nhoughto

9 ms·

31.
by nhoughto 4y ago
great to hear more details about snapshot/restore in the wild, plenty about firecracker but seemingly much less about this exciting feature in real usecases. looking forward to the unwritten details / future posts too, particularl
32.
by nhoughto 4y ago
Maybe I'm the outlier but at >15 years into this job I think back on what I was like/did/thought 6 years in and shudder, at that point I never mentored anyone my junior seriously enough to influence their perception (I don
33.
by nhoughto 4y ago
Yeah we gave up when we got to using lambda. Even if the api works it can’t scale the same way!
34.
by nhoughto 4y ago
Sympathy is what it all boils down to.
35.
by nhoughto 4y ago
ironically if this is successful we likely won't know how deep the hardening goes, unless security researchers who have legit access release something? but i'd love to know the series of changes that goes into something like this
36.
by nhoughto 4y ago
Worked on something very similar and we chose not to move ahead with this kind of feature because of the phishing risk, without redefining the problem (like webauthn, Fido etc) it is very easy to dupe a user / replay attack. Most users
37.
by nhoughto 4y ago
I built the exact same thing.. and had exactly the same outcome, crazy coincidence! Probably my biggest failure and learning experience so far.
38.
by nhoughto 4y ago
And also they already can, if you believe it: https://www.bloomberg.com/news/articles/2019-05-15/disney-wo...
39.
by nhoughto 4y ago
Good stat, anymore detail? Love to read more into why
40.
by nhoughto 4y ago
Australias NBN for all its painful mistakes at least had the structure right. National wholesale network with strong regulatory power to drive efficiency. Lots of problems but consistent, national problems that require solutions because hur
41.
by nhoughto 4y ago
Interesting they touched on a couple of pain points I’ve had with spark execution, not quite the same thing but I’ve wondered why some of the way it execute isn’t smarter. Bloom filters for cheap way to filter big joins, using single scans
42.
by nhoughto 4y ago
Interesting to see there is still some more to squeeze out of the old girl yet. Most seem like things that could have attempted before now but I guess weren’t the top pain points to fix, I wonder how many more 4-6% savings there are in ther
43.
by nhoughto 5y ago
So lambda invoke direct via api has 6mb limit: https://docs.aws.amazon.com/lambda/latest/dg/gettingstarted-... But api gw is 10mb, not for lambda tho, confuse: https://docs.aws.amazon.com/apig
44.
by nhoughto 5y ago
Yeah didn’t see anything about timeouts or body size limits.. is a good question. Lambda + apigw vs invoke vs alb vs the rest having different limits has got me a few times.
45.
by nhoughto 5y ago
Yeah you need an aws iam identity to call lambda invoke, for public consumer this is better.
46.
by nhoughto 5y ago
Works for lots of examples where you are packaging a statically-ish linked thing into a container, could be node/golang/python/java etc. Def lots of other scenarios where it doesn’t work, but sometimes you can push that to a
47.
by nhoughto 5y ago
Do it all in Userland with jib cli, doesn’t work for all cases but it’s constraints generally keep you honest.
48.
by nhoughto 5y ago
Ah I see, thanks for the insight. Didn’t realize they were like a storage primitive like that (obvious now..)
49.
by nhoughto 5y ago
ah right so they do this to avoid having a proper secure enclave, like a very focused secure storage capability related to what the efuses are logically related to. Makes sense, I guess i just assumed they would have a secure enclave like p
50.
by nhoughto 5y ago
Why does it need to physically modify the hardware via melting fuse when that fuse is read by enclave / boot loader code itself? If trusted code is trusted then couldn’t it store its state securely without melting fuses? I must be miss
51.
by nhoughto 5y ago
seemed more to be technical and organisational. More and more teams all the time, more and more products, more and more technical requirements / tuning etc, needed a strong abstraction to reason about it all. Homelab would be fine =)
52.
by nhoughto 5y ago
I had the same sentiment, I was thinking is this the localstack of the Firebase world? have you doomed yourself to replicate their problems? but no its a firebase _alternative_ not a replacement, well played. Lots of good explainer content
53.
by nhoughto 5y ago
Hmm feels like there is a bit of a misunderstanding on the meaning and cost of threads and how the OS schedules work from threads onto logical CPUs and what state they are in. It is expected that processes might use more than 1 thread, and
54.
by nhoughto 5y ago
yep agreed. I've often thought about what interesting findings you would get back if time and people were infinite and all those little things you thought, 'huh? thats weird' and put it in a queue somewhere and a few days lat
55.
by nhoughto 5y ago
I'll use this I think, very handy. I'd probably emphasis all the text/descriptions more too, lots of important nuggets in there that aren't obvious when glancing, like automating the top portions of the pyramid, humans s
56.
by nhoughto 5y ago
Recent convo with Cloudflare they said they are mostly container-less, debian packages installed and running on the the host, which is impressive at their scale and complexity. They did seem to think that this approach was hitting its limit
57.
by nhoughto 5y ago
ah good tip, I don't care about session affinity or custom balancing algos so that works. I'd imagine running in GKE or AWS you would also avoid the DNAT / conntrack overhead as pods by default use a routable VPC IP instead o
58.
by nhoughto 5y ago
Is this true of the OOTB GKE nginx ingress? Hard to tell by 'load balancer' do they mean nginx ingress reverse proxy? I can imagine the delay between updating the GCP global load balancer service from GKE would be much higher than
59.
by nhoughto 5y ago
I mean a monorepo is basically the modern answer to this, use technology to the solve the coordination problem, its just a different (better) answer. At Windows scale (like Google and FB etc) it would require effort but is do-able, and has
60.
by nhoughto 5y ago
interesting to think about the amount of 'information' that is present in an environment (like smells) that we aren't privy too and don't understand. Its easy to assume it doesn't exist and then be surprised when a

More ›