Denis, I fully support this idea.
First, looking back, I do not think it was a good design in the first place to build IGFS on top of Ignite caches. Second, I have never seen a case where IGFS provided significant performance boost. Usually it's either all data already fits buffer cache, and IGFS caching is not needed; or data does not fit buffer cache, and access pattern is close to full scan and additional caching in IGFS does not make sense. пн, 17 июн. 2019 г. в 11:28, Павлухин Иван <vololo...@gmail.com>: > Denis, > > I must say that aforementioned solutions for a Hadoop ecosystem appear > from time to time in questions on a user mailing list. So, it seems > that there is a practical need for such solutions. > > But of course it does not mean that we should continue a support of > IGFS and Hadoop Accelerator. If both are not solutions that fit well > common use cases then we should discontinue it. If any of them is very > good for it's purposes but we do not have a capacity to support it > without sacrificing main Ignite goals then we still should discontinue > it in my mind. > > P.S. Personally I am a fan of UNIX way. I like ideas of a single > responsibility and integrations. And I suppose there are other Ignite > features which could be dropped. > > ср, 12 июн. 2019 г. в 21:04, Denis Magda <dma...@apache.org>: > > > > > Igniters, > > > > I'd like us to move on and finish our conversation on the IGFS [1] and > > Hadoop Accelerator [2] support. > > > > To my knowledge, there is no single committer who maintains the > > integrations; they are no longer tested and, even more, the community > > stopped providing the binaries since Ignite 2.6.0 release (look for > > In-Memory Hadoop Accelerator table [3]). > > > > Why all of that happened? Because of a little value, something succeeds > > while something fails. Does it mean that Ignite cannot be used for Hadoop > > acceleration, in general? No, quite the opposite, it CAN be used, but a > > solution is different. Have Ignite with native persistence deployed close > > to your Hadoop cluster (replace GridGain with Ignite) [4]. > > > > So, I propose we remove IGFS and In-Memory Hadoop Accelerator from our > > master repository and rework existing public documentation showing how to > > achieve the acceleration with Ignite. > > > > Any supporters or objections? > > > > > > [1] https://apacheignite-fs.readme.io/docs/in-memory-file-system > > [2] https://apacheignite-fs.readme.io/docs/hadoop-accelerator > > [3] https://ignite.apache.org/download.cgi#binaries > > [4] > > > https://docs.gridgain.com/docs/bdb-getting-started#section-gridgain-data-lake-accelerator > > > > - > > Denis > > > > -- > Best regards, > Ivan Pavlukhin >
