couchdb-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Jan Lehnardt <...@apache.org>
Subject Re: [PROPOSAL] Allow rewrites to be JS function
Date Tue, 20 Oct 2015 19:09:27 GMT
What Dale and the others said.

And what I explained at length on the marketing@ mailing list to the very same people:

This project doesn’t have the capacity to get a distributed syncing database AND an application
server platform going at the same time. And the people who are doing the lion’s share of
the work here have long concluded that we need to get the database part going before we can
tackle anything else.

Like others said, CouchDB is flexible enough to let anyone build this on top of, or make a
plugin, or whatever, we can even make it an ASF sub-project, but this needs to happen separately
from what we’ve all agreed in 2012 and again in 2014 and 15 to do here, which is ship a
kick-ass database, that replicates.

The one “extra-curricular” thing we have to allow ourselves is pushing the sync protocol
forward, together with PouchDB & friends, and accommodate the new use cases that our syncing
cousins enable for our end-users.

Things like selective sync, a less chatty sync protocol for mobile users (_bulk_get/_revs
is only a small patch for a larger problem). These are the pressing issues for this project
that we have to get right before we can consider building an app server on top of it. And
we have to ship the 2.0 cluster first, too, so anyone with JavaScript chops here should join
the “The JavaScript Test Suite” thread and get hacking.

If you are interested in building an app server on top of CouchDB, do that: build it on top
of CouchDB. We are more than happy put you into the Apache family and help with naming, community,
security, distribution etc. issues, but this won’t be the core focus of this project.

And like I mentioned to you very same people on marketing@ before*,**: I’m tired of the
deliberate derailing of this project.

It stops here.

* http://markmail.org/message/cpfkxmd6i6qthcnr
** http://markmail.org/message/fns5gklnyddj6m7s

* * *

In the same email, I invited you to get your hands dirty, and you have followed that invitation
and you even got a feature in. That’s the right start, I welcome this. And now you are getting
some feedback about pursuing this road and you don’t like it, and then you start with the
derail again. Dynamic rewrites are feature with very limited use in the grander scope of things,
so I have no qualms about having this in CouchDB.

But as I outlined on numerous occasions now, the technical CouchApp architecture within CouchDB
is a _dead-end_. It has not been touched for 5+ years until the rewrites patch came in, it
has serious design flaws and abysmal performance characteristics. It is literally a cgi-bin,
which we collectively agreed at the end of the 90s is a terrible idea to serve dynamic content
from.

I’d welcome an app-server on top of CouchDB like you all want, but it won’t get any meaningful
adoption of users or support from other devs if you build it on couchjs.

I’ve long been in favour of porting our view server to Node.js. And if then every CouchDB
also has a Node.js, you can build a *fantastic*, modern, app server platform on top of *that*.
I wouldn’t even mind shipping this with CouchDB. 

But this won’t be CouchDB’s main story: A Database that replicates. A Database that has
sibling-projects it replicates with to enable even more use-cases. A Database, that defines
an open sync standard that powers the next 20 years in mobile/cluster computing. That’s
where the power of this project lies, not in another app server of which there are plenty
and of which many already work very well with CouchDB.

FWIW, I’d work full-time on porting our view server to Node.js if I didn’t realise that
shipping 2.0 and the work on the sync protocol are _way_ more important. So that’s what
I’m working on. And that’s what everyone else here is working on. We need to come together
as a team get this out, or 2.0 will never ship.


tl;dr:

- we have limited resources and we have to pick our battles, and what we need to be doing
is on a clear roadmap for the foreseeable future, and an app server is not it.

- you won’t change the main mission statement of CouchDB until after we shipped 2.0 and
3.0 and whichever other release we need to get some significant revisions to the sync protocol
into the user-base.

- if you want to build a modern app server into CouchDB, do it with a better technical foundation.
The road, at least to me, is obvious, it’s just a lot of work, like every fun endeavour.

- if you want to do this concurrently with the other efforts going on, we’d be more than
happy to establish a sub-project for you, where you can reign free, and not derail the efforts
happening here, that the majority of the people doing the work here deem more important.

* * *

One final note that I just copy-and-paste out of the aforementioned thread: before you consider
turning this into an ad-hominem attack, or some insinuation that I am abusing my PMC Chair
position to push through my personal agenda or vendetta against you and your loved ones, or
any of this sort of crap (that has come up before), keep it to yourself, thanks.

Best
Jan
--



> On 20 Oct 2015, at 11:10, Dale Harvey <dale@arandomurl.com> wrote:
> 
> This discussion has gone round and round a couple of times in different
> forms
> so I will avoid repeating my previous points but from working on PouchDB,
> the focus on having 'PouchDB' be a database only is fairly liberating, by
> not trying
> to add fairly arbitrary "application platform" features into the core
> codebase we can focus
> on integrating with what does provide those application platform features
> much better.
> 
> Users do not need to wait for reverse proxying or url rewriting to be an
> agreed, implemented
> and maintained feature, they can use the many that already exist and we
> will ensure
> that we integrate with their use case as best as we possibly can.
> 
> On 19 October 2015 at 23:57, Harald Kisch <haraldkisch@gmail.com> wrote:
> 
>> Hi Garren,
>> 
>> look at MSSQL (CLR) or ORACLE (JAVA Forms) any database was trying to
>> support their users with markup languages like XML, HTML, etc. for instance
>> directly out of the database core (performance, simplicity,
>> scalability,..).
>> Lotus Notes did also integrate JavaScript inside of their core (Do you know
>> which guy did take part of it?). This have different reasons, but one of
>> this reasons is to support users with dynamic mutable data directly into
>> their GUI in JSON format which in my opinion is the fundamental part of
>> CouchDB to be a database for the web.
>> Improvements get lost if we look at others and try not to be different. In
>> my opinion we should more think about replacing spidermonkey with the
>> google V8 engine and itegrate node completely into the CouchDB core to
>> consume npm-packages directly instead of using them in the local filesystem
>> outside of CouchDB, where unfortunatelly complexity rise up at scaling.
>> 
>> --Harald
>> 
>> On Mon, Oct 19, 2015 at 9:24 PM, Johs Ensby <johs@b2w.com> wrote:
>> 
>>> Hi Garren,
>>> thanks for the "not standing in the way", I hope for more volunteers to
>>> iron out some of CouchDB's old akward wrinkles.
>>> I am all with you for simplification:) ermouth's rewrite function is a
>>> huge simplifier.
>>> 
>>> Where I disagree with you is where you say "probably a sign that this
>> idea
>>> is not something worth pursuing".
>>> Whenever you discover that you have a differentiator, it's always a good
>>> idea to look closely before discaring it and blend in with the rest.
>>> It's all about attracting the next million web developers.
>>> 
>>> johs
>>> 
>>> 
>>> 
>>> 
>>>> On 19. okt. 2015, at 20.08, Garren Smith <gs@redcometlabs.com> wrote:
>>>> 
>>>> I’m really struggling with these proposals. I love the enthusiasm of
>>> everyone but I keep thinking we should rather simplify CouchDB.
>>>> CouchDB is ultimately a database. One with excellent sync capabilities.
>>> And combining that with libraries like PouchDB and Hoodie make it an
>>> amazing database to build applications with.
>>>> Adding routers and reverse proxies just makes it feel like we trying to
>>> push CouchDB into being more than it needs to be.
>>>> 
>>>> For example building Couchapp like functionality in Node.js is so
>> simple
>>> and way better. Languages like Go also do that really well. Far superior
>>> than what we can do with a database.
>>>> I would rather let the Node.js and Go web libraries serve content and
>>> let us focus on building a clustered replicating database. We will draw
>>> more people to this community if we can do that properly over creaky,
>> slow
>>> and limited web serving mashed on top of a database.
>>>> If I look at other popular databases, I don’t see any of them serving
>>> web content which is probably a sign that this idea is not something
>> worth
>>> pursuing.
>>>> 
>>>> However if there is a burning desire for this and developers raising
>>> their hands to code this functionality, I would not stand in your way. It
>>> is great to see the varied use of CouchDB out in the wild.
>>>> 
>>>> Cheers
>>>> Garren
>>>> 
>>>> 
>>>>> On 19 Oct 2015, at 4:47 PM, Johs. E <johs@b2w.com> wrote:
>>>>> 
>>>>> Thanks Andy,
>>>>> I will try and  get some use cases up on confluence.
>>>>> As for whoever would pick up the work after ermouth,  I have of course
>>> one big thing on the wish list that goes well with a new router
>> solution..
>>>>> reverse proxy
>>>>> I remember asking about it when I first started to work w CouchDB and
>>> there were some concerns regarding security.
>>>>> Since then I think node.js has paved the way with content scraping and
>>> all sorts of outgoing traffic.
>>>>> 
>>>>> Has anyone work on a reverse proxy solution for Couch?
>>>>> 
>>>>> johs
>>>>> 
>>>>> 
>>>>>> On 18 Oct 2015, at 21:36, Andy Wenk <andywenk@apache.org> wrote:
>>>>>> 
>>>>>> Hey Johs,
>>>>>> 
>>>>>> thanks a lot for this. I need some time to dig into it. We need a
>>> place to
>>>>>> write the user stories / use case down. So I suggest we find good
>>> place at
>>>>>> the cwiki. So I suggest to use
>>>>>> 
>>> 
>> https://cwiki.apache.org/confluence/display/COUCHDB/Enhancement+Proposals
>>> .
>>>>>> Do you have write access there? If not, please ping me.
>>>>>> 
>>>>>> Great work!
>>>>>> 
>>>>>> All the best
>>>>>> 
>>>>>> Andy
>>>>>> 
>>>>>> P.S.: Jan already mentioned the feature freeze. Please take it not
>> as a
>>>>>> demotivation but as the possibility to have a bit more time to work
>> on
>>> it.
>>>>>> 
>>>>>> On 17 October 2015 at 17:32, Johs Ensby <johs@b2w.com> wrote:
>>>>>> 
>>>>>>> Andy,
>>>>>>> I will make my first use case for function in _rewrite a high
level
>>> one:
>>>>>>> to create a standalone server that is an all-in-one DB server,
>>> application
>>>>>>> server, api server and web server.
>>>>>>> 
>>>>>>> I have played with the build of CouchDB 2 with rewrite function
>>>>>>> implemented that  ermouth put up on the irish AWS community AMI
list
>>> and
>>>>>>> the new use cases are endless.
>>>>>>> First, I find that there are a few things that people fail to
notice
>>> about
>>>>>>> ddocs.
>>>>>>> you need a tool to build a ddoc, editing JSON is not a viable
>> option.
>>> The
>>>>>>> Ddoc Lab of ermouth is in a class of its own. If you havent tried
it
>>> yet,
>>>>>>> do so from http://ddoc.me/ <http://ddoc.me/>. Installing
on your
>> own
>>>>>>> couch it is as easy as storing the application, all included
as one
>>>>>>> document in any database. Ddoc Lab is a component oriented IDE
with
>>> syntax
>>>>>>> checking, less preprosessor and other build tools that let you
keep
>> a
>>> well
>>>>>>> organized ddoc as a source project (in one couchdb document)
and you
>>>>>>> publich a ddoc to any target db.
>>>>>>> with this tool you can organize your js modules and templates
etc
>> and
>>>>>>> basically...
>>>>>>> set up the API of your application in a ddoc. You can switch
between
>>>>>>> databases and their ddoc functionality based on username, role
or
>>>>>>> geolocation and limit access to parts of the Couch API as needed
>>>>>>> 
>>>>>>> This is the method I would recommend to explore powerful simplicity
>>> with
>>>>>>> function in rewrites
>>>>>>> redirect port 80 directly to couch
>>>>>>> set up 2 vhosts, one for public access pointing to youdb/_design/api
>>> and
>>>>>>> one for sysadm pointing to /
>>>>>>> for admin use Fauxton and Ddoc Lab on the sysadm vhost
>>>>>>> you are set to develop great systems, no big tool stack to learn,
>> just
>>>>>>> bring in whatever js modules you like, the template engine you
like,
>>> the
>>>>>>> router you like, the HTML5 stuff you like..
>>>>>>> .. or just write some very compact js code in one place where
you
>>> ealier
>>>>>>> had to mess around with a whole stack of tools and systems
>>>>>>> 
>>>>>>> below is the req object that the function takes
>>>>>>> 
>>>>>>> Johs
>>>>>>> 
>>>>>>> 
>>>>>>> 
>>>>>>> The rewrite function has this syntax
>>>>>>> function(req) {
>>>>>>>     .. your code that will
>>>>>>>     return
>>>>>>>             path
>>>>>>>             // optional
>>>>>>>             headers
>>>>>>>             method // you can change this on the fly
>>>>>>>             code
>>>>>>>             body
>>>>>>> }
>>>>>>> 
>>>>>>> the function receives this req object
>>>>>>> method
>>>>>>> path
>>>>>>> raw_path
>>>>>>> query
>>>>>>> headers
>>>>>>>     Accept
>>>>>>>     Accept-Encoding
>>>>>>>     Connection
>>>>>>>     Host
>>>>>>>     Upgrade-Insecure-Requests
>>>>>>>     User-Agent
>>>>>>>     x-couchdb-vhost-path
>>>>>>> body
>>>>>>> peer
>>>>>>> cookie
>>>>>>> userCtx
>>>>>>> db
>>>>>>> name
>>>>>>> roles
>>>>>>> secObj
>>>>>>> 
>>>>>>>> On 1. okt. 2015, at 13.40, Andy Wenk <andywenk@apache.org>
wrote:
>>>>>>>> 
>>>>>>>> Johs,
>>>>>>>> 
>>>>>>>> Yes for sure! That's always great. Maybe you can also write
some
>> user
>>>>>>> stories (given when then) or scribble some graphics. Everything
is
>>> useful
>>>>>>> and will fasten the process ;-)
>>>>>>>> 
>>>>>>>> Cheers
>>>>>>>> 
>>>>>>>> Andy
>>>>>>>> 
>>>>>>>> On 1 Oct 2015 12:38, "Johs Ensby" <johs@b2w.com <mailto:
>> johs@b2w.com
>>>>> 
>>>>>>> wrote:
>>>>>>>> Thanks for this Andy,
>>>>>>>> 
>>>>>>>> I can contribute to the discussion of the feature seen from
a user
>>>>>>> perspective.
>>>>>>>> Would it be appropriate to present some use cases?
>>>>>>>> 
>>>>>>>> best
>>>>>>>> Johs
>>>>>>>> 
>>>>>>>>> On 1. okt. 2015, at 12.33, Andy Wenk <andywenk@apache.org
>> <mailto:
>>>>>>> andywenk@apache.org>> wrote:
>>>>>>>>> 
>>>>>>>>> Johs,
>>>>>>>>> 
>>>>>>>>> Let me please show the steps needed.
>>>>>>>>> 
>>>>>>>>> * discuss the feature very clearly on the dev@. Please
make sure
>>> that
>>>>>>> core
>>>>>>>>> developers as committers with commit bits are involved
>>>>>>>>> 
>>>>>>>>> * code the feature. Make sure to implement tests
>>>>>>>>> 
>>>>>>>>> * send a pull request and show it to dev@
>>>>>>>>> 
>>>>>>>>> * finally the community will accept or decline the feature
(this
>>> will
>>>>>>>>> involve refactoring and changes)
>>>>>>>>> 
>>>>>>>>> As Alex said. The PMC or Jan do not decide about the
feature.
>>>>>>>>> 
>>>>>>>>> All the best
>>>>>>>>> 
>>>>>>>>> Andy
>>>>>>>>> On 1 Oct 2015 11:21, "Alexander Shorin" <kxepal@gmail.com
>> <mailto:
>>>>>>> kxepal@gmail.com>> wrote:
>>>>>>>>> 
>>>>>>>>>> On Thu, Oct 1, 2015 at 12:07 PM, Johs Ensby <johs@b2w.com
>> <mailto:
>>>>>>> johs@b2w.com>> wrote:
>>>>>>>>>>> will you welcome ermouths rewrite contribution?
>>>>>>>>>> 
>>>>>>>>>> The decision is depends on the implementation. If
it will be
>> good,
>>> why
>>>>>>>>>> not? Finally, CouchDB is open source project: we
cannot forbid
>>> people
>>>>>>>>>> right for contributions, we only welcome them.
>>>>>>>>>> 
>>>>>>>>>>> Arguments against couchapps has to do with performance
and the
>>> folly
>>>>>>> in
>>>>>>>>>> competing with node.js.
>>>>>>>>>> 
>>>>>>>>>> Performance question for the new _rewrite implementation
is very
>>>>>>>>>> depends on query server. Once it can process this
kind of
>>> functions,
>>>>>>>>>> you may use something better than JS to gain better
performance.
>>> That
>>>>>>>>>> could be Erlang native query server, or luerl-based
one, or else
>>> you
>>>>>>>>>> like.
>>>>>>>>>> 
>>>>>>>>>> --
>>>>>>>>>> ,,,^..^,,,
>>>>>>>>>> 
>>>>>>>> 
>>>>>>> 
>>>>>>> 
>>>>>> 
>>>>>> 
>>>>>> --
>>>>>> Andy Wenk
>>>>>> Hamburg - Germany
>>>>>> RockIt!
>>>>>> 
>>>>>> GPG fingerprint: C044 8322 9E12 1483 4FEC 9452 B65D 6BE3 9ED3 9588
>>>>>> 
>>>>>> https://people.apache.org/keys/committer/andywenk.asc
>>>>> 
>>>> 
>>> 
>>> 
>> 

-- 
Professional Support for Apache CouchDB:
http://www.neighbourhood.ie/couchdb-support/


Mime
View raw message