Open chucklever opened 7 months ago
Note that the laundromat is already run by a work queue, but more parallelism would help with NFSv4 state recovery scalability.
Break laundromat into separate jobs that can run in parallel (and thus take less wall-clock time to complete).
This was bugzilla.linux.nfs.org 384
[Chuck Lever 2022-03-09 22:12:00 UTC] Instead of setting up a special kthread to manage nfs4_client housekeeping, we can schedule some (or all?) of this work using generic work queues to help reduce duplication and complexity. In particular, clients that expire can be unhashed and then handed to a background worker to perform tasks like the cltrack upcall without impeding an nfsd thread.