chris.blue
Writer of books (http://themissingreadme.com), code (http://slatedb.io), checks (http://materializedview.capital), and newsletters (http://materializedview.io)
2,514 posts
6,998 followers
181 following
Regular Contributor
Active Commenter
comment in response to
post
I do agree it’s kind of a trap for engineers (like stream processing). Graphs are such an elegant and pure structure (especially RDF triples). Low latency, partitioning, GPU operations (squaring matrices for friends-of-friends). It’s all very tempting.
comment in response to
post
Yea. My experience has been that realtime graph databases are very important in large social media, fintech, and mayyyybeeee recommended systems.
For everyone else, batch is fine.
comment in response to
post
Yea... IIRC, @frankmcsherry.bsky.social had a similar finding with Timely/Differential Dataflow. It was faster than GraphDBs on cyclic computation.
comment in response to
post
Windsurf is my go-to
comment in response to
post
Yup! CXL
comment in response to
post
Yes! That Philz is close to me, so it's very convenient. Already looking forward to the next one. 😀
comment in response to
post
This was my prompt
comment in response to
post
Brb, checking if it uses bitcoin-style encryption.
comment in response to
post
@qianli.dev @petereliaskraft.net Thanks so much for organizing
comment in response to
post
Search and social graph were actually quite similar at LinkedIn. They were part of the same org. The distributed graph was deployed and had a topology quite similar to search.
comment in response to
post
Yea.. I think these two are interconnected. To get good perf, everything needs to be in memory. But that doesn't scale, so you need to keep it data in memory across multiple machines. To do traversal, you have to scatter gather. Now you've invented a search index.
comment in response to
post
I think they might be coming back. Have you seen www.graphiumlabs.com?
comment in response to
post
"Rewriting a project of this size from Python to Rust would require a complete reimplementation of each module, the parser generator, and the tests. That’s a significant effort and far beyond what can reasonably be delivered in a single answer or small pull request."
Interesting...
comment in response to
post
But which order?
comment in response to
post
Are any fully remote?
comment in response to
post
comment in response to
post
Hah, yea I was really resistant to this as well. But it's too convenient to ignore.
comment in response to
post
Not much I can do. It's Substack. 😢
What happens if you switch into reader mode?
comment in response to
post
Odd. One other person mentioned this when I first published, but refreshing made it reappear.
comment in response to
post
Yea, it drives me nuts. We learned all these lessons with Samza 15 years ago. TBH, though, CFLT was _forced_ to go all in on Flink; they were losing the marketing battle.
comment in response to
post
I find this quote about the stream processing winner fascinating when comparing to the original motivation behind Kafka Streams.
The truth is clients WERE the real winner. Majority of apps are using those APIs and not much more.
comment in response to
post
Yea, someone on Discord was using it. I can test it out if needed. 😅