Ah, really good question :) I believe it is stored locally on the monitor host. Saving the cluster map into RADOS would result in a chicken or egg problem.
This is supported by the following two sections in the docs: 1. https://docs.ceph.com/docs/master/rados/configuration/mon-config-ref/#background 2. https://docs.ceph.com/docs/master/rados/configuration/mon-config-ref/#data I can confirm that on my testing cephadm installation (octopus release) the `/var/lib/ceph/<cluster_id>/mon.<hostname> dir contains different timestamps on different monitors (it was flushed to disk at different times). G. On Tue, Aug 4, 2020, at 1:34 PM, Edward kalk wrote: > Thank you Gregor for the reply. I have read that page. It does say what a > Crush map is and how it’s used by monitors and OSDs, but does not say how or > where the map is stored in the system. Is it replicated on all OSD, vis a > distributed hidden pool? Is it stored on the local linux disk of the host > operating system on which the monitor daemons run? > -Ed > > -Ed > >> On Aug 3, 2020, at 4:43 PM, Gregor Krmelj <[email protected]> wrote: >> >> The CRUSH map makes up the so called "Cluster Map", for which the Ceph >> Monitors maintain a master copy. This is precisely why you have multiple >> monitors - for high availability in case a monitor goes down. >> >> This is all explained quite well in the architecture documentation: >> https://docs.ceph.com/docs/master/architecture/#cluster-map. >> >> Regards, >> G. >> >> On Mon, Aug 3, 2020, at 2:01 PM, Edward kalk wrote: >>> The metadata that tells CEPH where all data is located, my understanding is >>> the crush map. Where is it stored, is it redundantly distributed so as to >>> protect from node failure? What safeguards the critical cluster metadata? >>> >>> -Ed >>> _______________________________________________ >>> ceph-users mailing list -- [email protected] >>> To unsubscribe send an email to [email protected] >>> >> >> _______________________________________________ ceph-users mailing list -- [email protected] To unsubscribe send an email to [email protected]
