Hacker News new | comments | show | ask | jobs | submit login

I work on memcache at Facebook, and cache invalidation is a hard, hard problem. It's easy to think that we just overlooked this trivial fix, but the reality is that this just doesn't solve all possible cache consistency problems. Namely, how do you deal with racing writes where the clocks are skewed? How do you deal with multiple copies of the same data in different caches (different data centers, for example). What about axe-murdering your top level db because that data is inherently uncacheable using this scheme?

I'm not trying to hate on the post at all, I think it's awesome that product people are thinking about this problem, and the more of them that do the better the world is going to be for guys like me. I just want them to continue to think hard about it :)




good points. but to be fair most problems become hard, hard problems when your site operates at large, large scale. for small traffic sites, caching architecture is easy and most of the subtle vulnerabilities don't manifest.

background: i've been a senior staff engineer at Orbitz and i did a lot of work on the caches at Cheaptickets.com




Guidelines | FAQ | Support | API | Security | Lists | Bookmarklet | DMCA | Apply to YC | Contact

Search: