You could try repartitioning your RDD using a custom partitioner (HashPartitioner etc) and caching the dataset into memory to speedup the joins.
Thanks Best Regards On Tue, Apr 14, 2015 at 8:10 PM, Wang, Ningjun (LNG-NPV) < ningjun.w...@lexisnexis.com> wrote: > I have an RDD that contains millions of Document objects. Each document > has an unique Id that is a string. I need to find the documents by ids > quickly. Currently I used RDD join as follow > > > > First I save the RDD as object file > > > > allDocs : RDD[Document] = getDocs() // this RDD contains 7 million > Document objects > > allDocs.saveAsObjectFile(“/temp/allDocs.obj”) > > > > Then I wrote a function to find documents by Ids > > > > def findDocumentsByIds(docids: RDD[String]) = { > > // docids contains less than 100 item > > *val *allDocs : RDD[Document] =sc.*objectFile*[Document]( > (“/temp/allDocs.obj”) > > *val *idAndDocs = allDocs.keyBy(d => dv.id) > > docids.map(id => (id,id)).join(idAndDocs).map(t => t._2._2) > > } > > > > I found that this is very slow. I suspect it scan the entire 7 million > Document objects in “/temp/allDocs.obj” sequentially to find the desired > document. > > > > Is there any efficient way to do this? > > > > One option I am thinking is that instead of storing the RDD[Document] as > object file, I store each document in a separate file with filename equal > to the docid. This way I can find a document quickly by docid. However this > means I need to save the RDD to 7 million small file which will take a very > long time to save and may cause IO problems with so many small files. > > > > Is there any other way? > > > > > > > > Ningjun >