Hacker Newsnew | past | comments | ask | show | jobs | submit | leetrout's commentslogin

You should almost never stop testing in prod.

https://www.honeycomb.io/blog/testing-in-production


Related check out chain of draft if you haven't.

Similar performance with 7% of tokens as chain of thought.

https://arxiv.org/abs/2502.18600


That's a comparison to "CoT via prompting of chat models", not "CoT via training reasoning models with RLVR", so it may not apply.

This seems remarkably less safe?

Would would we want to purposely decrease interpretability?

Very strange.


Turn on power saver and try again :-/

It is also terrible to scroll on iOS with power saver turned on.

I looked around for reasonable datacenter colo in the triad and triangle and didn't get anything promising. Where are you physically hosting your infra?

Curious if you are having to buy bandwidth as well. Some of the Midwest data centers include over 30TB of bandwidth in the rack rentals.

And if you are willing to go into the details curious how you are handling bare metal provisioning. MaaS or home grown tooling? Or are you just installing proxmox by hand?


We don't have on-demand API-based bare metal provisioning right now. Sorry if that is misleading on the website. Our bare metal is OTC right now (over the counter).

For the rest of our provisioning (VM and container) I wrote the software myself. It's based on a Django app called the "master" that hosts the console and keeps track of who has rented what etc + a bunch of "host" nodes that listen for instructions from the "master". Pure python, the only thing that's in Go is the CLI.

I looked into Proxmox but ultimately decided I wanted full control. ZFS storage from Proxmox is something I do sometimes wish we had — going to offer s3-compatible storage very soon but I know Proxmox does ZFS out of the box really well.


How do you sanitize hardware between workloads?

> left my job earlier this year

They are probably reselling from hetzner/vultr


We own all our own hardware but we rent from a colocation provider in Charlotte. They provide the backed-up power and network + the IP block, we do literally everything else. We used to be in my house and I was considering installing my own generator (actually not that expensive) but we got a great deal on colocation, shout out Ivan Teoh at Tier.net!

That's a surprising amount of investment, well done and godspeed.

The Hetzner US server on the east coast is in Virginia not North Carolina.

Which actually isn't that far. But I don't think they could fake it if they say the servers are in NC.


Why would they need to buy bandwidth? Many places would offer you a gigabit pipe directly.

Sure, some places charge to upgrade. Some places are metered.

100Mbps is 33TB / mo. 1Gbps is 330 TB / mo. I'm curious what they have and how they prevent saturation from a client or they just pass on networks managed by the DC.


Replying to myself since OP shared elsewhere they are using Tier.

$950.00/month for 100TB @ 1Gbps

20A @ 208V

61 Usable IPv4 or free BGP

And half rack is $650 / mo for 30TB @ 1Gbps (so basically the full month at 1/10th full pipe).

https://www.tier.net/colocation


> Many places would offer you a gigabit pipe directly.

i.e. you don't have redundancy? A gigabit is nothing these days sadly. You can get a surge in traffic (DDoS or not) and be stuck quite quickly. You always need extra capacity to deal with issues unforeseen and that adds to the cost.


Have you tried Echo instead of Gin? I find it to be much more friendly and approachable with its docs compared to Gin.

Echo is really great framework. Blazzing fast and great documentation.

Tbh not yet, I heard that it’s more user friendly, but I go with gin because it has larger community and support!

Would love to explore different libraries


> Two have sadly taken their own lives despite seemingly good FAANG careers

Sorry to hear that.

Unfortunately I think we have way over indexed on "success" being tied to money and seeking these careers at companies that drive people to exhaustion and let the competitive environment drive everyone harder and harder with a ratchet effect.


IMO it’s less about success but a lack of reliable safety nets. Absent a good supporting environment, what choice does an individual have than to maximise their own outcome.

Not just safety nets, but the disappearance of the middle class at least in the US. It increasingly feels like people either make twice or more what they need to live, or half what they need to live.

I can absolutely see why parents see the way things are, try to extrapolate out another 20 or 30 years, and feel like they have to make sure their child is in the "well-to-do" group. It feels like the days are gone where you could be an average performer at an average job and live an okay life.


Fighting for the crows nest on the titanic

Do countries with reliable safety nets have lower rates of taking one's own life? What does reliable mean for a safety net?

Curious if anyone in the thread has / is using windmill?

They don't seem to have jumped for AI hype (yet?)...

https://www.windmill.dev/


I've tried it, but there's too much "sorry not in the open-source edition, please buy the entreprise edition" stuff all around, which makes it quite unusable

have used it, and i do like it, but the licensing situation is not great. It open source but its not free software by any means.

Have you used AWS bedrock? I assume these get pretty affordable with prompt caching...


The often excluded option is dynamically generating JSON and feeding that to TF instead of HCL.

You can combine it with tools like Dhall or my personal preference Jsonnet instead of imperative languages for an interesting experience for reusable pieces outside of module concepts.


Any particular libraries you use to generate TF-JSON from jsonnet?

I wrote a generator a little while ago that can create jsonnet libraries from the TF schemas: https://github.com/Duologic/soysonnet

Example lib here: https://github.com/Duologic/soysonnet-aws

I only needed it for AWS so I didn't spend more time on it.


By hand :( But I like your project. Do you use Tanka?


I work at Grafana and did quite a bit of work on Tanka in the past, so yes.


Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: