-
Bug
-
Resolution: Unresolved
-
Medium
-
None
-
None
-
None
-
3
-
9223372036854775807
The client hashes every stripe object of every cached inode in a single
rhashtable, and that table has no upper bound. When it grows past 100.7M
objects the next resize asks for 2 GiB, which the kernel refuses; the resize
worker retries the allocation forever and the node spends its CPUs there.
Independently, the code path that reacts to a full table cannot make progress
and would spin forever, so any attempt to cap the table without also fixing
that path turns a warning storm into a hang.
- relates to failure
-
EX-16400 Loading...