IPFS Core Implementations Weekly, 30 Sep 2019

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: Core Implementations 2019-09-30

Description

Description: https://github.com/ipfs/team-mgmt/issues/992#issuecomment-536622085
Notes: https://cryptpad.fr/code/#/2/code/edit/BvXRqE95u8KbwU+FvpAt6Cq1/

A

First, up upgrade the release process. I, don't think, there's been any progress within the release process. A JSF, a fast release is still in progress and there has been no new press release, testing infra.

B

A

It went but just now, okay, his head checked like ten minutes ago. Okay,.

B

No one's approved the blog post. If I'm not most piling, doesn't go past it, but it's been written. Bloopio hasn't been published. Yet we definitely check on him and everything. So I will certainly go back for the release process. Okay, Missy I, like Tessa third party, great I'm, using a nifty little tool, dat cube, oh right off the broker found problem is worth using a diamond. That's what I've made into a currently running the tests of a bunch of the contaminants stuff. So we're gonna.

B

Add all the people doing the common early early testers so that we be intimated.

A

A

So upgrade testing in front. We had done by a rule, I, don't think we published the recording yet, but it's coming along and it's really cool yeah. If you want to take a look at that, I can't find the actual bill here.

A

No, that's not the repo, that's!

A

Okay! That's the repo forecast ground.

A

Its own progress, but it's it's coming along. If you have or one ask, is we have an issue here about you know what request? Yes,.

C

Identify ten.

A

Tests you have comments on test just going to be testing initially, please post them here: okay,.

A

Garbage collection job has been progress, but that's it. hmm Wait comes to.

D

A

That means we can remove. This have removed a section from the core quotations, meaning in awhile.

A

We send an email, shipped shipped I,.

E

Would keep it I would like extend the scope. Okay.

A

Let's see something giveaways I know, updates.

E

Yeah, so, like brief update, is that we merely inserts of synchronously looking at adding support for paths on the developing and that's sort of raised priority for adding support for base 32 in IP and s content post. So generally, a background on this is that we have CID p1 in verse, 42 as a safe representation for that we can use in subdomains.

E

Unfortunately, the peer ids cannot be for, like formally, we do not support base 52 representation of pure IDs and I propose a change to lead p2p specs to make that economical representation like basically wrap multi hash in CID, so that RFC is waiting for a final approval.

E

I think we've addressed more of roles, concerns and separately from that adding support for base 32 in the p2p and bubbling that up to ago, ipfs may take some time so for now, for this specific case around give a blink, we may look at handling that conversion at the engine next level, unless someone from defrosting wants to pick up those go changes.

E

So that's the update and separately from that I Eric asked me to write like more human focused guide around addressing basically like how do you make addresses for gateways that are path or subdomain focused, and how do I construct those native URLs for IPS, : /, so I wrote just a short guide for dogs, ipfs IO and it's appreciate like proofreading, because my English is garbage. But that's that's all from me.

A

A

Let's see distributed, signaling yeah is still on hold.

A

Okay, I pns, the happiness over cups up stuff is waiting for a view for me. Ip answer for dns to go.

C

Yeah not done yet finish this week.

A

Okay, migration to montage keys in the lock store, I, don't think, there's my progress, then I'll spit out so.

A

Mount Dominic.

F

um So yeah I have a PR up for mount an experimental mount thing over the nine P protocol and the read only support for ipfs seems to be alright, there's some bugs with it that still need to be ironed out. But besides that performance seems alright and just general behavior.

F

So fine, so I'm gonna keep looking at that and making sure that that's like stable and protocol accurate and then I also have a fork that has IPS right support, which is functional at the moment, but also a little bit buggy, there's just something to do with how I'm handling the offset that needs to be fixed and then how publishing is dealt with so that when we actually make the writes, we commit them back to the key that they belong to.

F

But, besides that progress coming along pretty well and that's it for me. Okay.

A

Turkish Turkish Turkey, sir.

G

Hey yeah, no, no big updates. Last week, I kind of spent a lot of the week debugging basically I was with connection management. You know stimulative and then on Thursday and Friday I was gonna track down a bug and I ready ton of tests when I eventually realized that I was testing the wrong piece of code. So it was good way to motivate writing tests.

F

Did those tests pass, they.

A

Did yeah that's great? Actually, we could use people in this call to take a look at situation linked. There's, a simultaneous connect funkiness that I introduced could have basically the way we currently had all Simon tipster and go we dialed without everything at once, yeah, and then we see a connection succeed. We cancel all the other dials yeah, unfortunately, if you'll dial each other. At the same time, you see the inbound connection before you see the outbound connection, which means that you canceled your out.

A

There's like you see the impact venture succeed before you see your a potential succeed. You can do at the same time, which means you cancel to your app bound connection and both sides. The same site things they see. The connect is the scene, huh so we're not quite sure the right fix for this one those'll I have is just like say: look like you only connect gifts, connections. When you see one of your connections succeed and the you cancel all your other outbound connections, but you never cancel a basis.

A

Someone else email connection, because you don't know what they're going to do yeah. This would, in some cases, lead to multiple connections, but we do have some logic to try like deterministically pick the best connection to use. We can prove that logic as well, but comets on the sauce on this. If anyone has like prior art on how other systems handle this, this would be wonderful yeah. This gets even trickier when you deal with actual simultaneous connectivity. Spee. Has this wonderful feature where like, if you both sense?

A

In fact it's the same time you don't create do connects to create one connection and you both think he initiated the connected name either you're lucky, but you've all did initiation that's a different book. So, like the bug attachment or the book I was first talking was much more calm or a comment in this. Other books, like the tea, sometimes connection requires, could be precisely the exact same time. This house house.

A

His connect bug requires to need to be a bit less precise because, like the kitchen boots process takes time mini round trips, did he think the TCP when you have to be like within one RTP of each other, so usually with a hundred milliseconds? The bug that's linked here is within, like a second or so.

A

A

G

On Jacob I think didn't we didn't I raise an issue about this in jail, it'll, be a new fixture, simultaneous connection, no.

H

So we still have the issue we solve this same issue. We basically did like connection consolidation, so once we have established a connection, if you have multiple connections we'll end up just using well we'll switch to just using a single connection, so all of our requests will go through the other one and then eventually, like the other, one will naturally get pruned, provided the connection manager kicks on, but yeah we're not explicitly prep prohibiting it.

A

Okay, but you don't have this exact same problem because you'll preemptively kill connections, our problems will preemptively kill all the dials yeah. Oh you.

I

Know it's, it's I think it's.

A

Less thank you for GSP. Don't you download sequence? Yes,.

H

Currently, although that will change with async/await, because we get abort control, okay,.

A

A

H

So we pushed a bunch of updates to live. P2P released several of the modules, including a multi stream select, which will also include header consolidation, because Jess was being really inefficient in someone sending the multi stream packet and then the select packet we're not doing that in one fell swoop, so that'll be more efficient, which is great.

H

We also started a guide for DJ, a slippy to be internal refactor, to kind of break down our approach for that, because the intern also massive that doing it all at once will take forever and will end up getting all kinds of conflicts and it's gonna be nasty.

H

So we have an approach for splitting that out into its individual components, we also are starting a guide to cut for streaming intervals, so we're gonna link there to all the modules that Alan has already created for edible streams like it push push of all all those will all be linked there and then we're also gonna have a usage guide, and so that's going to be for both contributors and users down the line and then eventually we're also going to create a migration guide.

H

So take this pull stream code and you can convert this to this async it or bowles code, so we'll be working on that over the course of this week and we're kicking more off of the lip p2p internals so and we should have a few more of the transport stuff today or tomorrow.

I

Jim I have a question: I was trying to do some juice with Peter Peter stuff on the weekend, and just going through the tutorials and stuff and I was getting into the lit that the current version of the do slip. Ptp is caught using callbacks, but then a lot of the dependencies are now converted to missing collate and right now, if you tried to do the tutorials, if you use for the current version of the Peter Pina police dependencies. Yes,.

H

So if you go to the jail ooh, p2p, async/await PR, there's one right there on yeah.

C

H

Issue 266: we actually have a list of the modules that currently work with the latest version of lit p2p because we are releasing those out. So if you run into issues where you're pulling in async modules and yeah.

I

I'm just wondering this is like anybody pulling the production version, just let Peter be right now it's gonna be pulling in, is gonna, have a broken system right now. Is that intentional, or is that they.

H

Should never be I can affect the tutorials, because it should they all should be minor. Bumps today, single way, see.

I

How the shooter called it was called like the the p2p TC TCP transport. It was trying to do listen on that line, but Allah is expecting a call back, but it simply I.

A

Think the issue here is that, like, if you create a new project, I need to say ham ham, install this new dependency, you could get something that requires a single wait. Yeah.

I

Just broken right now, like it I'm just wondering, maybe we should just put like a big warning on everything like wait for the transition or or is there something we can do with the versioning.

D

Versions too, we have we've actually in a usage, important note that we had in this regard. Well, that's.

H

I mean Jimmy hit it. So it's obviously a problem. Yeah.

I

I mean I'd down great I downgraded the modules I needed two older ones and then they work but yeah.

A

Is there any way to like stop people from upgrading in NPM like mark these experimental unique versions like we have to say, probably go what we usually do. There is like we just say like. Basically, you merge everything as fast as possible in just do everything all at once, so we.

H

Could go in and change all of the NPM tags to point back latest, so we can set all the latest for everything that works on the current version, and then we can just put beta on all the new ones. Yeah then.

A

Like you could still use the new ones, they just won't flip. We even have EMS all. Let's do that. That's down.

H

All those this week.

B

As well trying to do some of the upgrades to the first core, which is pulling in more doors, oh cool, like the test files great and then try to run it and nothing worked.

A

It was like we need tests.

B

A

Design interview from visibles and.

B

A

New ones, that's re.

B

Not played a cyclist as well, and so I landed a PR kind of based on the walk was doing. He was trying to upgrade the eye because when you played to the async/await version and it just spoiled my severe control and I, basically rewritten all of Josiah verse to be chromatids in your core buttocks. So you know I was going to get a bit apart when it comes to my neck. I think.

A

It'll be happiest.

B

But yeah, so that's quite cool, so we chose like the first kind of done and the only thing that I couldn't do the upgrade the pair IDs, because the async/await version, because basically I, have to wait for the pizza P to update a new version then uses the next version of pair IDs, because the stable version I can actually use users like version 112 and they synclavier chosen 13. And then you have multiple versions of things.

B

Then then, like it stops and starts denying the things appear: IDs when they're, actually parodies and so on the circle and once that gets released, and then we can.

A

B

A

There any way in NPM to say there must be a single version of this package like telling him like you are not allowed. Can you just like have a function did like some, how to text? No, you can't know.

B

The whole told the magic is that you always get the version that you want. You should have no.

A

No, but like there's no way of saying I look like there cannot be two persons. So the way this person go is like you can have multiple major versions. You are not allowed to have just can't happen. No, okay, that's Oh!

A

A

Design review proposals anything else this week, I'm so blink lint on my gimmick, specimen plate, 5 spec, which hopefully come out sweet I, have a claim right so tried to hint them. Okay blockers asks anyone blocked it, something you review, you talk, someone need some coffee, okay parking lot. Oh so I guess questions first met questions. A parking lot. Hugo is.

C

Just too plain, no last week well browsers team went to Malta to hang out a little bit do some Road mapping and stuff, and we need also some hacking and from that hacking came out a new browser they restored with around 50% improvements on adding getting and also, we added support to UNIX FS to do batching, which also gives us around 50% improvement, I'm still running proper benchmarks to to validate these numbers and to also make sure if we can compound the to at least in the browser and get a hundred percent improvement in the browser.

C

A

Updated I'm concerned I, remember hearing something about how JSF FS was as fast as go fast in some ways: daddy does this make it significantly faster the browser, oh, that was on dude, okay, got it nope.

C

Yeah, but do the batching is, is for an old and browsing improves box. Okay, good.

B

Faster for a 5 gigabyte data set to the point where it kind of negated goes: look startup, speed and bonds. That just say got it. Okay, Eid crown is still on the guy's head for the time for the topic.

A

For now, okay, that covers everything. Please leave here a sink updates, no well I and I guess weekend and five minutes early on someone else's. Something else talk about Oh.

A

See all thank you.

F