mds: cache limits should be expressed in memory usage, not inode count
Cached inode count (mds_cache_size) is an imperfect limit for what we really want. We frequently have to tell users/customers to tune this based on the amount of memory they want to consume for the MDS which is unacceptable.
Instead, use mempools with the cached objects to track memory usage and limit based on that.
- Assignee set to Patrick Donnelly
See #4504 and associated MemoryModel tickets.
- Related to Feature #4504: mds: trim based on total memory usage added
- Status changed from New to In Progress
- Priority changed from Immediate to Urgent
- Status changed from In Progress to Fix Under Review
- Target version set to v12.2.1
- Backport set to luminous
- Status changed from Fix Under Review to Pending Backport
- Copied to Backport #21384: luminous: mds: cache limits should be expressed in memory usage, not inode count added
- Related to Bug #21402: mds: move remaining containers in CDentry/CDir/CInode to mempool added
- Status changed from Pending Backport to Resolved
- Related to Feature #36663: mds: adjust cache memory limit automatically via target that tracks RSS added
Also available in: Atom