Table of Contents
Scope/Description
This article discusses the key garbage collection tunables to play with with large delete workloads on a RGW Ceph cluster
Prerequisites
- Ceph cluster with RGW(s) deployed
Steps
The tunables for garbage collection below. Here is description of each
For heavy delete RGW workloads it is recommended to start with the following.
rgw_gc_max_concurrent_io = 20 rgw_gc_max_trim_chunk = 64
Further tunables to consider are below
- Increase the amount concurrent io the cluster will spend on gc requests. (rgw_gc_max_concurrent_io)
- Decrease the amount of time rgw will wait before purging an object (rgw_gc_obj_min_wait)
- Decrease the amount of a RGW will hold a lease on the data to gc’d (rgw_gc_processor_max_time)
- Decrease the amount of time between the start of consecutive garbage collector threads (rgw_gc_processor_period)
- Increase the # of objects the gc data is spread across (rgw_gc_max_objs)
Verification
Troubleshooting
Views: 1240