59):
I hope so.
Shay Nehmad (17:01):
All building the Internet.
Jonathan Hall (17:03):
We've got a couple more items to talk about
a little bit quicker,
Shay Nehmad (17:06):
I think.
Jonathan Hall (17:07):
The next one is
Shay Nehmad (17:08):
Wake up.
Jonathan Hall (17:09):
Uh-huh. Okay. So the next one is from Datadog.
Datadog had a blog post, how Goone point two four Swiss tables
saved us hundreds of gigabytes.Now when I first read the
headline, I thought, oh, Iunderstand this story.
They do a Swiss tables thing,which we talked about in the
past briefly, the way they dothe new implementation for maps,
(17:31):
how hash mapping works in Go,saved a bunch of memory, that's
cool. So I started reading itand they start by saying, We had
some terrible memory regressionsin Go 124. Like what? This is
the opposite of what I wasexpecting to read. So they
upgraded from 123 to 124 andtheir memory usage got worse and
they started investigating,working with the Go team and
they did find the regressionsand solve that.
(17:51):
But in the meantime, once theysolve the regressions, they
noticed that some of theirservices were using much less
memory than before. And it turnsout that's because of these this
new Swiss tables implementation.So the article goes into some
pretty low level detail like bitmapping of memory registers and
stuff. So if you're into thatsort of thing to want to
(18:12):
understand the nitty gritty ofwhy, I'm not going to read that
to you. You have to go read ityourself.
But I thought just the TLDR wasinteresting. They went after a
regression, which was using muchmore memory process and ended up
saving a ton of memory as wellin some cases. That's kind of a
fun story if you like techsleuth stories.
Shay Nehmad (18:30):
It is, it looks detailed and super well written.
Like there's charts explainingthe old table and the new table
and the hashes and the addressesand the groups. Very, very cool.
So they upgraded and it made itworse, but then they fixed the
part where it got worse and gotbetter. So overall Yeah.
Jonathan Hall (18:48):
Overall, a net win.
Shay Nehmad (18:50):
Yeah. Net win for the Swiss tables. Right? Yes.
Yes.
One other thing they mentionedis that they use the Go
community a lot. Like, theycollaborate a lot with other
people in the community, whichis pretty cool from from,
Datadog. They've, reached out,and now they're sharing their
results as well. That's the wayto go when you're a big company
(19:11):
like Datadog to make sure thatthe ecosystem you use is
actually you know what I mean?It's like they they give back
sorta.
That's what I I
Jonathan Hall (19:21):
It's one
Shay Nehmad (19:21):
of the
Jonathan Hall (19:22):
nice things about open source is, everybody
benefits from from changes likethis.
Shay Nehmad (19:27):
And they have a few tricks as well, like setting the
go memory limit and looking atobviously, being Datadog,
they're like, oh my god. Runtimemetrics and and profiling were
critical to our investigation,which you can do with Datadog.
Baa, ba, ba. But the bottom lineis they reduce 70% in map memory
(19:48):
usage, basically withoutchanging the logic.
Jonathan Hall (19:52):
Yeah. Yeah. That's pretty amazing.
Shay Nehmad (19:55):
I also have a company who did, like, a
optimization blog post. Again,we're not gonna go super deep
into the detail, but Terso Tech,they do, like, I don't know if
you heard of it, but it's likeSQLite for every they spin up,
like, a tiny SQLite instance forevery AI agent in your thing.
(20:18):
Okay. And they reimplementedSQLite in Rust. Okay.
Yeah. Kinda kinda cool. Theyhave some some very techy things
going on. I haven't actuallyused it, but they had a headline
that, like, caught my eye. I waslike, We rewrote large parts of
API in Go using AI.
I was like, Oh, cool. Did theywrite like a language generator
(20:38):
or a tool or whatever? No, itturns out the developer is
called the Avinash. So it'sAvinash Intelligence. Okay.
They got the clickbait, but theylike had to fully rewrite their
API, just because of performanceissues. They have a thing where
they have a database per agentand then people started using a
(20:59):
ton more agents and more AI andwhatever, so they had a lot more
tiny databases to set up. Andthey used a trick we shared in
the show called interning.
Jonathan Hall (21:08):
If you
Shay Nehmad (21:09):
remember, we mentioned it with a pretty
recent thing. The blog post, Ithink was 2024, maybe it's not
that recent, almost a year ago,Jesus. But we actually took our
sweet time reporting that. Butthey used unique. Handle because
you know, it's the same usertalking to a different, DB.
(21:32):
So it it was it made sense tointern, like, the app, and that
saved them a ton of memory. Andthe second thing they did as a
SQLite company is, okay, wewill, like, do a cache and we
will store the, you know, thethe second layer of that cache,
we'll store it in SQLite. Andthe cache miss is, like, very
(21:53):
fast because they use like fastdisks instead of keeping
everything in memory becausethey want to scale to a billion
databases. So just a these areway I think way more normal and
useful tricks to deal with likememory issues in Go. Datadog
obviously have like real hugecompany problems.
(22:14):
Yeah, yeah. This is slightlymore basic. So if you're earlier
in your like scaling outjourney, maybe this blog post is
useful for you as well. Allabout managing memory. People
don't have enough memory.
Why don't they just downloadmore RAM?
Jonathan Hall (22:27):
I can't remember why I didn't do that last time.
Went fishing. Went fishing andforgot everything I knew.
Shay Nehmad (22:34):
Alright. No ad break. Let's jump to the
lightning round. Let's do it.Lightning round.
Jonathan Hall (22:45):
First up in the lightning round, you probably
remember a few weeks ago, wesaid we're never gonna talk
about error handling again inGo. I'm breaking my word. We
found something by Daniel Harp.He posted on LinkedIn a few
weeks ago, two weeks ago, thatthe Go team has given up on
error handling and that kind ofupset him a little bit. So he
(23:05):
decided to just write his ownversion of error handling.
I'm not sure what problem thatreally solves, but it's kind of
neat to see someone try this. Heimplemented a throw operator. So
anybody who follows Go on socialmedia probably already knows
this has happened, but the Goteam, specifically Robert
Grisemere, has published a postabout this error handling in Go.
(23:26):
TLDR, they've we've been talkingabout improving error handling
for years with dozens orhundreds, possibly thousands of
proposals directly or indirectlyrelated to it and cannot get
anywhere close to consensus. SoGo team has decided for the
foreseeable future, we will stoppursuing syntactic language
changes for error handling.
We'll also close all open andincoming proposals that concern
(23:50):
themselves primarily with thesyntax for error handling
without further investigation.It's not what you think. It's
not like try catch or whatever.It's a little bit different than
that. You don't even have towrite the word throw, use the
caret and basically use thecaret in place of the error
value in an assignment.
Whenever that gets assigned to,it immediately returns. So
(24:13):
there's a link in the show notesto to his code. It's working
code. It's a fork of Go. You caninstall it if you want to.
It's just kind of a cool thingto look at. I don't I don't
imagine anybody's going to usethis for production stuff since
it's an unofficial fork of Go,but it's kind of a neat thing to
look at. So that's that's mine.
Shay Nehmad (24:30):
The I I love the comments. This is, so LinkedIn
coded. People are just like, Ithink a language like Java will
be more suitable for you.There's nothing wrong with Go's
verbose error check. I Iactually think LinkedIn is a
SciOp parody site, and they'rejust waiting to let
Jonathan Hall (24:51):
us know.
Shay Nehmad (24:52):
I like it.
Jonathan Hall (24:54):
It's cool. It's it's
Shay Nehmad (24:56):
a cool, thought experiment.
Jonathan Hall (24:58):
Yes. Exactly.
Shay Nehmad (25:00):
My Lightning Round, bone to throw is it's the reason
it's Lightning Round and not amain thing, it's because not
actually about Go, it's aboutKubernetes. But I assume a lot
of our listeners are actuallylike sort of platform engineers,
DevOps kind of people justbecause a lot of the people
working Go tend to be that sortof person. So there's a pretty
(25:21):
well written blog post by AhmedAl Balkan, I hope I'm saying
that correctly, about theKubernetes list API performance
and reliability. They runhundreds of thousands of bare
metal servers over hundreds ofKubernetes clusters. Jesus.
And just the API call to listthe resources has some pretty
(25:42):
funky defaults with, like,pagination and watch cache,
which make it when you combinethem both, it makes it, like,
super slow. And then it's apretty cool blog post because it
lists, like, some upstreamimprovements in, like, some
various versions and lists like,points you towards the actual
GitHub discussions. And finally,it's, the recommendation is
(26:04):
upgrade your clusters, enableaudit logs, and, like, you know,
use these privileged RBACs,experiment with a GoGC
environment variable to tunegarbage collection and cube API
server processes. And they werelike, we found that setting it
to 200 is better than notsetting it at all. I mean, a
useful trick.
If you end up managing so manyKubernetes resources, look back
(26:28):
at your life, see what you didwrong, and choose another
profession. But as long as youhave to deal with it, I think
this blog post could be usefulfor you. Also, like, pretty well
written. Alright.
Jonathan Hall (26:38):
Well, I think that wraps it up. No, break
today. Enjoy your Following
Shay Nehmad (26:42):
your feedback.
Jonathan Hall (26:43):
Yeah. See you next time.
Shay Nehmad (26:45):
Program exited.
Popular Podcasts
Stuff You Should Know
If you've ever wanted to know about champagne, satanism, the Stonewall Uprising, chaos theory, LSD, El Nino, true crime and Rosa Parks, then look no further. Josh and Chuck have you covered.
The Joe Rogan Experience
The official podcast of comedian Joe Rogan.
Dateline NBC
Current and classic episodes, featuring compelling true-crime mysteries, powerful documentaries and in-depth investigations. Special Summer Offer: Exclusively on Apple Podcasts, try our Dateline Premium subscription completely free for one month! With Dateline Premium, you get every episode ad-free plus exclusive bonus content.