def get_dfa()

in cluster-trace-gpu-v2020/analysis/utils.py [0:0]

• 16 lines of code
• 3 McCabe index (conditional complexity)

def get_dfa(dft, dfj, dfi, dfg):
    print('dft + dfj ...')
    dfa = dft.merge(dfj, on=['job_name'], suffixes = ['','_j'])
    dfa.loc[dfa.start_time==0, 'start_time'] = np.nan
    dfa.loc[dfa.start_time==0, 'end_time'] = np.nan
    dfa['runtime'] = dfa.end_time - dfa.start_time
    print('dft + dfj + dfi ...')
    dfia = get_dfia(dfi)
    dfa = dfa.merge(dfia, on=['job_name','task_name'], suffixes=['','_i'])
    dfa['duration_min'] = dfa.runtime_i / 60  # duration of instances
    dfa['wait_time'] = dfa.start_time_i - dfa.start_time # task wait time
    dfa['start_date']=dfa.start_time.apply(pd.Timestamp, unit='s', tz='Asia/Shanghai') # task start time
    # dfa = dfa[dfa.status=='Terminated']
    print('dft + dfj + dfi + dfg ...')
    dfa = dfa.merge(dfg[[x for x in dfg.columns if x != 'user']], on='inst_id', how='left')  # reserve NaN ones by how='left'
    dfa.loc[dfa.group.isnull(),'group'] = dfa.loc[dfa.group.isnull(), 'user']  # fill group==NaN ones with user
    return dfa