Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

GCP Incidents

blog.railway.app

Post reply on HN
Print view View original

161–167 of 167 posts

Re: GCP Incidents

#161

Post by Bluecobra »

londons_explore wrote: post #151
betaby wrote: post #111

Earlier quoted context omitted.

Google doesn't do any 'black magic' even though they do presentations and publish papers. Their edge infra is very boring, and yes, they shuffle edge a log and their edge routers like everybody's' else - off the shelf Juniper/Cisco without any tcp session preservation.

Migrating an in-use TCP session from one host to another is far from easy. I did it for a project and the number of corner cases is insane - both in the TCP protocol (what of the window has holes in? What if the connection is half closed?), but also in the OS's handling of the TCP state and interactions with userspace (will be correctly wake up a process poll()'ing a socket if we migrate the socket after a packet is…

I’m not saying it’s easy, but a company like Google should have no problem implementing this. If you ever used Vmotion it does a good job of migrating a live VM to another physical host. Also enterprise firewalls have no problem moving TCP/NAT state from an active to passive firewall.

Re: GCP Incidents

#162

Post by berniedurfee »

NineStarPoint wrote: post #143
vel0city wrote: post #95

It's hilarious people are bashing GCP for having one compute instance go down and the author acknowledges it's a rare event. On AWS I've got instances getting forced stopped or even straight disappearing all the time. 99.95% durability vs 99.999% is way different. If they had the same architecture on AWS it would go down all the time IME. AWS primitives are way less reliable than GCP, according to AWS' docs and my ow…

This is very different from my experience. In my years with AWS I’ve only had an instance get stopped once for a reason that was weird AWS background stuff that had nothing to do with my application. I don’t think I’ve ever had or even heard of an instance just disappearing.

Same. 12+ years of using AWS and there’s been 1 instance of a server (RDS) going down due to something outside of our control.

Restoring a snapshot got us back running quickly. If we were multi-az, we probably wouldn’t have noticed.

Re: GCP Incidents

#163

Post by yolovoe »

vel0city wrote: post #146
NineStarPoint wrote: post #143

Earlier quoted context omitted.

This is very different from my experience. In my years with AWS I’ve only had an instance get stopped once for a reason that was weird AWS background stuff that had nothing to do with my application. I don’t think I’ve ever had or even heard of an instance just disappearing.

By "disappear" I mean the instance failed hard and couldn't be restarted. It's just gone. Usually related to the EBS volume dying. But yeah, usually when they die they can just be relaunched. Still they die way more often on AWS than in GCP, and will just end up staying stopped. Until very recently they couldn't even migrate the instances when the underlying hardware had some maintenance, you had to stop and relaunch…

I don’t know why you’re getting downvotes. What you’re saying sounds true to me, and I work in the core of EC2.

I am guessing you’re using newer instance types if their reliability is still questionable. Or you have a huge fleet of instances so you see a steady rate of failures every year.

Our failure rate on the commonly used instance types if fairly low. We have several types of failures and in some bad failure cases, live migration isn’t possible and your instance won’t even be restarted.

AWS already asks people to expect failures and plan around this with multi AZ deployments.

If you want stability, sign an NDA with AWS and ask for fleet wide reliability metrics for various instance types. There’s a surprisingly huge variance.

Re: GCP Incidents

#164

Post by nijave »

jwnin wrote: post #102
twisteriffic wrote: post #75

Earlier quoted context omitted.

Azure is great if you stick to the three golden oldies: DTU-based SQL, app service and service bus. Maybe table storage if you're feeling lucky. Anything else leads to pain and $$$, because it's likely that no one at Microsoft is being forced to use it.

vCore SQL is solid and predictable. Azure's vm offering is also highly reliable, and they host more linux workloads than Windows now.

> Azure's vm offering is also highly reliable

Not in my experience. Running just under 100 VMs, they'd randomly fail and restart about once a year. One month, something went terribly wrong with our k8s cluster and nodes were becoming unhealthy and being replaced every few hours to the tune of 500 replacements in a month for a 60 node cluster.

Premium SSD v2 is fairly good released generally about a year ago. Premium SSD was pretty painfully slow.

Re: GCP Incidents

#165

Post by BrutalCoding »

markdog12 wrote: post #79
BrutalCoding wrote: post #51

Earlier quoted context omitted.

I’d say Flutter, a GUI framework backed by Google and it’s open source. I’ve recently ported over a popular project called “llama.cpp” to Dart (language behind Flutter) and I’ve recently made YT video’s showing it running natively on macOS, Linux, Android, iOS, iPadOS and next up is Windows. The official Ubuntu installer is made with Flutter too nowadays. But to be fair, last time I tried QT was somewhere in 2018, it…

Is the port open-source? I'd like to try it, if so.

My bad, didn’t notice your comment earlier. Yes it is, here’s the repo: https://github.com/brutalcoding/aub.ai

Re: GCP Incidents

#166

Post by ForHackernews »

wg0 wrote: post #69
ForHackernews wrote: post #46

Earlier quoted context omitted.

I hear this over and over, but what is the alternative cross-platform stack for building rich local GUI apps? Qt?

I see Electron, Tauri/Capacitor and friends as a viable route. Second choice in terms of DX is Flutter. There's nothing like that out there in terms of DX although there are tons of other issues notably around performance. Third would be Qt but that's not a viable tool because of licensing. The first two options have no such licensing issues.

Electron is just webdev but you have to ship a whole freaking browser along with your app. I haven't heard of Tauri/Capacitor, maybe I'll check them out.

Edit: https://github.com/capacitor-community/tauri

" This platform is in alpha, expect broken parts. "

So I guess the answer is still "nothing, really. Maybe Flutter"

Re: GCP Incidents

#167

Post by yencabulator »

Bluecobra wrote: post #161
londons_explore wrote: post #151

Earlier quoted context omitted.

Migrating an in-use TCP session from one host to another is far from easy. I did it for a project and the number of corner cases is insane - both in the TCP protocol (what of the window has holes in? What if the connection is half closed?), but also in the OS's handling of the TCP state and interactions with userspace (will be correctly wake up a process poll()'ing a socket if we migrate the socket after a packet is…

I’m not saying it’s easy, but a company like Google should have no problem implementing this. If you ever used Vmotion it does a good job of migrating a live VM to another physical host. Also enterprise firewalls have no problem moving TCP/NAT state from an active to passive firewall.

> Also enterprise firewalls have no problem moving TCP/NAT state from an active to passive firewall.

Middlebox TCP state is a lot simpler than end host TCP state.

Post reply on HN