enz

Btw, models are available on HuggingFace

> The trigger was outages in cloud services with sometimes significant impacts on other internet services. Shortly before, an approximately 15-hour outage of the AWS cloud in the US meant that not only Amazon's own streaming services but also Atlassian, Docker, Epic Games, and the Signal messenger were unavailable or severely restricted.

If I remember correctly, it was a us-east-1 issue specifically. Why is everyone hosted in us-east-1, especially in Europe where stable and reliable regions are available (eu-west-1, eu-west-3, ...)?

> MariaDB (and MySQL‑family engines) avoid this entire class of problems by cleaning up row versions at transaction time. There is no background janitor. No vacuum lag. No wraparound timer. No need to tune autovacuum workers or throttle I/O to keep the system responsive.

The article seems a bit misleading. AFAIK, MariaDB (InnoDB) have to "vacuum" too. The implementation details are different between InnoDB and PostgreSQL, and maybe the InnoDB's Undo Log approach is less subject to bloat and maintenance cost, but it still exist as the InnoDB Purge Thread: https://mariadb.com/docs/server/server-usage/storage-engines...

1. Yeah, AWS and Cloudflare suffered from bad outages a few weeks/months ago. In my experience AWS has been very stable in the regions I use (us-east-2, eu-west-*), though.

For those using LevelDB, what characteristics made you choose it over SQLite as an embedded key-value store? Speed?

I worked at companies in Paris where it’d be considered "uncommitted" as well, if done consistently. Especially in "small" companies (let’s say fewer than than 20 people). I guess it’s a matter of company culture.

It may be 1000 comments, but including answers to comments, and answers to answers to comments and so on. Since it’s possible to fold "sub threads" of answers in which I am not interested, it becomes pretty manageable.

I say Claude like this: /klod/ which is the French standard pronunciation for this name. https://en.wikipedia.org/wiki/Claude_(given_name)

To read more. I know it sounds cliché, but here is the plan: instead of setting a quantitative bar (e.g., read 20 books in 2026), I have 5-6 topics I want to explore and get reasonably knowledgeable about. That’s the goal.

> Viral traffic from Hacker News, Twitter, etc. fades quickly; One-time spikes provide no long-term value; Focus on sustainable organic growth instead

I guess it depends on the audience. Our audience is tech-savvy and like RSS feeds, and it can change everything.

You need to make one big "spike", then some people will subscribe to your RSS feed, and some of them will silently follow you and read the future articles that won’t make it to the HN front page.

But I still agree with the point.

Got some alerts about unreachable websites and APIs about an hour ago (Europe region). Looks settled now.

Looks pretty bad... Hackers on BreachForums are claiming they did that and now have criminal records (wanted persons, victims files, ...) data, and emails from +16M people. If the files contain info on key witnesses, they are now at risk.

macOS/iOS 26.1 had some minor (but annoying) UI bugs. Some of them are still here after having upgraded to 26.2, e.g., the menu displaying wrong Bluetooth device statuses (despite the device working as expected).

The problem with this approach is that you now have to manage a secret key/secret for a (maybe) a very long time.

I shared this article a few weeks ago, discussing the problems with this kind of approach: https://notnotp.com/notes/do-not-encrypt-ids/

I believe it can make sense in some situations, but do you really want to implement such crypto-related complexity?

Cool. Looks like V8 can use it: https://chromium.googlesource.com/v8/v8/+/93f85699e22df95861...

On what aspect?

> This appears in the release notes as a fairly minor change but it significantly boosts the defense against one of the sneakiest problems in data management - silent data corruption.

I’ve always been puzzled about the fact there was no checksum by default. Integrity of data is a core job of a DB after all. I’m curious to know if there was a technical justification for that.

> What You Need: A computer with a Chromium-based browser such as Chrome, Vivaldi, or Brave [...]

I believe the book focuses on client-side TS apps?

Not completely down, but have experienced high latencies here in France. It seems better now.

The Linux kernel and (neo)vim.

I use macOS which is a UNIX (certified as such: https://www.opengroup.org/openbrand/certificates/1223p.pdf) and it may already be very popular at least on laptops ;)

At $WORK-1 we had to log into the console almost every day as part of our normal daily operations anyway. However I rarely log into my personal AWS account console, the bill is 70-80cts/mo (mostly S3 storage fees, pretty stable over time).

If I were hosting something subject to potential egress overconsumption, it’d be interested I guess

Working on a small embedded vector database lib (à la SQLite). Just finalized the file format, it works pretty great so far. Besides, it’s an excuse to learn rust and low level programming in general.

At previous $WORK we had similar bills. Our Account Manager got us some deals on S3 storage and egress fee (via CloudFront), in exchange for some usage commitment. It was AWS Europe though, it may be different in the US.

Using Fresh (v1) for a commercial project, mostly SSR with a very few islands. So far it has been great for my productivity and it’s reasonably fast for our purposes (we run it on a VPS, not Deno Deploy)

I believe EU could think of making adoption mandatory among merchants (UE-based or not) doing business with UE citizens.

I am playing with a Tillitis TKey, to use it as an HSM. I write its firmware in Zig (works great on that small RV32 chip), while learning it. I am also launching a consulting firm specialized on cryptographic solutions.

I guess encrypted swap would be a good option (easier than messing with mlock()) According to the docs, this is enabled by default.

I put pink noise in my ears. I made the track with Audacity with a duration of 1h, so I know 1h passed when the noise stops. I can take a break and re-iterate.

I believe Go wants to support "ARM64" that just works among the heterogeneous fleets of Arm machines (from Raspberry Pi to Graviton EC2 instances) and let the programmer not to worry about it. GCC seems to do the same if I believe the article linked on "MySQL on ARM": it emits code that dynamically decides to either use CAS or LL/SC depending on the LSE support.