Hi, if there isnt a problem, why are now 358 objects inside the cache pool running multiple times dd if=file of=/dev/zero while every full read of this 1.5 GB file produces around 8 objects inside the cache pool.
Its the same file, read again and again. But at no point its read from the cache. Its always read from the cold pool. -- Mit freundlichen Gruessen / Best regards Oliver Dzombic IP-Interactive mailto:i...@ip-interactive.de Anschrift: IP Interactive UG ( haftungsbeschraenkt ) Zum Sonnenberg 1-3 63571 Gelnhausen HRB 93402 beim Amtsgericht Hanau Geschäftsführung: Oliver Dzombic Steuer Nr.: 35 236 3622 1 UST ID: DE274086107 Am 14.06.2016 um 02:25 schrieb Christian Balzer: > > Hello, > > On Tue, 14 Jun 2016 01:57:49 +0200 Oliver Dzombic wrote: > >> Hi, >> >> yeah, for sure you are right ! >> >> Sorry... >> >> Its Centos 7 ( default kernel 3.10 ) >> ceph version 10.2.1 (3a66dd4f30852819c1bdaa8ec23c795d4ad77269) >> >> ---- >> >> In the very end i want/need a setup where all is going to the cache tier >> ( read and write ). >> >> Writes shall be, after some time without modification, written to the >> cold pool. >> > That's not how cache tiering works (by default). > See * below. > >> Read shall be, as long as they are frequently red, remain on the cache >> pool. >> >> To me, writeback mode should do that. But somehow does not. >> >> Since my thread opening mail 1,5 houres has been passed by. >> >> And its still not evicted. So i still have this objects inside the cache >> pool, while after 3600 seconds they should be gone. >> >> So there is some major problem here, as it seems to me. >> > No there isn't. > You did read my "Cache tier operation clarifications" thread a few months > back, time to re-read it. > > Flushes (writes to the cold pool) happen when the dirty ratios are > reached, never before that. > > Evictions are just zero'ing clean (previously flushed) objects when the > full ratio is reached. > > * > You might be able to achieve that by setting cache_target_dirty_ratio to > 0, but I don't see the advantage in having objects flushed half an hour > later by default. > If your cache pool fails, you'll still be stuck with tons of broken PGs, a > dead pool for all intents and purposes. > While running the cache tier "normally" may allow you to do flushes during > off-peak hours if you cache pool is large enough. > > Christian > _______________________________________________ ceph-users mailing list ceph-users@lists.ceph.com http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com
