Hi Alexander, Thanks for your response! I'll give it a shot. On Sat, Jan 20, 2018 at 10:22 AM Alexander Dejanovski < a...@thelastpickle.com> wrote:
> Hi Brian, > > You should definitely set unchecked_tombstone_compaction to true and set > the interval to the default of 1 day. Use a tombstone_threshold of 0.6 for > example and see how that works. > Tombstones will get purged depending on your partitioning as their > partition needs to be fully contained within a single sstable. > > Deleting the sstables by hand is theoretically possible but should be kept > as a last resort option if you're running out of space. > > Cheers, > > Le sam. 20 janv. 2018 à 15:41, Brian Spindler <brian.spind...@gmail.com> > a écrit : > >> I probably should have mentioned our setup: we’re on Cassandra version >> 2.1.15. >> >> >> On Sat, Jan 20, 2018 at 9:33 AM Brian Spindler <brian.spind...@gmail.com> >> wrote: >> >>> Hi, I have several column families using TWCS and it’s great. >>> Unfortunately we seem to have missed the great advice in Alex’s article >>> here: http://thelastpickle.com/blog/2016/12/08/TWCS-part1.html about >>> setting the appropriate aggressive tombstone settings and now we have lots >>> of timestamp overlaps and disk space to reclaim. >>> >>> >>> >>> I am trying to figure the best way out of this. Lots of the SSTables >>> with overlapping timestamps in newer SSTables have droppable tombstones at >>> like 0.895143957 or something similar, very close to 0.90 where the full >>> sstable will drop afaik. >>> >>> >>> >>> I’m thinking to do the following immediately: >>> >>> >>> >>> Set *unchecked_tombstone_compaction = true* >>> >>> Set* tombstone_compaction_interval == TTL + gc_grace_seconds* >>> >>> Set* dclocal_read_repair_chance = 0.0 (currently 0.1)* >>> >>> >>> >>> If I do this, can I expect TWCS/C* to reclaim the space from those >>> SSTables with 0.89* droppable tombstones? Or do I (can I?) manually >>> delete these files and will c* just ignore the overlapping data and treat >>> as tombstoned? >>> >>> >>> >>> What else should/could be done? >>> >>> >>> >>> Thank you in advance for your advice, >>> >>> >>> >>> *__________________________________________________* >>> >>> *Brian Spindler * >>> >>> >>> >>> >>> >> -- > ----------------- > Alexander Dejanovski > France > @alexanderdeja > > Consultant > Apache Cassandra Consulting > http://www.thelastpickle.com >
