Having worked on a number of message brokers, but not really fucked with append-only log streams, what's the actually selling point here?
Feels like the simplicity wins are made by only solving one half of the problem space (only writes), and it introduces new problems of its own (like how do we truncate the append-only log once readers are done with it?)
Fun write-up. Good stuff, very nice contribution to to compio of compio-ws. Looks like the Iggy switched to thread-per-core compio back in July.
https://github.com/apache/iggy/pull/1960
One thing I saw mentioned in the 0.5.0 release notes (which happened shortly after migrating to compio): trying to switch to a Sans-IO model design (but maybe just for clients?). It feels like there's similar ownership concerns that might arise, as seen here with the poll-vs-completion challenges faced here. A lot of good learnings to be had, as we go forth to faster better io, eager to keep tuning in & hearing more. Thanks for the write up!
https://iggy.apache.org/blogs/2025/08/10/release-0.5.0/#what...https://www.firezone.dev/blog/sans-io
4 comments
[ 0.21 ms ] story [ 15.7 ms ] threadFeels like the simplicity wins are made by only solving one half of the problem space (only writes), and it introduces new problems of its own (like how do we truncate the append-only log once readers are done with it?)
One thing I saw mentioned in the 0.5.0 release notes (which happened shortly after migrating to compio): trying to switch to a Sans-IO model design (but maybe just for clients?). It feels like there's similar ownership concerns that might arise, as seen here with the poll-vs-completion challenges faced here. A lot of good learnings to be had, as we go forth to faster better io, eager to keep tuning in & hearing more. Thanks for the write up! https://iggy.apache.org/blogs/2025/08/10/release-0.5.0/#what... https://www.firezone.dev/blog/sans-io