Hi all, I recently started a Masters in Computer Science (and not at the institution in my email address). One of my courses is "Advanced Databases" - yummy I thought - it's not even compulsory for me but I just *_had_* to take this module. The lecturer is a bit of an Oracle fan-boy (ACE director no less... hmmm...) and I want(ed) - becoming less enthusiasic by the minute - to do my dissertation with him. So, we're having a chat and I make plain my love of good 'ol PostgreSQL as my RDBMS of choice and he tells me that there are problems with random block corruption with PostgreSQL. I said "really" and before that conversation could go any further, another student came over and asked a question. So, I toddled off and did some research - I had heard something about this before (vague fuzzy memories) of a problem with the Linux kernel so I searched for a bit and duly dug up a couple of pages https://lwn.net/Articles/752063/ : PostgreSQL's fsync() surprise - and https://news.ycombinator.com/item?id=19238121 : Linux Fsync Issue for Buffered IO and Its Preliminary Fix for PostgreSQL So, this week I go back to my lecturer and say, yep, there was some issue but it was a Linux kernel problem and not PostgreSQL's fault and has been resolved. He tells me that he knew about that but that there was another issue (he had "spoken to people" at meetings!). I said "well, why isn't it fixed?" and he replied "where's the impetus?" to which I responded (quite shocked at this stage) something like "well, I know that the core team values correctness very highly" to which he came back with "yes, but they have no commercial imperative to fix anything - they have to wait until somebody is capable enough and interested enough to do the work". He then muttered something about this mysterious flaw having been fixed in EnterpriseDB. At this point, I lost interest. Having lurked on lists and going by my general "gut feeling" - if there was a serious issue causing irrecoverable block corruption, I'm pretty sure that it would be "all hands on deck" until this problem had been solved and "nice-to-haves" (GENERATED AS... for example) would have been parked till then. Now, I have four questions: 1) Is my lecturer full of it or does he really have a point? 2) The actual concrete acknowledged problem with fsync that affected PostgreSQL - why didn't it affect Oracle? Or MySQL? Or did it but it was so rare that it never became apparent - it wasn't that obvious with PostgreSQL either - one of those rare and intermittent problems? 3) Were there ever any problems with BSD? 4) What is the OS of choice for *_serious_* PostgreSQL installations? I hope that I have been clear, but should anyone require any clarification, please don't hesitate to ask me. Tia and rgs, Pól...
