keith-turner commented on pull request #2096: URL: https://github.com/apache/accumulo/pull/2096#issuecomment-843314472
Thinking through plugin interfaces and alternative designs, I thought of a very different way compactions could work in Accumulo. We could have a process (like the manager or compaction coordinator) that continually scans the files in the metadata table looking for compaction work that needs to be done per tablet and gives that to external compaction processes. Under this model tablet server would have no compaction code at all and there would only be external compactions. The tserver would only need a mechanism to replace a set of files in a tablet to support this model. I think this is an interesting model to think about, but I am not advocating for it. It has pluses and minuses over the current model. The reason I am sharing it is that I think it informs the general discussion about API and SPI interfaces around compaction and supporting alternative designs. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: [email protected]
