New metric: `ceph_crash_reports` which counts the entries returned by
`ceph crash ls` by daemon name and archival status.
This is not the same as `ceph_new_crash_reports` which is the value of
the `RECENT_CRASH` health check, and that only counts the non-archived
errors of the past two weeks. The new metric counts errors as long as
they are not purged (which is done after 1 year by defaults).
In TestClusterHealthCollector, test all supported versions by default,
and split the osdmap tests for Nautilus vs. Octopus+. There were a
number of tests that included an osdmap that didn't need it, and the
osdmap was removed from them so that version-specific testing would not
be required.
Misplaced ratio equals to misplaced_objects deviding misplaced_total, not misplaced_objects / num_objects.
So add a separate metric to show misplaced ratio.
Signed-off-by: haoyixing <haoyixing@kuaishou.com>
- `ceph.Conn` interface no longer depends on go-ceph/rados,
now defines its own `PoolStat` structure for our use.
- New separate `rados` package that implements the interface
- Merged `mocks` package into `ceph` to avoid circular import
