I heard about this from many ex-FBers. 1. Poorly documented in-house tools. 2. The API is inferior to public clouds. 3. Too many half baked features due to impact related stuff. #facebook
Fwiw, you could say this for almost every company running their own datacenters. The public cloud providers would likely always beat them in documentation of features, SDKs, APIs etc. Can't comment on half-baked features though.
Doesn't even have to be specific to data centers. This is our industry. Half baked shit everywhere.
Most of the public clouds list "99% uptime". Unfortunately, you certainly get roasted for the 1% downtime event.
Who gives a fuxk tho Honestly
OP is probably prepping for a Meta SRE interview.
Surprised you have many votes but not a single comment from a meta employee?
Yeah it seems there a lot of experts on Meta’s engineering practices but oh wait none of them work at Meta 😂
Yeah, weird. Either Meta employees silently agreeing or the entire poll is bogus.
As an SWE, imperfect implementations imply job security.
That is true except for amazon. At amazon there is no such word called "job security".
Yeah, bruh. My statement was about about the broader demand for the job (profession) called SWE. Not specifically about an individual’s role in an organization. But point taken.
Basically no one is paying FB for nice infra documentations, so it's just good enough for internal engineers to be productive. Can it be better? Of course. But the current inventive isn't there. Also to note, what you described are the user facing side of things, in terms of raw scale and technical design, it's absolutely top of the industry.
Finally, someone from Meta spoke and she’s some light on the above. Thank you, Meta!
Shed*
I used to work at Meta, and would say the infra was quite good. Maybe the documentation was a bit lacking, but honestly not that bad imo. APIs seemed fine honestly, and some of them were honestly quite excellent (EntSchema and logger configs are great). Haven't been there in a bit so won't comment on half baked features. I agree with what others said, AWS is literally selling infra so of course they're going to invest the time in documentation and widely-usable APIs.
what’s entschema and logger configs?
Come and find out 😉
Tupperware is poorly documented, but using it is way easier and better than Kubernetes.
How can it be easier to use if the documentation is so poor?
Because everything is integrated together well. Also documentation might be poor, but instead you just use code search to find examples of what other people did (that’s true for lots of Facebook infra). For tupperware, I would disagree with point 2 as well. Its api is superior to kubernetes. First of all, everything is programmatically generated so you don’t have to deal with crap like helm or yaml.
Fitness
Yesterday
928
Gain muscle without protein powder
Tech Industry
11h
702
Is meta really worth the wait?
Tech Industry
2d
19987
What happens when most of your team is Indian?
Ask Blinders
Yesterday
875
Why is Bay Area terribly dressed?
World Conflicts
16h
1067
Why do Indians support Israel so much( on blind surprisingly) when Israel really thinks 💩of them ?
AWS just had its 3rd outage this month.
It’s funny that you’re commenting IBM.. your cloud reliability is by far the worst by any metric
That might have been true before today's outage. Even our crap cloud hasn't been that unreliable.