Andrew Werner

🏴‍☠️

db: WAL failover to deal with transient unavailability · Issue #3230 · cockroachdb/pebble We see transient write unavailability of block devices in the cloud (< 60s) that are sometimes detected as disk stalls resulting in node crashes. Whether the node crashes or not, this negatively im...

@alexmillerdb.bsky.social have you seen this software adaptation: github.com/cockroachdb/...? Cockroach was seeing ebs volumes that become transiently write unavailable. Folks regularly run many ebs volumes with one vm (apparently it gives a better iop/$ tradeoff with gp)

The other day I confidently asserted the debugger I work on hits better for more mature software than small, new projects -- thought being code changes are more painful. This morning a user from a tiny startup hit me up to say how helpful it was to debug a 3rd party lib. Feels good to be wrong!

FWIW, the intuition I was missing in that problem was the need to massively tighten the feedback loop by making smaller decisions faster to eliminate such delays, but neither book drilled that thought into me

Feedback Control of Computing Systems This is the first practical treatment of the design and application of feedback control of computing systems. MATLAB files for the solution of problems and case studies accompany the text throughout. ...

IIRC I preferred www.amazon.com/Feedback-Con... due to the treatment of systems with delays (which are common...)

Janert is like Z-transform is hard and out of scope, whereas
Hellerstein is like Z-transform is hard but 1st order approximation isn't and usually works sufficiently, here's an example

Yugabyte uses docdb, a rocksdb fork. docs.yugabyte.com/preview/arch...

Facebook, iiuc, uses MyRocks, a rocksdb based storage engine for MySQL. engineering.fb.com/2016/08/31/c...