3,448 karma · joined November 4, 2013
formerly: Principal Engineer of "Core Technologies" at Discord. https://discord.com/
[ my public key: https://keybase.io/jhgg; my proof: https://keybase.io/jhgg/sigs/NfC1mAm4fGJgKsLzYcshaSlYAfvMnxgLUZgee3nWqzI ]
[1]: https://www.irs.gov/charities-non-profits/public-disclosure-...
One could of course debate whether or not OpenAI would be the best stewards of that knowledge or aligned with the best interests of humanity. However, it is important to recognize that building a successful business is key to funding the research, H100s aren't cheap. It's also important to note that as with all things tech, price of hardware will go down, and OSS models continue to get more capable every week.
As for the QR code login, I built that feature. Although it does offer a venue for social engineering, we've done a lot since launch to ensure people understand that they're logging into a new device using it. From day 1, it's always included red text that said roughly "you're logging into a new device using this" and to "not scan codes that random people have sent you." Of course, some people don't really read. That being said, millions of users every week use the QR code login legitimately, and it's a feature that most other chat platforms offer. It's also very very beneficial when you're using a shared device (i.e. you live in South Korea and visit PC bangs often.)
As well, you'd be surprised, webauthn adoption within companies is not nearly as ubiquitous as you'd think. Shipping out yubikeys around the world during a pandemic was a gargantuan task. Either way, any post advocating for more broad adoption of webauthn and also showing success does the industry as a whole good.
Also, I'm not sure what's given you the impression that there has been any push-back on the feature?
Picking words people are familiar with in order to convey a concept is a generally smart thing to do. A discord server was meant to be analogous to a ventrilo server. Calling them servers made that idea clear to our users, which helped users understand our product and what purpose a server served. I would say that the term has served its purpose well.
Internally we call them guilds, because having a "server server" is a bit ridiculous.
so pgbouncer connects to cloud sql proxy, which does the connection pooling concern.
everything but postgres runs on one machine. we use haproxy to load balance replica connections across different pgbouncer backends for replicas of the same db.
however, i think that you may have a lot left on the table in terms of optimizations that can be made. these numbers seem far under the performance of what i know beam to be capable.
My point is, the capability is useful to some people, and there are many other ways that doing arbitrary things at build/compile time can be useful or make things easier. The sqlx example is one of many.
Another usage, is calling out to another tool, e.g. a protobuf code generation tool. That requires the build toolchain interacting with another tool, that would "break the sandbox."
Ours is quite simple! We take the token range (int64 min -> int64 max), and sub-divide that into N equally sized ranges within the range that span the entire range. We use sqllite to store a database of "work to do" this contains the token ranges, some stats about work done so far, status, and last pagination token (this lets us pause/resume the migration.) We run N many tokio tasks that then just pluck work from the sqllite database, and update the database as work progresses, occasionally snapshotting the pagination token. Each task will insert to the destination database with a configurable amount of concurrency, however will only work on one token range at a time. Each task gets its own "connection" that is not shared with the other tasks to minimize contention. Distributing this workload across machines is very simple. We have a command called "sunder" which just subdivides the sqllite database into N many "shards" that contain 1/N the rows of the original database. We can then just copy these shards to other machines and start the migrator on them. We just spun up 5 chonky VMs on GCP and sundered the database, copied the binary and ran it in screen.
Since we already have the types of the rows described in Rust, we simply just plug the relevant types into the migrator. Everything just works, and you're using the exact same code that's reading/writing the data in the live service to also migrate the data (which further reduces the possibility of bugs). Since the migrator is built on-top of our data service library, we also get all the metrics/telemetry built in as well, so we can monitor the performance of the migration in great detail.
Unfortunately, I suspect this is something that happens quite often for Linode, which is that, their most successful clients will eventually graduate off of them.
This may reposition their business in such a way that their most successful clients stay with them.
This attack vector is significantly harder to pull off if a hardware authenticator will assert that the user is logging into the correct domain.