In my view the approach in implementing LOCAL caches isn’t supposed to be highly efficient – just a functional workaround for the existing users of LOCAL cache. Moreover, if the workaround is easy but slightly awkward it isn’t a bad thing – a user needs to understand that their use case isn't directly supported and they shouldn’t expect too much of it. That’s a drawback of the existing LOCAL cache – it appears as a well-supported use case in the API, but if one actually tries to use it they’ll see lower performance and more awkward behavior than what they could expect.
Stan From: Pavel Kovalenko Sent: 25 июля 2018 г. 15:27 To: dev@ignite.apache.org Subject: Re: Deprecating LOCAL cache It's not easy to just make such caches as PARTITIONED with NodeFilter. Even in the case when a node is not affinity node for this cache we create entities like GridClientPartitionTopology for such caches on all nodes. These caches participate in the exchange, calculate affinity, etc. on all nodes. If you create 1 instance of local cache on N nodes you will get N^2 useless entities which will eat resources. So, this approach should be carefully analyzed before the proposed implementation. 2018-07-25 11:58 GMT+03:00 Dmitrii Ryabov <somefire...@gmail.com>: > +1 to make LOCAL as filtered PARTITIONED cache. I think it would be much > easier and faster than fixing all bugs. > > 2018-07-25 11:51 GMT+03:00 Dmitriy Setrakyan <dsetrak...@apache.org>: > > > I would stay away from deprecating such huge pieces as a whole LOCAL > cache. > > In retrospect, we should probably not even have LOCAL caches, but now I > am > > certain that it is used by many users. > > > > I would do one of the following, whichever one is easier: > > > > - Fix the issues found with LOCAL caches, including persistence > support > > - Implement LOCAL caches as PARTITIONED caches over the local node. In > > this case, we would have to hide any distribution-related config from > > users, like affinity function, for example. > > > > D. > > > > On Wed, Jul 25, 2018 at 9:05 AM, Valentin Kulichenko < > > valentin.kuliche...@gmail.com> wrote: > > > > > It sounds like the main drawback of LOCAL cache is that it's > implemented > > > separately and therefore has to be maintained separately. If that's the > > > only issue, why not keep LOCAL cache mode on public API, but implement > it > > > as a PARTITIONED cache with a node filter forcefully set? That's > similar > > to > > > what we do with REPLICATED caches which are actually PARTITIONED with > > > infinite number of backups. > > > > > > This way we fix the issues described by Stan and don't have to > deprecate > > > anything. > > > > > > -Val > > > > > > On Wed, Jul 25, 2018 at 12:53 AM Stanislav Lukyanov < > > > stanlukya...@gmail.com> > > > wrote: > > > > > > > Hi Igniters, > > > > > > > > I’d like to start a discussion about the deprecation of the LOCAL > > caches. > > > > > > > > LOCAL caches are an edge-case functionality > > > > I haven’t done any formal analysis, but from my experience LOCAL > caches > > > > are needed very rarely, if ever. > > > > I think most usages of LOCAL caches I’ve seen were misuses: the users > > > > actually needed a simple HashMap, or an actual PARTITIONED cache. > > > > > > > > LOCAL caches are easy to implement on top of PARTITIONED > > > > If one requires a LOCAL cache (which is itself questionable, as > > discussed > > > > above) it is quite easy to implement one on top of PARTITIONED cache. > > > > A node filter of form `node -> node.id().equals(localNodeId)` is > > enough > > > > to make the cache to be stored on the node that created it. > > > > Locality of access to the cache (i.e. making it unavailable from > other > > > > nodes) can be achieved on the application level. > > > > > > > > LOCAL caches are hard to maintain > > > > A quick look at the open issues mentioning “local cache” suggests > that > > > > this is a corner case for implementation of many Ignite features: > > > > > > > > https://issues.apache.org/jira/issues/?jql=text%20~%20% > > > 22local%20cache%22%20and%20%20project%20%3D%20IGNITE% > > > 20and%20status%20%3D%20open > > > > In particular, a recent SO question brought up the fact that LOCAL > > caches > > > > don’t support native persistence: > > > > > > > > https://stackoverflow.com/questions/51511892/how-to- > > > configure-persistent-storage-for-apache-ignite-cache > > > > Having to ask ourselves “how does it play with LOCAL caches” every > time > > > we > > > > write any code in Ignite seems way to much for the benefits we gain > > from > > > it. > > > > > > > > Proposal > > > > Let’s deprecate LOCAL caches in 2.x and remove them in 3.0. > > > > As a part of deprecation let’s do the following: > > > > - Put @Deprecated on the CacheMode.LOCAL > > > > - Print a warning every time a LOCAL cache is created > > > > - Remove all mentions of LOCAL caches from readme.io, if any, except > > for > > > > the page about cache modes > > > > - On the page about cache modes explain that LOCAL is deprecated and > > can > > > > be replaced with a PARTITIONED cache with a node filter > > > > > > > > Thanks, > > > > Stan > > > > > > > > > >
