Hacker News Reader

Comment by mattdeboard | Hacker News Reader

I am actually pretty excited about the native JSON support, and overall I am a huge fan of Postgres, but this is the most press-release-y press release ever* . By that I mean that the quotes are way too "perfect", the kind you only see in press releases. Some PR or marketing guy wrote them then showed them to the person to whom they'd be attributed to get their ok. Nothing inherently wrong with it, just struck me as funny.

* having written more than my share of press releases in my time

View on HN

> Some PR or marketing guy wrote them then showed them to the person to whom they'd be attributed to get their ok

I happen to know the guy that AFAIK was in charge of preparing the press release and he's actually a major contributor to the codebase, a geek par excellence and have been nagging people to get him quotes on the development mailing list.

The fact that apart from hacking C code he also knows how to write a catchy press release just makes him all the awesomer :)

This. Good devs who are able to better pimp their great work (such as anything in this Postgres release), are the devs that get the coolest job offers. So props to the great work by the Postgres team and their tasteful fact based self promotion. :-)

Hey man, I was PR & marketing for 6 years before I became a programmer. Valuable skillset to have.

That was some good combination. I always admire programmers than can also produce good design. Now, I will add marketer + programmer combination to my "admiration" list!

Yep, the people that can both produce and sell are a rare and very valuable breed, if only because they can actually understand what the other marketers need (or will pay for!).

I'm one of those quotes (talking about NewsBlur) and I just tried to give back to an amazing tool in my tech stack by being effusively honest. Honestly, I think they deserve more credit and was a bit surprised that they chose my quote over the dozens of undoubtedly heavier stacks that also use Postgres. Surprised in the good way.

I think Postgres doesn't do enough to toot it's own horn. Look at it's website. Come on. Spice things up a bit. I'm inclined to think hardcore users purposely downplay the product because they see it as their secret sauce.

It actually looks extremely professional. In my experience, when you're trying to convince a large company to use a platform, you want a clean, business-case-focused site without excessive flair. Postgres nails it on those points, I think.

It's reminiscent of IBM, which is a huge compliment in this context.

The last major web site update was in late 2004, according to my bisection of the wayback machine. It's a project, not a company, and there is no marketing or design budget. For example, Magnus, who does some Windows maintenance, is a principal author of the new pg_receivexlog, and knows a lot about the SSL support also worries about the web-based mailing list archive program and the CSS of the manual, if memory serves.

Nobody's holding back anything. It would be wonderful if someone wanted to contribute their design skills to the web page or documentation -- talking to the mailing list pgsql-www is probably the way to do that. I am positive such a person would receive profuse appreciation, and maybe can parlay that into personal advantage if the fulfillment of the impact such improvements would have is not quite material enough by itself.

The project runs its own web site infrastructure (many regimes have come and went in Postgres' history, only semi-recently was pgfoundary decommissioned for the purpose of new projects), and it's a little arcane -- but if someone really wanted to take ownership to move things beyond maintenance into progress, perhaps changes could be made.

> I am actually pretty excited about the native JSON support

There really isn't much to it yet, it's basically a varchar field with JSON validation, it's not like you can query or index it.

I actually find the `row_to_json` and `array_to_json` functions more interesting for now (though strangely enough there's no `hstore_to_json`)

Eh, if you want to index it you have the hstore. I do wonder why they don't deserialize to an hstore type, though.

I'm sure something like that will be added. One of the issues there facing is that hstore is a string => string map, so a conversion from json to that will be lossy (i.e. 'that "true" in hstore, is that the string "true" or a converted boolean true from json?').

What you can do right now though is use PL/V8 to query the JSON fields. Then you can use functional indexes to still being able to speed up queries. Yes. You could that before but now there's a guarantee that a field of type json contains just that, meaning that your application logic will get simpler.

> One of the issues there facing is that hstore is a string => string map, so a conversion from json to that will be lossy

Or the insertion/conversion routine can assert that the JSON object is a string:string mapping only.

Why would that be helpful? I'd want arrays, hashes, integers, etc.

You won't get them in an hstore[0] column, which was the context of my reply. Of course in a JSON column you want full JSON support, but my and my parent's post were about converting back and forth between hstore and json.

Since hstore only handles string:string mapping, the choice is either to silently corrupt data by stringifying all values in the json->hstore encoding, or erroring out if the input data is anything other than string:string.

The latter would be what I'd prefer, and more in line with usual Postgres behavior.

[0] http://www.postgresql.org/docs/9.2/static/hstore.html

JSON isn't much like hstore, as JSON can be recursive. I don't think hstore can.

Still, there's no reason why you couldn't write a function to index JSON however you like.

A hstore field can only contain strings, so it's not a one-size-fits-all indexing solution in this case.

hstore is much more limited, it's a flat mapping of strings to strings (so a single level of JSON object). And as I noted, there does not seem to be any convenience method for converting between hstore and JSON data (or the other way around)

JSON is much more powerful than hstore. Not only can it store different JSON data types (boolean, numbers) but also you can store documents with multiple levels deep, which is not possible on the hstore type.

Having said that, hstore is perfect when storing simple key/values. It's been battle tested and used for years and has some powerful and native indexing possibilities.

I think you can index JSON using a function based index, which will probably cover a lot of use-cases.

In addition, I don't know what you'd index if it wasn't through a function. Indexing JSON data as JSON? Why?

Not indexing JSON data as JSON, indexing a record based (in part) on the content of its JSON field(s) selected via e.g. an xpath-like query.

Hence using a function to extract that and then indexing the function's output. That can be done in PostgreSQL right now.

You can query and index json columns using plv8. http://lanyrd.com/coverage/?speakers=leinweber&topics=js... has more details.

As one of the main contribs noted in a previous HN article, the important part was to get JSON conversion into core, full stop - because extensions (like hstore) can't declare dependencies on other extensions. Having JSON in/out in core is an enabling feature for all the rest.

That limitation (not-lucid dependencies between contribs) is rather sucky, but jumping into the world of dependency management will require a deep breath for sure.

Hence, instead of waiting for that, JSON support in un-adorned Postgres to solve a common use case in the interceding years.

The jargon for what this gives you is "a stable oid". Also more or less equivalent to a "system OID" at this time. These "Object IDentifiers" in Postgres are unsigned int32s that are (almost?) never reused (only accrued, or removed), and are all under the number 10000, and all assigned statically by hand. A-priori knowledge of these numbers can simplify writing extensions dramatically, but clearly this is not scalable for a future with dependency chains in extensions.

If the proposed HTTP REST API (http://wiki.postgresql.org/wiki/HTTP_API) makes it into the product then the native JSON support is going to be even more useful. (I'm excited to see that there's an experimental attempt at it, too: https://github.com/bjornharrtell/postgresql-http-server)

Here is another attempt, which exposes all your SPs as a JSON-RPC API. http://search.cpan.org/~joeljac/DBIx-Pg-CallFunction-0.012/s...

Very interesting, the rest api would make psql into a total feature-for-feature replacement of mongo for us.

You're kidding right ?

PostreSQL can't horizontally scale easily or properly handle JSON right now compared to MongoDB. It also has a fixed schema which makes database migrations a necessary evil again. Sorry but the idea that PostgreSQL is a feature by feature replacement is pretty laughable.

> PostreSQL can't horizontally scale easily

Sure you can. Skype did it, Hitachi¹ did it.

¹http://www.pgcon.org/2008/schedule/events/57.en.html

> properly handle JSON right now

Well it has V8 running on it.

> It also has a fixed schema

You don't need to use it like that if you don't want to.

Easily being the appropriate word here. In that article you wrote they had to manually setup sharding. With something like CouchDB I can press a button to add a new node in the cluster. Again. Easily is what I am talking about.

And again V8 is nice and all but it is rather bolted on as opposed to something that is was built from scratch to support JSON. This shows in terms of feature support and most importantly ease of use. I can't just annotate HashMaps, Lists etc in my Java classes and have them serialized to PostgreSQL in JSON format.

I am just saying that PostgreSQL is a great and all but it is not a proper JSON document store style database anymore than hacking SQL on top of MonogDB would make it a RDBMS.

I noticed that Mattdebord's post qualified it by saying feature-for-feature replacement of Mongodb for us. He didn't claim it was for everyone. Why all the hate?

That's just taligent begin taligent.

Not sure "press the button" is the case for some. I can't speak for CouchDB but I know many people finding it not so easy for MongoDB.

You don't actually know how we use it, so you are thoroughly unqualified to criticize what I've said.

Maybe not easily but it can far more flexibly than MongoDB. Take a look at Postgres-XC which offers a two-tiered clustering solution for Postgres inspired by Teradata. A bit complex, but you don't give up your formal proof of data integrity in order to get it.

Reply on news.ycombinator.com