Hi,
I have an architecture question regarding the union of more than two streams in Apache Flink. We are having three and sometime more streams that are some kind of code book with whom we have to enrich main stream. Code book streams are compacted Kafka topics. Code books are something that doesn't change so often, eg currency. Main stream is a fast event stream. Idea is to make a union of all code books and then join it with main stream and store the enrichment data as managed, keyed state (so when compact events from kafka expire I have the codebooks saved in state). The problem is that enriched data foreign keys of every code book is different. Eg. codebook_1 has foreign key id codebook_fk1, codebook_2 has foreign key codebook_fk2,…. that connects with main stream. This means I cannot use the keyBy with coProcessFunction. Is this doable with union or I should cascade a series of connect streams with main stream, eg. mainstream.conect(codebook_1) -> mainstreamWihtCodebook1.connect(codebook_2) - > mainstreamWithCodebook1AndCodebook2.connect(codebook_3) - > ….? I read somewhere that this later approach is not memory friendly. Thx. BB. |
Hi, With a.connect(b).coprocess(xx).connect(c).coprocess(xx), there would create two operators, the first operators would union a and b and output the enriched data, and then .connect(c).coprocess(xx) would pass-throught the already enriched data and enrich the record from c. Since the two operators could not get chained, the performance seems would be affected. Another method is to first label each input with a tag, e.g., ("a", a record), ("b", b record), .. and then use a.union(b).union(c).union(d).process(xx) then in the process operator, different logic could be chosen according to the tag. If adding tag is hard, then it might need to use the new multiple-inputs operator, which somehow would need to use the low-level API of Flink, thus I would recommend the above tag + union method first. Best, Yun
|
In reply to this post by B.B.
Hello BB, Just want to share you some of my immature ideas. Maybe some experts can give you better solutions and advice.
You can try to use Flink temporal table join to do the join work here. [1][2]. For such approach, you are cascade the join to enrich the mainstream. This seems to be fitting into your use case since your enrich stream doesn’t change
so often and contains something like currency. For such join, there should be some internal optimization and might get rid of some memory consumption issues, I guess? Maybe I am wrong. But it worth to take a look. Reference: [1]
https://ci.apache.org/projects/flink/flink-docs-release-1.12/dev/table/streaming/joins.html Best, Fuyao From:
B.B. <[hidden email]> Hi, BB. |
Hello BB,
Class CodebookData{ private Currency currency; private OrganizationUnit organizationUnit;
... }
Reference: [1]
https://ci.apache.org/projects/flink/flink-docs-stable/dev/stream/state/broadcast_state.html Best, Fuyao From:
B.B. <[hidden email]> Hi Fuyao, thanks for you input. I have follow up questions regarding your advices. In your DataStream suggested solution in a) case could you elaborate a little bit more. When you create that kind of generalized type how would you join it with main stream? Which key would you use. I was thinking of creating wrapper class that inside will have all the data from code books. For example Class CodebookData{ private Currency currency; private OrganizationUnit organizationUnit ... } But then I have problem which key to use to join with main stream because currency has its own key currencyId and organization unit has also its key organizationId and so on. Regarding your 2. suggested solution with Flink SQL what do you mean by “ For such join, there should be some internal optimization and might get rid of some memory consumption issues”. Thx in advance BB On Mon, 5 Apr 2021 at 07:29, Fuyao Li <[hidden email]> wrote:
-- Everybody wants to be a winner |
Free forum by Nabble | Edit this page |