On Mon, 4 Aug 2014, Kenneth Waegeman wrote: > Hi, > > I have been doing some tests with rados bench write on a EC storage pool with > a writeback cache pool(replicated, size 3), and have some questions: > > * I had set target_max_bytes to 280G, and after some time of writing, the > cache pool stays filled around 250G of data, rados df output: > > pool name category KB objects clones > degraded unfound rd rd KB wr wr KB > cache - 244371669 60685 0 > 0 0 0 0 3521019 7329726464 > data - 0 0 0 > 0 0 0 0 0 0 > ecdata - 7211409408 1760598 0 > 0 0 0 0 1760598 7211409408 > metadata - 2 20 0 > 0 0 0 0 21 8 > total used 11019283916 1821303 > total avail 83093579980 > total space 94112863896 > > I thought target_max_bytes was raw bytes in the cache pool, but I suppose it > is the actual data then ? (I did not find this in the docs) . So the cache > pool will fill up to about 3*280G? Correct. > * Another question about that rados df output: There are no reads on the cache > pool, but of course there are when data gets evicted to the ecdata pool. I > guess this is just not taken into account for the rados df output? Correct. > * After the rados test has been done, rados starts to cleanup the written > data. I see that it then reads data from the ecdata pool (is this for > verification?) but it is not read from the cache: > > > pool name category KB objects clones > degraded unfound rd rd KB wr wr KB > cache - 147251372 2975965 0 > 0 0 1 1 8976363 12472766465 > data - 0 0 0 > 0 0 0 0 0 0 > ecdata - 9528655873 2326333 0 > 0 0 2919104 0 3763883 12472758273 > metadata - 2 20 0 > 0 0 0 0 21 8 > total used 16216299992 5302318 > total avail 77896563904 > total space 94112863896 The EC pool "reads" are probably the objects getting promoted just before they get deleted. It's (currently) normal, although not particularly ideal. :) sage
 |