Should we just kill this option entirely? And/or also kill mds_cache_memory_limit, in favor of mds_memory_target? We could probably do either/both in one release if we make the mon translate/create the newer options out of the old ones during upgrade... s
On the OSD side I kept the bluestore cache settings around just in case folks wanted to disable autotuning and force a manual configuration. I wouldn't be surprised if we eventually take them out though once the double-cache avoidance PR merges (after sharding) followed by cache age-binning. I'm hoping that will resolve any remaining reasons to do manual tuning. On the MDS side I'm not sure there's much reason to keep the old settings, though I'll defer to Patrick. If we do keep the old settings let's try to keep the conventions the same across daemons. Mark On 9/17/19 9:43 AM, Sage Weil wrote:
Should we just kill this option entirely? And/or also kill mds_cache_memory_limit, in favor of mds_memory_target? We could probably do either/both in one release if we make the mon translate/create the newer options out of the old ones during upgrade...
s _______________________________________________ Dev mailing list -- dev@ceph.io To unsubscribe send an email to dev-leave@ceph.io
On Tue, Sep 17, 2019 at 7:43 AM Sage Weil <sweil@redhat.com> wrote:
Should we just kill this option entirely?
I agree with removing mds_cache_size.
And/or also kill mds_cache_memory_limit, in favor of mds_memory_target?
I agree with Mark that it could be conceivably be useful to turn off autotuning in which case this option will remain necessary. https://tracker.ceph.com/issues/41951 -- Patrick Donnelly, Ph.D. He / Him / His Senior Software Engineer Red Hat Sunnyvale, CA GPG: 19F28A586F808C2402351B93C3301A3E258DD79D
participants (3)
-
Mark Nelson
-
Patrick Donnelly
-
Sage Weil