Sergi, Dima, In the scope of Ignite 1.x it is perfectly fine to have "schema = cache". Nobody suffers from it because nobody use Ignite as database. But in future, thanks to page memory, we are going to target real database use cases. Users will have multiple tables in Ignite. Plus views, triggers, constraints, etc.. All these features are very useful and easy to implement provided that we already have table and index implementations. And in databases all related objects are *logically *grouped in a "schema". This is convenient for users: less boilerplate in SQL, better manageability (remember that database users will definitely need some console and/or UI tools to manage Ignite as a database).
What you offer is to group database objects *physically *rather than logically. It will lead to: - Boilerplate in queries - Inconvenient database management. All the things database users are used to - import/export tools, UIs, "USING" keyword, etc, will look weird in Ignite as there will be no way to group arbitrary objects logically. With this approach almost every user will have to use two schemes instead of one - one for operational data (PARTITIONED) and one for reference data (REPLICATED). No conventional database works this way. Vladimir. On Fri, Jan 13, 2017 at 9:18 PM, Dmitriy Setrakyan <dsetrak...@apache.org> wrote: > Vova, > > I will join Sergi here. It seems like "schema = cache" will take care of > all different configuration properties required for different groups of > caches. In addition, it cleanly maps into current Ignite architecture. We > will need to have a very strong reason to move away from it. > > D. > > On Fri, Jan 13, 2017 at 2:39 AM, Vladimir Ozerov <voze...@gridgain.com> > wrote: > > > Correct, it worked, because Ignite has never had real database use case > in > > mind. Unfortunately, if our global plans go as expected, it will not work > > for Ignite 2.x+. > > > > On Fri, Jan 13, 2017 at 11:53 AM, Sergi Vladykin < > sergi.vlady...@gmail.com > > > > > wrote: > > > > > Lets move on with SQL schema == Ignite cache. It worked always like > > this, I > > > see no reasons to change this. > > > > > > Sergi > > > > > > 2017-01-13 11:20 GMT+03:00 Vladimir Ozerov <voze...@gridgain.com>: > > > > > > > "Tablespace" (Oracle, PostgreSQL) is what maps better than "schema" > to > > > our > > > > cache. But not ideally still. > > > > > > > > On Fri, Jan 13, 2017 at 11:10 AM, Vladimir Ozerov < > > voze...@gridgain.com> > > > > wrote: > > > > > > > > > Alex, > > > > > > > > > > Currently Ignite is not used as database. It is used as search > > engine - > > > > > several types, several tables, several joins. This is why having > > > "SCHEMA > > > > == > > > > > cache" was never a problem. Users have never build complex SQL > > > > applications > > > > > on top of Ignite. But we are going towards database. And my > question > > > > stands > > > > > still - suppose it is Y2019, how is user going to migrate his > > database > > > > > containing 20-30-50-100 tables in a single schema in Oracle to > > Ignite? > > > > > > > > > > Single cache for all tables? Doens't work - not flexible. Users > will > > > > > definitely require different cache modes, different co-location > > rules, > > > > > different number of backups, etc.. > > > > > Schema per table? Doesn't work either - unmanageable and not > > convenient > > > > > for users even for relatively small databases. > > > > > > > > > > From user perspective schema is logical grouping of database > objects, > > > > > nothing more. > > > > > > > > > > For Ignite schema could be a logical group of resources (nodes, > > memory > > > > > pools, caches, etc.). And multiple tables over multiple caches > should > > > > > reside in it. To the contrast, table definition governs how data is > > > > stored. > > > > > This is similar to, for example, MySQL approach, where you define > how > > > you > > > > > store data on per-table level, and on schema level you define only > > > minor > > > > > things like collation. > > > > > > > > > > Vladimir. > > > > > > > > > > > > > > > On Fri, Jan 13, 2017 at 10:33 AM, Alexander Paschenko < > > > > > alexander.a.pasche...@gmail.com> wrote: > > > > > > > > > >> Vova, > > > > >> > > > > >> 2017-01-13 4:56 GMT+08:00 Vladimir Ozerov <voze...@gridgain.com>: > > > > >> > I am not quite sure I understand the idea of "SCHEMA == cache". > > > > Consider > > > > >> > some small database with, say, ~30 tables. And user wants to > > migrate > > > > to > > > > >> > Ignite. How is he supposed to do so? 30 schemas leading to > rewrite > > > of > > > > >> all > > > > >> > his SQL scripts? Or 30 key-value pairs in a single cache leading > > to > > > > >> lack of > > > > >> > flexibility and performance problems? > > > > >> > > > > >> But currently schema *is* semantically equal to cache while table > is > > > > >> equal to type descriptor (i.e. type of stored entities), nothing > new > > > > >> here. > > > > >> > > > > >> Say, in single cache we may have entities of types Person and > > > > >> Organization, those map to two tables with same names, and can be > > > > >> accessed within the same cache (i.e. schema). > > > > >> > > > > >> If we want to limit the user with having single type descriptor > per > > > > >> cache (i.e. cache has only one type of stored entities - BTW, > where > > we > > > > >> are with this 2.0-wise?), then this notion could change. But > > currently > > > > >> what has been suggested already fits quite good with what we do > have > > > > >> at the moment regarding semantic of SQL objects. > > > > >> > > > > >> - Alex > > > > >> > > > > >> > Another example is how to deal with referene tables? Lots > database > > > has > > > > >> > small reference tables which is best to fit REPLICATED cache, > > while > > > > >> others > > > > >> > are usually bound to PARTITIONED mode. "SCHEMA == cache" will > > force > > > > >> users > > > > >> > to split them into separate schemes leading to poor user > > experience. > > > > >> > > > > > >> > I understand that we may have some implementation details around > > it > > > at > > > > >> the > > > > >> > moment. But from user perspective "SCHEMA == cache" doesn't make > > > > sense. > > > > >> As > > > > >> > we are going towards AI 2.0 we'd better to rethink this > approach. > > > > >> > > > > > >> > On Thu, Jan 12, 2017 at 11:46 PM, Denis Magda < > dma...@apache.org> > > > > >> wrote: > > > > >> > > > > > >> >> > > > > >> >> > On Jan 12, 2017, at 12:35 PM, Dmitriy Setrakyan < > > > > >> dsetrak...@apache.org> > > > > >> >> wrote: > > > > >> >> > > > > > >> >> > On Thu, Jan 12, 2017 at 9:47 AM, Sergi Vladykin < > > > > >> >> sergi.vlady...@gmail.com> > > > > >> >> > wrote: > > > > >> >> > > > > > >> >> >> The xml config was only for example. We can put in this > > > > >> configuration > > > > >> >> >> string cache config parameters directly like this: > > > > >> >> >> > > > > >> >> >> CREATE SCHEMA "MyCacheName" WITH > > > > >> >> >> "cacheMode=REPLICATED;atomicityMode=ATOMIC" > > > > >> >> >> > > > > >> >> > > > > > >> >> > This approach makes sense, if it can be easily supported with > > H2. > > > > >> >> > > > > >> >> What’s for affinity keys? Can we make an exception for them by > > > > >> defining in > > > > >> >> this part of the statement > > > > >> >> > > > > >> >> CREATE TABLE employee ( > > > > >> >> id BIGINT PRIMARY KEY, > > > > >> >> dept_id BIGINT AFFINITY KEY, > > > > >> >> name VARCHAR(128), > > > > >> >> ); > > > > >> >> > > > > >> >> or that l > > > > >> >> > > > > >> >> CREATE TABLE employee ( > > > > >> >> id BIGINT PRIMARY KEY, > > > > >> >> dept_id BIGINT, > > > > >> >> name VARCHAR(128), > > > > >> >> CONSTRAINT affKey AFFINITY KEY(dept_id) > > > > >> >> ); > > > > >> >> > > > > >> >> ? > > > > >> >> > > > > >> >> — > > > > >> >> Denis > > > > >> >> > > > > >> >> > > > > >> > > > > > > > > > > > > > > > > > > > >
