RE: https://mastodon.jepsen.io/@jepsen/117535799639100730
This is a hell of an indictment of an LLM-generated database that fails to adhere to basic rules of logic and arithmetic, let alone reliably storing data.
RE: https://mastodon.jepsen.io/@jepsen/117535799639100730
This is a hell of an indictment of an LLM-generated database that fails to adhere to basic rules of logic and arithmetic, let alone reliably storing data.
I built a database that’s got 100 billion times more write performance than Postgres. It’s got built in consistency, instant replication, and a perfect security record. Works across all major distributions and doesn’t require any special adapter for your language or framework. And it is agentic AI ready, quantum proof and blockchain compatible.
Just pipe everything to /dev/null.
Can I get some VC money now?
@tommorris /dev/null is webscale
@tommorris I’ve decided that vc money is a bug, not a feature.
@tommorris embarrassingly bad. Reminds me of a bug my team found in MySQL where, although you couldn't INSERT 100 into a YEAR(2) column you could get it in via some indirection (INSERT 99+1 or INSERT 99 then UPDATE x=x+1 or something, I forget) and create a value that comes out as '00' in a query but sorts above 99. We looked into the MySQL source as we were writing a custom table storage engine and the code was generally shockingly bad and inconsistent, so it wasn't a huge surprise!
@tommorris to give a taste of the state of MySQL at the time (now long ago), the guts of a storage engine was a function to get the next row in the table, which returned a flag to signal end of table. We implemented it and it worked right up until we tested joins, where we (through stepping with a debugger and inspecting other table engine source) found there was also an undocumented global errno-like value that could be used to signal end of table, which was ignored in normal table scans but...
@tommorris ...the nested loop that implemented joins checked for it. Our code never set it so when a MyISAM table in a join with us ran out of rows and signalled it, our table was seen as out of rows too.
There were several other "different data paths used for the same thing in different contexts", where any inconsistency between them would (a) be easily done and (b) cause really weird behaviour that only crops up in certain combinations of situations, it was a real train wreck!
@kitten_tech early MySQL was incredibly rough. Still better than this mess though.