text stringlengths 0 1.24k |
|---|
**Johnny Boursiquot:** So how do you deal with failures? Basically, you have a situation where a user makes a comment. You reflect the comment on the page, so as far as the user is concerned, the application responded very quickly, great UX, and they're on about their business... But now you have an asynchronous operat... |
**David Hernandez:** So once the comment in question is ok to the user... it's basically okay to the user; we don't report it back. So if the Pubsub failed for some reason, it has an in-built mechanism to do retries. We check if the error is temporal or retriable. If it's retriable, we try again for a certain number of... |
\[56:23\] It's also not very complicated, but it's something that you have to continuously revise, especially -- we released a couple of weeks ago, or even a week ago... We've been using the product for a while ourselves; one thing that we did a few months ago is turn off Slack, turn off any other project management to... |
So yeah, it's a matter of strength to do your best, trying to retry if the error is retriable, log it and report it to analyze it after a while, when you have a few samples, and decide that error you should keep it, you should improve it, or you just leave it. |
**Mat Ryer:** Yeah, that comment is a great example, because if the delivery -- essentially, there's a notification that gets created for each person in a conversation. If one of those fails to get created - because it's a comment, and there's a conversation - probably someone else is gonna say something after, and tha... |
And so if the user was to click lots of times, of being like "Interested"/"Not interested" (you can toggle it in the UI), if the user were to click that and we didn't have any sort of debounce protection, that could end up with there being lots of messages in the system (some of them could be lost, and things), but bec... |
\[59:43\] Some things when it comes to design - we have a lot of experience of building various systems. David doesn't like me saying things like this, but he worked on a project for the Olympics, which is massive scale... And think about all the messaging that's going on in a system that's there to support the Olympic... |
So yeah, then you think "Okay, let's assume it's gonna fail. We'll design for that", and you can kind of build systems that are somewhat self-healing. It's really amazing to see these things just -- yes, we see error reports, but by the time we go and look it sort of self-healed, just because of the design of it, which... |
**Johnny Boursiquot:** This leads into my next question, which is sort of -- you're using a non-relational data store, and that means you can't easily join things and provide the latest and greatest to the user... So did you factor in some sort of eventual consistency model to the data you're returning? To use a common... |
So if the user ever changes their name or their avatar, how do you remediate, how do you resolve that across all the different copies of that data you've made? |
**David Hernandez:** It's kind of best-effort. There are some times that you should do it, and some times that you just show it as historic data. If you've been commenting-- |
**Johnny Boursiquot:** It's not a bug, it's a feature. \[laughs\] |
**David Hernandez:** Yeah, exactly. It's like "Oh, this is the story, what you did with your old name. And this is -- you changed your name, it's fine. The newest things. You have it with a new name, a new profile picture", or whatever you used it's new, isn't it. It's kind of an original blockchain if we see it. |
**Johnny Boursiquot:** \[laughs\] |
**David Hernandez:** That works pretty well sometimes. In that case, probably nobody's going to complain that "Oh, old cards display with all the information about me." It really depends, but in other cases we struggle to issue updates to just "Oh, if you rename..." We have for example tags or labels for categorizing t... |
But this is just a trade-off. You sacrifice reading for writing, and you sacrifice writing for reading. So it's not a good solution for everything. You just need to design your application accordingly. You probably do some mistakes in both cases. When you have joins, you have to probably know "I made too many joins. Wh... |
**Mat Ryer:** One of the nice things is we are acting really in the product role and the technical at the same time... And in the past, when I've worked at places where they've separated those functions out - that creates a lot of friction, because you then have a situation where you've got the product person fighting ... |
\[01:04:23.06\] Often you don't have that, and what's nice about -- since we both fulfill those two functions separately, or rather we're kind of fulfilling them at the same time, we kind of get to think about "What's the user experience we want with the realities of the system?", and that allows us to if not always de... |
**Jon Calhoun:** I think it's also worth noting that -- like, for our listeners who might not be familiar with both a SQL database that has these joins, versus a document store where you normalize data and copy it over, that problems exist on both sides of the table. Both sides will have their own separate, similar pro... |
I guess the one that comes to mind is let's say you're Amazon, you're shipping packages to a user, and they have an address. And you have a relational database where you query the user and do a join with their address, and pull the address up... But if the user goes and edits an address and changes it, depending on how... |
**Mat Ryer:** Yeah, it's true. I think you're at risk of not being as rapid. If you use just a SQL data store, you can prototype and you can sort of throw data in and query in different ways. You can do a lot more of that. Having said that, I haven't found that our creativity has in any way been stifled by having a sch... |
So you need a dedicated index for every kind of query you're gonna be doing, essentially. It creates by default single-field ones for you, and your job then is to go and exclude any that you're not gonna need. But having that - that's quite strange initially, because if you wanna just rapidly prototype in the browser a... |
But honestly, it hasn't slowed us down at all, and of course, the trade-off is if you go to Pace.dev, if you actually play with it and use it, it's lightning fast, and it'll stay fast, because of the nature of that choice. The reason why that data store is so limited is so that if you use it properly, you can deliver m... |
**Jingle:** \[01:07:50.09\] |
**Jon Calhoun:** David, Mat, do either of you have an unpopular opinion you'd like to share? |
**Johnny Boursiquot:** Is pairing with Mat an unpopular opinion? |
**Mat Ryer:** \[laughs\] |
**David Hernandez:** Well, it's one of the things that we usually do while we are pairing. Mat gets the guitar and we try to do something completely different and build a little song, or something like that. |
**Mat Ryer:** Yeah, just do some songwriting instead. \[laughter\] But actually, an unpopular opinion I have is you should try and work in tiny teams. A lot of the problems when it comes to software engineering come at scale. And that's not just code scale, but people scale. So if you can have tiny little teams working... |
You can be so effective in such a small group, because you cut out a lot of the work needed really to marshal the team. You can't always do it, and it sounds a little bit anti-social, but that would be my unpopular opinion. Tiny teams. Do you have an unpopular opinion, David? |
**David Hernandez:** I don't have an unpopular opinion. I'm very populist, probably. \[laughter\] Going with the flow. |
**Jon Calhoun:** You'd probably have to be "Go with the flow" to work with Mat that much... |
**Johnny Boursiquot:** Oh, man... |
**Mat Ryer:** I'm getting a grilling on this. |
**Jon Calhoun:** I'm just kidding, Mat. |
**Johnny Boursiquot:** Yeah, we kind of warned you though. You knew this was coming. |
**Mat Ryer:** Yeah, but if you knew David... |
**David Hernandez:** \[laughs\] Yeah, that's what I was about to say. You don't know me. I'm congratulating Mat for working with me for a while also. \[laughter\] |
**Jon Calhoun:** So you're telling us that if he turned his camera we'd see a blank wall right now, but really if he turns it the other way it's a bunch of unpopular (I don't know) posters, or something? \[laughter\] |
**Mat Ryer:** No, he's alright. |
**David Hernandez:** I usually tell Mat that he's a terrible developer. His ideas are bad, so I'm not sure why he's working with me... |
**Mat Ryer:** Yeah, I thought it was a language thing when you would say that my ideas are a terrible idea. I thought that was just the language, but no; it turns out he does think that, and he will tell you. \[laughter\] |
**Jon Calhoun:** Is that why you spend so much time trying to prove him wrong? \[laughter\] |
**David Hernandez:** Pretty much... |
**Mat Ryer:** Actually, the serious point of just honesty about things is a big shortened when you want to be rapid as well. No one wants to hurt anyone's feelings, but if you can very quickly just have a very open and honest discussion about things, I think it does save a lot of time. |
**Jon Calhoun:** I suspect a lot of that comes with working -- like, small team helps, but partially in the sense that you're probably working with similar people every time, so you build a good relationship and trust, and when somebody says "I don't agree with this", I guess it almost feels like you don't have to worr... |
**Mat Ryer:** Yeah. Well, I've worked in places where that's how I've always been, just completely honest about it... Because that's the idea - we'll all just put the best ideas out, and we can all figure it out as a team. And a few times in my career that's hurt me, where I've just been doing that and I've been kind o... |
So yeah, I then got a bit sensitive about that, because it is important - you want to make sure that the ideas are there, but there are people that hold their ideas very personally, and will feel personally attacked if you disagree with their ideas, and stuff... So it's definitely worth watching out for, but it is nice... |
**David Hernandez:** Yeah, it's definitely about trust. I kind of trust Mat, and I can say "This looks terrible. I dislike this completely", or something like that; he doesn't get hurt. But don't forget that when you communicate in open source, when you're war mates, you try to be exactly the opposite, isn't it? You tr... |
We get so much reports -- I don't know about you, but you can feel the tone in the words, in written words, in many issues... And you know when something doesn't smell right in the tone, in the messages, when something is wrong. Trying to be honest, we try to be very clear and very respectful in that way. Every time th... |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.