Closed louwrentius closed 2 years ago
How about first grep
ing for candidate json files which at least contain fio
output keywords such as slat
, clat
, etc. These candidate files can then be checked against a JSON schema matching the full fio
output format. If they match, they can be parsed into an in-memory JSON tree.
Thanks for thinking with me! I think there is no problem to just import every json we find and maybe indeed validate them against a JSON schema. In a second step, we validate that all keys are present that we require as specified on the command line.
Thanks a lot @jsrolon, code is merged, and a new version is released on pypi.org (1.0.17)
If you want to plot charts with fio-plot, the input file names all have to adhere to a special format as documented. There is no way around this for .log files (recorded performance over time) as they don't contain any metadata.
However JSON files, by their nature and also because of the fio-plot output format, can be parsed for the variables we need, like iodepth, numjobs and the benchark type (read/write/randrw, and so on).
The key problem we would like to solve here is that from what I can tell, if files are not generated with bench-fio, mistakes are made, which in turn may lead to problems.