Andrey Satarin

@asatarin.bsky.social

Staff SRE at Google. Distributed systems / databases / reliability / correctness. Views my own. Repost / like is not an endorsement. http://asatarin.github.io

We've been testing Raft implementations, and have found bugs in every implementation we've tested. Here's an in-depth report on one set of issues.

Bild

------- "Jepsen 18: Serializable Mom", Kyle Kingsbury/@aphyr.com , #sd25 Engineers are tasked with building towers of abstraction, building everything higher and higher above the towering tire fire that is databases. "I professionally set those tires on fire".

Good old Microsoft published some work: - "Deploying a Steered Query Optimizer in Production at Microsoft" dl.acm.org/doi/abs/10.1... - This great talk "The Cascades Framework for Query Optimization at Microsoft" touches on correctness youtu.be/pQe1LQJiXN0

Deploying a Steered Query Optimizer in Production at Microsoft | Proceedings of the 2022 International Conference on Management of Data

You will be notified whenever a record that you have chosen has been cited.

dl.acm.org

To counter balance your argument some big tech work on correctness of (SQL) database in a thread below. Spanner has incredibly sophisticated random generated checks internally, this just scratches the surface: - "Randomized Testing of Cloud Spanner" medium.com/@jcorbett_26...

Randomized Testing of Cloud Spanner

One of the secrets behind Cloud Spanner quality is randomized testing. SQL databases like Cloud Spanner have complex APIs. Complete unit…

medium.com

Deterministic simulation has been entirely pushed by startups. Jepsen is jepsen. All the SQL fuzzing stuff I know of comes from academia or database startups. Larger companies seem to only be winning in the application of formal methods (because they can afford to hire a team of ex-professors?)

When I think of advancements in quality and correctness of databases, it feels most things have come from startups or individuals, and not large well-established companies. Which seems... backwards? We talk of startups hacking out code and megacorps crawling to keep a high quality bar.

MariaDB is improving their isolation level guarantees following the recent @jepsen.mastodon.jepsen.io.ap.brid.gy report. "Isolation level violation testing and debugging in MariaDB" post talks about how they specialize Jepsen tests for InnoDB, make it reproducible and debuggable.

Isolation level violation testing and debugging in MariaDB | MariaDB

Explore how the InnoDB team tackles isolation violations in databases, highlighting debugging strategies and a new option for Snapshot isolation compliance.

mariadb.com

Alexey Gotsman talks about database isolation levels. Starts off with consistency in general: - what is strong consistency - common misunderstandings and pitfalls - serializability and linearizability - weak isolation levels - safety and liveness Video links below.

Isolation level specification

Hi, I’m a distributed systems researcher at MongoDB, photographer, Zen Buddhist in the White Plum lineage, rock climber living in New Paltz NY. I hope to meet fellow researchers here, and I hope we can all settle on BlueSky as our Schelling point for a couple years before we have to do this again.