The droppable tombstone ratio continues to be very high, and thus far, strategies for user-defined compactions have failed to move the needle much. Major compactions, while not a viable long-term strategy for this problem, do however result in a significant reduction to the droppable ratio, as well as reclaiming significant disk space (and it is something we can begin doing //now//).
- [ ] eqiad
- [x] rack a
- [x] restbase1007.eqiad.wmnet
- [x] a (0.748) (0.082)
- [x] b (0.023)
- [x] c
- [x] restbase1010.eqiad.wmnet
- [x] a (0.840) (0.085)
- [x] b (0.024)
- [x] c
- [x] restbase1011.eqiad.wmnet
- [x] a (0.844) (0.079)
- [x] b (0.025)
- [x] c
- [x] rack b
- [x] restbase1008.eqiad.wmnet
- [x] a (0.837) (0.084)
- [x] b (0.025)
- [x] c
- [x] restbase1012.eqiad.wmnet
- [x] a (0.807) (0.082)
- [x] b (0.025)
- [x] c
- [x] restbase1013.eqiad.wmnet
- [x] a (0.806) (0.084)
- [x] b (0.025)
- [x] c
- [ ] rack d
- [x] restbase1009.eqiad.wmnet
- [x] a (0.818) (0.028)
- [x] b
- [x] c (0.053)
- [x] restbase1014.eqiad.wmnet
- [x] a (0.833) (0.023)
- [x] b
- [x] c (0.056)
- [ ] restbase1015.eqiad.wmnet
- [x] a (0.807) (0.026)
- [x] b
- [ ] !!c!!
- [x] codfw
- [x] rack b
- [x] restbase2001.codfw.wmnet
- [x] a
- [x] b (0.023)
- [x] c
- [x] restbase2002.codfw.wmnet
- [x] a (0.055)
- [x] b (0.023)
- [x] c (0.019)
- [x] restbase2007.codfw.wmnet
- [x] a (0.156)
- [x] b (0.022)
- [x] c (0.018)
- [ ] rack c
- [x] restbase2003.codfw.wmnet
- [x] a (0.091)
- [x] b (0.021)
- [x] c (0.017)
- [ ] restbase2004.codfw.wmnet
- [x] a (0.081)
- [x] b (0.047)
- [x] c
- [x] restbase2008.codfw.wmnet
- [x] a (0.662)
- [x] b (0.021)
- [x] c (0.018)
- [x] rack d
- [x] restbase2005.codfw.wmnet
- [x] a (0.109)
- [x] b (0.021)
- [x] c (0.037)
- [x] restbase2006.codfw.wmnet
- [x] a (0.131)
- [x] b (0.026)
- [x] c (0.017)
- [x] restbase2009.codfw.wmnet
- [x] a (0.071)
- [x] b (0.021)
- [x] c
NOTE: These compactions will all be split output (`nodetool compact -s`), resulting in 4 SSTable files (one each of 50% and 25%, and two each of 12.5%)
NOTE: In an effort to limit compaction throughput, only one compaction will be run on a host at a time