-
Notifications
You must be signed in to change notification settings - Fork 107
Description
Karl Cronburg wrote fiologparser_hist.py to postprocess latency histogram logs collected by fio --client so that we could get latency percentile variation over time for an entire cluster, but he wrote it with a dependency on python-pandas, which he thought greatly improved the efficiency and simplicity of the code. When I go to install pbench-fio RPM that contains it, pbench-fio RPM fails because of this dependency. When I install python pandas from EPEL, it pulls in 75 RPMs which include everything from X-windows to tex to you-name-it.
I propose that we remove the dependency on python-pandas and simply put a test into fiologparser_hist.py such that if python-pandas is not installed (i.e. the "import pandas" fails, then it tells you to install it. This tool is not intended to be run on every workload generator host, just on the one that fio --client runs on.
Do people agree with this approach? If not, what approach should we take? This change would have to be pushed upstream to fio repo.