loadtest

Towards automated loading test. Strategy defined here.

This package submits jobs to load data run by run in the way user did, and report loadability.

Usage

In some cvmfs conda environment:

python batch.py <str_run_mode> <bool_load_peaks> <bool_load_events>

and here is an example

For example: python batch.py sr1_bkg True True

It will load test all runs in official SR1 background run runlist.
It will test loading both peaks and events level offline data.
You can define more details for the test in configuration, especially must_have and targets.

There will be two outputs in <level>_result_folder:

<run_mode>-<level>-<datetime>-loadable.txt which is a list of runs passing a specific load test.
<run_mode>-<level>-<datetime>-err.txt in which we record all the failure traceback for runs who failed the load test.

Configuration

Two most important things to configure:

must_have: everything in the list are assumed to have existed, and if not it will directly fail the loading test.
targets: all tuples inside will be tried by get_array as targets. It should mimick how users load data.

An example here:

[general]
debug = False

[utilix]
runs_per_job = 50
max_num_submit = 200
t_sleep = 1
peaks_ram = 40000
events_ram = 5000
peaks_cpu = 1
events_cpu = 1
container = xenonnt-development.simg
peaks_log_dir = /dali/lgrandi/yuanlq/loadtest/logs
events_log_dir = /project/lgrandi/yuanlq/loadtest/logs

[context]
peaks_result_folder = /dali/lgrandi/yuanlq/loadtest/results
events_result_folder = /project/lgrandi/yuanlq/loadtest/results
peaks_storage_to_patch =
events_storage_to_patch =
peaks_output_folder = /dali/lgrandi/yuanlq/pb
events_output_folder = /project/lgrandi/yuanlq/pb

[computation]
allow_peaks_computation = False
allow_events_computation = True

[load]
must_have = {"peaks": ["peaklets", "lone_hits", "merged_s2s", "peaklet_classification", "peak_basics", "peak_positions_mlp", "peak_positions_cnn", "peak_positions_gcn"], "events": ["peak_basics", "peak_positions_mlp", "peak_positions_cnn", "peak_positions_gcn", "event_pattern_fit", "event_basics", "event_shadow", "event_ambience"]}
targets = {"peaks": [["peaks", "peak_basics", "peak_positions"]], "events": [["event_info", "cuts_basic"], ["peak_positions", "peak_basics"]]}

Some tips:

debug = True will not submit any jobs to slurm.
Typically peaks_ram = 40000 and events_ram = 5000 are good enough.
Be sure to check container is the correct one you want every time.
For all the directories, please use absolute path. Also make sure that events always go to /project and peaks always go to /dali

FaroutYLq/loadtest

loadtest

Usage

Configuration