Separation of crawl phase from processing phase in Storm Crawler

Question

I am currently working on a Storm Crawler based project. In the current project, we modified some Bolts and Spouts of the original Storm Crawler Core artifact. For example, we changed some parts of ParserBolt or etc. In addition, we develop some processing steps in the above project. Our Bolts has been mixed with the original Storm Crawler project. For example, I have an image classifier that gives some images from the Storm Crawler and does some classification on that. Now, I am going to separate the crawl phase from the processing phase. For the crawl phase, I want to use the latest version of Storm Crawler and save its results into a Solr collection named Docs. For the second phase (which is independent of the crawl phase), I have another Storm based project which has not any relation to Storm Crawler. The input tuples of the second topology need to be feed from the Docs collection. I have not any idea about feeding documents from the Solr collection to the second storm topology. Is it a good design architecture or not? If yes, what is a good way for importing data to the second topology? It also should be noted that I want to use these projects without any downtime.

Answer 1

That is an opinion-based question but to answer it, you definitely can separate your pipeline into multiple topologies. It is a good practice when you need different types of hardware e.g. GPU for image processing vs cheaper instances for the crawl.

You could index your documents into SOLR but other solutions would also work, for instance queues etc... What you will need on the second topology is a bespoke SOLR spout. If you want the 2nd project to be independent from SC, you won't be able to leverage the code from our SOLR module but you could take it as a source of inspiration.

There might be better approaches depending on your architecture in general and whether the 2nd topology needs to ingest the content of the images. That's beyond the scope of technical questions on StackOverflow though.