shachaf

vor 2 Jahrendiscuss

I use signalfd when I can, and this argument doesn't make much sense to me. The main goal of most signals you'd use it with -- like SIGCHLD/SIGPIPE -- is to wake you up and tell you that something happened, not to give you all the information. If you get a SIGCHLD, you can call wait() to get information. If you get SIGPIPE you can poll your file descriptors (though you might be doing that anyway so the signal isn't very useful). For the purpose of waking you up, coalescing isn't a problem -- only missed notifications are a problem, and signalfd handles those correctly.

Of course, for "real" signals that you must handle synchronously, like SIGSEGV, signalfd is less useful and makes less sense (and arguably something closer to Windows's SEH might make more sense). It's an odd historical artifact of Unix that SIGCHLD and SIGSEGV use the same mechanism.

shachaf

vor 3 Jahrendiscuss

The perspective I have in the post is that Multi-Paxos and Raft and so on are still doing pretty much exactly consensus for each log entry, they're just sharing lock IDs/ballots/terms across multiple log entries, and having the leader reuse the first phase -- acquiring the lock, reading the existing state -- across multiple instances of the second phase -- writing with the lock.

I think this is compatible with what you're saying, but maybe makes them seem more similar than treating them as two different problems.

shachaf

vor 4 Jahrendiscuss

Hmm, do you mean N+K+1 (to have enough points for both the data and parity shards)? Why isn't N+K sufficient, fitting a polynomial to N points and emitting K more?

shachaf

vor 4 Jahrendiscuss

I would suggest against relying too closely on this article in its current state. The "precedence climbing" code is needlessly complicated -- it has a nested loop which doesn't do anything, only one loop is necessary -- and it doesn't really explain what's going on (which is a relatively simple idea), just writes some pseudocode and mechanically evaluates it.

shachaf

vor 6 Jahrendiscuss

I didn't see this reply until today either.

On laziness, I only mean that the language feature isn't important. You can certainly represent infinite data using laziness (and you must represent it as some form of codata, such as a function), but it's not really the crucial thing here.

Here's a simpler version of seemingly impossible programs, which is better for explaining what I mean: Instead of bit streams, use conatural numbers, i.e. the natural numbers + infinity. You can define these in various ways (e.g. monotonic bit streams), but we'll use a regular Haskell type:

  > data Conat = Z | S Conat

Ignoring bottoms (which we won't permit), this has all the natural numbers as inhabitants, but it also has infinity:

  > inf :: Conat
  > inf = S inf

Note that, if I give you a conatural, all you can do with it is peel off successors; you can't distinguish infinity from "a really big number", extensionally.

A predicate on conaturals is a total function that takes a conatural and returns a boolean. As in the bit stream case, totality is a very strong requirement. For example, "is n even" isn't a valid predicate, because the way to check for evenness is to peel off pairs of successors until you reach either 0 or 1, but that doesn't halt on infinity. In particular, to be total, every predicate must give up eventually: After seeing k successors (for some unknown k), it has to give up and return true or false. If there was no such k, the predicate wouldn't terminate on infinity.

Now you can ask: Given a predicate, how can I decide whether there's any input that it holds for? And this is "seemingly impossible": You can test the predicate on infinity, and, say, on every finite number up to some N, but that doesn't give you any guarantee about the numbers you haven't tested. If all you can do is query p with any finite set of Conats -- even interactively, deciding what to ask it based on its previous responses! -- you don't get enough information to decide if the predicate is satisfiable.

The trick is to construct a special conatural which is defined in terms of p itself. If you think of this more imperatively, you can imagine a program that queries the predicate about increasingly larger natural numbers -- 0, 1, 2, 3, ... -- and prints out an "S" each time the predicate returns false, and a "Z" when the predicate returns true. If the predicate is false for all inputs, it prints out "S"s forever -- but that's still a valid (productive) conatural number.

This conatural is -- by construction -- the smallest input that satisfies p, if there is one, and infinity if there isn't. Now you can apply p to this number, and, if there's any input that satisfies p, p will return true.

The point I was making above is that it's crucial for this conatural to be able to call p. It doesn't introspect p, in the sense of being able to access its source code, but it's not independent either. And, although laziness expresses this elegantly, I wouldn't really say it's the core of the trick here. You could express the same thing using Turing machines that write out "S"s to their output tape, for example (as long as you don't require them to terminate overall, only to be productive). You do need to guarantee that the "conatural" Turing machine that you give to p has access to p (either by embedding its source code or by being able to call it as an oracle).

Here's the implementation in Haskell (which is simpler than the bit stream case given in the article, and doesn't need mutual recursion):

  > -- find p returns a conatural that satisfies p, if one exists.
  > -- (If one doesn't exist, it returns infinity.)
  > find :: (Conat -> Bool) -> Conat
  > find p = if p Z then Z else S (find (p . S))

  > -- Does any conatural satisfy the predicate?
  > exists :: (Conat -> Bool) -> Bool
  > exists p = p (find p)

shachaf

vor 6 Jahrendiscuss

No, the article's p takes a function:

> type Cantor = Natural -> Bit > (#) :: Bit -> Cantor -> Cantor > x # a = \i -> if i == 0 then x else a(i-1)

(You could represent bit streams with something like data Stream = Cons Bit Stream in Haskell, of course, but that's not what the article does. The code in the article works in e.g. Python almost as-is.)

And you can certainly pass closures to the predicate. Not closing over a mutable variable, of course -- I just mean that the bit stream function itself has to have access to p, which is the trick that makes this possible. If you couldn't do that -- if you could only query it with bit sequences that were defined independently of p -- this wouldn't be possible.

shachaf

vor 6 Jahrendiscuss

Laziness isn't the essential thing here -- the article's construction specifically doesn't rely on laziness, and would work in a strict language (almost verbatim -- you might need to write "lambda i: f(i)" instead of "f" in a few places).

I'd say that the thing that makes the article's version work is that you can pass a closure to the predicate, which is a channel for communicating extra information that makes it not quite a black box. In an imperative language, you could imagine passing in a function that mutates some state to get information out (or throws an exception, the way you set it up). The article shows that even that isn't necessary: You can pass in a sequence (represented as a function from index to value) that closes over the predicate itself, and recursively uses it to build a bit sequence that satisfies it, if one exists.

By the way, there's a clearer (I think) version of the core idea here that uses conatural numbers (the naturals + infinity) that people don't usually present, but might be worth writing up.

Submissions

Expectation and Copysets

The FLP Theorem

Show HN: Mustardwatch: Detect what files a program uses, rerun when they change

Git Workflow Is Snapshot-Isolated

Constraining writers in distributed systems

Consensus Zoo

comments