Skip to content

Draft: Make fstrim execution a `b2c.cache_device` parameter

Martin Roukala requested to merge fstrim_rework into master

Turns out that fstrim is not something we should run continuously for the following reasons:

  • It takes up to 3 minutes, every time it gets run
  • It might degrade and brick some poor-quality SSDs (the ones re-issuing erase commands for already-trimmed areas)

This series disables it by default, and allows the user to re-enable it using b2c.cache_device=auto,fstrim=pipeline_start.

TODO: Add support for periodic fstrims, and write tests

Merge request reports