Vaughan Reid

Another option I haven't seen yet, is using testing centres. Cheaper for companies to screen without flying them to the office.

Oh I see, so either way we need to be careful.

Btw I started reading your blog when I worked at AWS a few years ago. Its hard to find the same level of information on running distributed systems since I've left. Thanks for wading through the weeds for us!

Great post Marc, it makes a lot of sense that your team chose SI. Most clients most probably have to deal with write skew anyway.

I was wondering, if we would expect the adjudicators to abort one of the transactions if they used SELECT FOR UPDATE? Would that then classify it as a write?

I have a new blog post on Amazon Aurora DSQL and it's choice of snapshot isolation leading to the potential of write skew. It's a good choice based on what they want to achieve but it's important to understand what concurrency bugs a weaker isolation can produce. www.vaughanreid.com/2024/12/amaz...

Great talks Marc! Really interested in the approach you all took with DSQL. Am curious how the adjudicators will be tuned to minimise write skew across so many different workflows! Will lookout for more papers from your teams.