b9b10eb736

Nobody's asking for mainline submissions though. Just publishing the drivers source code under a FLOSS licence when they stop supporting it would be enough to let the community take over the maintenance.

While I'd agree with the general stance of avoiding MongoDB for any new project, I find the statement that postgres always gives better and faster solution dubious. They don't really solve the same problems. If you happen to really need horizontal scaling, actual HA (not failovers) or documents with many field-level atomic operations, MongoDB might still be a better fit than postgres. For on-premise hosting, not having any HA out-of-the-box can be a major painpoint.

I works great (with the usual requirement of having the expected kernel options/modules enabled). The main issue right now is the lack of image support for RISC-V. I hope this is going to change if hardware starts becoming accessible.

Thanks for the clarification, you're completely right. I didn't realize the SG2042 used the same core. So, Scaleway's offer really does seem to make sense!

Fully agreed.

The JH7110 is usually about 10% faster on system level real world tasks e.g. building software.

Pretty sure the U74 CPU from the JH7110 is way worse than the C910 from the TH1520 on pretty much all aspects. So my guess is that your metric is mostly explained by the fact that the JH7110 has a PCIe bus which allows plugging in an SSD rather than an eMMC or SD Card. But such SSD also has a cost. I think this gives some perspectives.

Still way better than a VisionFive 2, and more than enough to get your hands in the thing (especially with the hourly pricing). You have to acknowledge that the RISC-V ecosystem isn't as mature as that of ARM64 and AMD64. You don't get the best performance/price ratio with RISC-V (yet). Many extensions were ratified so recently that hardware doesn't even exist (yet).

As far as I'm concerned, the LicheePi 4A from Sipeed which is also equipped with a TH1520 offers a decent value for a RISC-V SBC. If you want more performance than that, it's going to cost you a lot more.

According to my tests, it's billed by the hour with no minimum commitment. It's attractive indeed for transient workloads. €15.99 for one month commitment sounds okay-ish when comparing with competitor's offering at this price (but you don't get RISC-V servers there (yet)).

I totally upvote and agree with your comment. Thanks a lot for working on JIT, that's invaluable. We (as a community) will never thank the PG contributors enough for their work. I'd love to contribute to the postgres project, though I couldn't make it at work at the moment unfortunately. And the cost of entry is quite expensive for what I can give in my free time. I'm very sorry about this. I know well that it's always easier to complain about things rather than fixing them. Doesn't mean there's absolutely no content in it. I hope my comment did not hurt you, it was not my intention.

People are smart enough to install the software and the plugins they need. If only .1% of the users use a feature, it doesn't have to be maintained by the core contributors (and slow down the other 99.9% users from getting important features and bugfixes). It's a tradeoff and FWIW, while I think postgres is great, I also reckon the tradeoff could be improved.

Thanks for replying. Honestly, I expected a more closed and rude answer because I know my opinion on this is quite unpopular. The context of my answer: I love PostgreSQL, I've been using it in production, day-to-day, at work and for personal projects. Still it happens that I have also used more modern databases and they fixed some flaws postgres has (while having alot of other flaws that's not the point). I was answering to the question "what would make anybody choose another database given that Postgres can do pretty much anything?".

This is something that's actively being worked on. It's also useful to look at the systems that are more easily distributable, and compare other features they posses with Postgres. Aphyr has provided a wonderful service with Jepsen showing just how difficult correctness in general (with and without the added complexity of distributability) actually is.

Thanks. I'll definitely have a look into this. I'm definitely not saying that everything would be easy and ideal to implement in such a robust and mature software (I am even sure it would be a nightmare for the developers to develop and stabilize and that for this reason it's unlikely to ever happen). Other databases provides this out of the box, postgres just don't, and that's a good reason "why would anyone use anything else". Still, not a requirement for everybody, but more frequent than JIT IMO.

Do you mean something by this beyond as an aspect of distributability?

I mean that physical replication is great, but you always get into troubles when it comes to failover. Pretty much any modern DB system can be put in front of a basic HTTP LB, and replicate data on n nodes. It's indeed tightly coupled with distribution, but I meant distribution in term of performance and sharding (horizontal scalability) and resiliency in term of reliability. Relying on DNS for failover induces a downtime, relying on a proxy induces a hack. I'm wondering why it would be so hard to fix this in postgres though.

The story here has been much improved, both with pg_upgrade and though features such as logical replication.

This is definitely true. While I think logical replication could/should go even further to really fix the upgrade issue, PG 10 is a huge move in the right direction. Thanks for pointing that out.

"Schemas remains at postgres core" Would you elaborate what you mean by this?

Maybe it's more a general feeling than something really concrete, but for instance, would you create a table with only one JSONB column. It would feel weird, like abusing/hacking around the schema. From the user perspective, it's impractical, not natural, and it doesn't look like a document store at all (as we see in nosql DBs). In this respect, I don't think I will ever use postgres as a document-only database because I feel like it's just not how postgres is built (unless I'm missing some great feature?). That's a subject were document-only database may have the point.

In terms of binary size? Feature set? Actual on-disk data representation? If so, what are you comparing it to?

I'm only speaking in term of feature set. A lot of documentation comes with them. A lot of side-effects. A lot of wrong expectations from the user. A lot of complexity. A lot of maintenance burden which makes postgres more complex to evolve on huge topics.

Also, this is a common problem across the most used relational DBMS (MySQL, Oracle, SQL Server and Postgres). To give another perspective, ElasticSearch dropped a massive amount of features in version 6 just so that they can focus on the core things. I personally think it was a good move.

Having used Postgres over the years, I can certainly say that things have become much easier. What in particular are you thinking of?

I have also used PG for years and while I reckon you are completely right (a lot of improvements have been made in a tremendous amount of places), I was pointing out that the core topic I listed were still such a pain to handle, while so important since day 1.

I also struggle to square what appears to be complaints of capability while also expressing frustration with "postgres bloat and overall complexity of the system." As with all things, there are tradeoffs and resource limitations. What would you in particular are you looking for?

Ok, let me fix this: I'm not complaining at all, really really not. PG guys are doing a fantastic job, PG is a great software, no doubt about it. I would even go further: in my opinion postgres is the currently the best mature RDBMS in the place. I am just saying that postgres is not and probably never will be the answer to all the DB use-cases in the world. More than that: postgres is even failing on very basic stuff needed in almost every modern app (mostly because it was started at a time were HA was not really a concern). Put another way: some modern DBs are solving some modern problems better than postgres and might disqualify postgres for some of these use-cases. I did not meant anything else, believe me, and I apologize sincerely to all the developers that have worked on the project if my comment was taken like that.

"nobody needs" is to take in the unix philosophy sense here. If you take it first degree, you're part of the reason why PG has been missing these essential points for so long (IMHO). In comparison to the items I've listed, those features are nice to have but definitely not used by the majority of users. And I do use these advanced features (nobody needs) in PostgreSQL. I use them extensively and really love them. It's just that I could workaround them easily and properly. That's just not the case for the things I've listed.

From the top of my head, I can list at least four reasons to actually use anything else:

- Not distributable

- Poor resiliency

- Hard to upgrade

- Schemas remains at postgres core

Every solution to address any of these points in PG are hacks (bucardo, londist, slony, pgbouncer, things relying on triggers or proxies). Quite honestly, PG contributors are doing an impressive job. PG is arguably the best relational database in the market as of today and still improving very fast. However it's also one of the most (if not the most) bloated. And despite all these features nobody needs and after all these years, it's still lacking some very basic yet essential items anyone would ask for in any modern application. It feels like things haven't changed that much since the 90's. DBAs have to hack around things and root issues never gets addressed. Implementing JIT optimizations looks like fun. It will undoubtedly improve some queries by 2x, 4x or even 10x, but it won't be a game changer anyways. It'll just increase the postgres bloat and the overall complexity of the system. Meanwhile, we'll still be lacking essential things that would make PG suitable for pretty much any use-case.