Skip to main content

JuiceFS Performance Statistics Watcher

JuiceFS pre-defined a lot of monitoring items to show internal performance statistics when the system is running. These items are exposed by Prometheus API. However, when diagnosing performace issues, users may want a real-time monitorig tool to know what is actually going on within a certain time. Thus, the stats command is developed to display selected items every second, similar to the Linux tool dstat. The output is like:

stats_watcher

By default, this command will monitor the JuiceFS process corresponding to the specified mount point, showing the following items:

usage

  • cpu:CPU usage of the process
  • mem:physical memory used by the process
  • buf:current buffer size of JuiceFS; its limit can be adjusted by the mount argument --buffer-size

fuse

  • ops/lat:number of operations handled by FUSE per second, and the average latency of them
  • read/write:read/write bandwidth handled by FUSE

meta

  • ops/lat:number of metadata operations and the average latency of them; please note operations returned directly in cache are not counted, so that the result is closer to real performance of metadata engines

blockcache

  • read/write:read/write bandwidth of client local data cache

object

  • get/put:Get/Put bandwidth between client and object storage

Moreover, users can acquire verbose statistics (like read/write ops and the average latency) by setting --verbosity 1, or customize displayed items by changing --schema. For more information, please check juicefs stats -h.