@fribbledom i'm getting the impression that either linux's implementation of fsync, or posix's definition of it, is hopelessly broken... is this not the PCLSR problem (qv. "worse is better") all over again?

@thamesynne @thamesynne My impression is that POSIX fsync is too open-ended to be considered broken. 😀 This is not PCLSR: That was a design choice with a specified correct behavior to go with it, pushing complexity to userspace. This is just broken. According to that Postgres talk: Pre-4.13 Linux? You're just screwed. Do not put fsync in the situation of failing.
sounds really interesting, added to my watch list

@fribbledom The HN thread had some interesting discussion on the experiences and expectations of contributing code and suggestions the project, and in general:

@fribbledom Wow. Before the www wore me down into using e.g. mysql/mariadb I was a postgresql religious nut and very absorbed with how well ACID was implemented and so on.
I can see how the problem was masked by blaming NFS et al.
I've long used DRBD and RAID so I never realized that postgresql might be the actual problem.
Does oracle use direct i/o? I don't know.

@gemlog Postgres people are the ones who discovered this. Is there a reason to believe that My and Maria are doing any better? It's possible that they are using unbuffered I/O of course, but if they do, it's not because they knew about most POSIXes being broken.

@clacke No, you are right. I was more musing about my previous 'religious' notions about ACID.

Sign in to participate in the conversation

Server run by the main developers of the project 🐘 It is not focused on any particular niche interest - everyone is welcome as long as you follow our code of conduct!