uvdn7

14 ms·

121.
by uvdn7 4y ago
This is fantastic question! We face something very similar (if not identical). There are two parts to solve this problem 1. monitor and measure how consistent the cache is 2. figure out why they are inconsistent I will focus on #1 in this c
122.
by uvdn7 4y ago
> which is to figure out when to do it. Can you elaborate?
123.
by uvdn7 4y ago
> For any anomaly in a stateful service, it is an anomaly only if clients can observe it one way or the other. Otherwise, we argue that it doesn’t matter at all. Thanks for the comment! I stand by this claim; I would love to hear more ab
124.
by uvdn7 4y ago
I am the author; so I am obviously biased here. I am serious when I say cache invalidation might no longer be a hard thing in computer science. In the post, I explained why cache invalidation is hard; and how we solve/manage its unique
125.
by uvdn7 4y ago
Yep. You're exactly right. And the approach is generic and I think it should work for everyone. The idea is that with all these observability capabilities, debugging cache inconsistencies is getting very actionable and close to how we
126.
by uvdn7 4y ago
I am the author of the blog post. I believe the methodology described should be applicable to most if not all invalidation-based caches. I am serious when I say that cache invalidation might no longer be a hard thing in computer science. AM
127.
Cache made consistent: Meta’s cache invalidation solution(engineering.fb.com)
280 points by uvdn7 4y ago|193 comments
128.
by uvdn7 4y ago
There is this philosophy about software that it needs to be redesigned if the workload scales by 10x. The same applies here. When we are studying a topic, context (scale in particular in this case) matters a lot. Just like our physical worl
129.
by uvdn7 4y ago
fzf is an amazing piece of software. Thank you, Junegunn Choi.
130.
by uvdn7 4y ago
Exactly. The definition of “one thing” is unclear. One house is one thing, which also has a thousand things in it. Car is does the “one thing” — driving, but steering is arguably a thing as well.
131.
by uvdn7 4y ago
At the time, it's very hard for Google to predict how bad things could be. Ads business tends to get hit very hard when the overall market is in a bad shape, no matter how great the search engine is.
132.
by uvdn7 4y ago
There is no distributed system without Lamport. His Time, Clocks and Ordering paper to distributed system is like Ted Codd’s A Relational Model paper to database (which is also math!). His elegant analogy between the special relativity, spa
133.
Feedback Loops(blog.the-pans.com)
1 points by uvdn7 4y ago|0 comments
134.
by uvdn7 5y ago
And not to mention better CPU cacheline hits!
135.
by uvdn7 5y ago
Abstract a fleet of machines as single super computer sounds nice. But how about partial failures? It's something that a real stateful distributed system would have to deal with all the time but a single host machine almost never deals
136.
by uvdn7 5y ago
This is actually what makes handling the distributed system in a single computer easier – everything crashing together makes it an easier problem. E.g. you have multiple CPU cachelines, caching different values of a main memory location. An
137.
Little's Law and Service Latency(blog.the-pans.com)
2 points by uvdn7 5y ago|0 comments
138.
by uvdn7 5y ago
I am skeptical to say the least. TrueTime, PTP, FB’s time card, all solve one problem that is to provide a bound where “true time” falls in. You need to be able to _guarantee_ the precision. If you say the true time lies between [T - d, T +
139.
Don’t point out something wrong immediately(blog.the-pans.com)
371 points by uvdn7 5y ago|169 comments
140.
by uvdn7 5y ago
This is a fork. How is it different from the source repo?
141.
by uvdn7 5y ago
My impression is that MariaDB has good ARM support out of the box. I am not sure about ARM support status of other popular DBs.
142.
by uvdn7 5y ago
What I like about vim (and other terminal editors alike) the most is the predictability. For any decent tools, you will get used to them over time, things become muscle memories; and that’s good. You learn it once, and you can benefit from
143.
by uvdn7 5y ago
They are not mutually exclusive? If you look at China, both are sort of happening at the same place in the same place.
144.
by uvdn7 5y ago
The same argument can be applied to Linux and apply to virtually all companies having a backend?
145.
by uvdn7 5y ago
I don’t think Paxos is superior than 2pc. 2pc is dramatically simpler to implement. And if you layer 2pc on Paxos (something highly available) you can get the best of both worlds — simplicity and fault tolerance. The thing with Paxos is tha
146.
by uvdn7 5y ago
Yes, it is correct. At a distance both 2pc and paxos are solving the consensus problem - a group of entities agreeing on _something_. In the 2pc case, the _something_ is commit or not. In the paxos case, the _something_ is the chosen value.
147.
by uvdn7 5y ago
It’s not entirely useless. It costs big companies little to make this type of investments to avoid a potential big miss longer term (think of IBM, Nokia, etc.) Big companies are paranoid about missing the next big thing, so it makes sense t
148.
by uvdn7 5y ago
> When you're at Roblox's scale Yet a regional Consul deployment is the single point of failure. I apologize if that sounds sarcastic. There’re obviously a lot of lessons to be learned and blames have no places in this type of
149.
by uvdn7 5y ago
https://brooker.co.za/blog/2022/01/19/predictability.html This might be something you are looking for.
150.
by uvdn7 5y ago
The relational model is orthogonal to transactions. Ted Codd literally called out consistency challenges in the paper. Transactions were not invented then.

More ›