Hi list, there is CASSANDRA-18773 we want to merge to 4.0 up to trunk (hence it will be in 5.0 (alpha2)) and I want to be sure we are all OK with that (especially for that 5.0 alpha release).
The patch is significantly speeding up the compaction throughput for cases when you have a lot of SSTables in a key-value table without secondary index. My colleague Cameron Zemek has identified and fixed the issue together with help of Branimir Lambov. It is a little bit hard to believe but for cases when your table contains thousands of SSTables and it does not have any 2i's, (tested on around cca 2500 SSTables), we saw the speedup of 50x (fifty times) on compaction throughput for major compactions. It is also, reportedly, affecting operations when switching from STCS to LCS. As mentioned, we plan to merge this to 4.0, 4.1, 5.0 and trunk. Any objections to that? Regards