How to use print_job_result method in autotest

Best Python code snippet using autotest_python

ts_2019_ncov_dataprep_us.py

Source:ts_2019_ncov_dataprep_us.py Github

copy

Full Screen

...11loader = Loader(client=client)12extractor = Extractor(client=client)13geo='us'14#### Functions ####15def print_job_result(job, client, max_results=20):16 job.result()17 bq_table = client.get_table(job.destination)18 df = client.list_rows(bq_table, max_results=max_results).to_dataframe()19 print(bq_table.full_table_id)20 print(df.head(max_results))21#### Load Data ####22confirmed = pd.read_csv('https://raw.githubusercontent.com/CSSEGISandData/COVID-19/master/csse_covid_19_data/csse_covid_19_time_series/time_series_covid19_confirmed_US.csv')23deaths = pd.read_csv('https://raw.githubusercontent.com/CSSEGISandData/COVID-19/master/csse_covid_19_data/csse_covid_19_time_series/time_series_covid19_deaths_US.csv')24dt_cols = [datetime.datetime.strptime(dt, '%m/%d/%y').strftime('dt_%Y%m%d') for dt in list(confirmed.columns)[11:]]25# confirmed26confirmed_cols = ['UID', 'iso2', 'iso3', 'code3', 'FIPS', 'county', 'province_state', 'country_region', 'latitude', 'longitude', 'combined_key'] + dt_cols27confirmed.columns = confirmed_cols28confirmed_schema = [29 ('UID', 'INT64'),30 ('iso2', 'STRING'),31 ('iso3', 'STRING'),32 ('code3', 'INT64'),33 ('FIPS', 'INT64'),34 ('county', 'STRING'),35 ('province_state', 'STRING'),36 ('country_region', 'STRING'),37 ('latitude', 'FLOAT64'),38 ('longitude', 'FLOAT64'),39 ('combined_key', 'STRING'),40] + [(dt_col, 'INT64') for dt_col in dt_cols]41loader.load_df(confirmed, 'stanleysfang.surveillance_2019_ncov.ts_2019_ncov_{geo}_confirmed_raw'.format(geo=geo), schema=confirmed_schema)42# deaths43deaths_cols = confirmed_cols44deaths_cols.insert(11, 'population')45deaths.columns = deaths_cols46deaths_schema = confirmed_schema47deaths_schema.insert(11, ('population', 'INT64'))48loader.load_df(deaths, 'stanleysfang.surveillance_2019_ncov.ts_2019_ncov_{geo}_deaths_raw'.format(geo=geo), schema=deaths_schema)49for job in loader.job_history:50 print_job_result(job, client)51#### Dataprep ####52# restructure raw data53for metric in ['confirmed', 'deaths']:54 array_query = ''55 for dt_col in dt_cols:56 array_query = \57 """{array_query}58 STRUCT(PARSE_DATE('dt_%Y%m%d', '{dt_col}') AS dt, {dt_col} AS total_{metric}),59 """.format(array_query=array_query, dt_col=dt_col, metric=metric)60 array_query = re.sub('[,\n ]*$', '\n', array_query)61 62 query = \63 """64 SELECT dt, a.* EXCEPT(arr), total_{metric}65 FROM (66 SELECT67 county, province_state, country_region, combined_key, latitude, longitude,68 [{array_query}] AS arr69 FROM `stanleysfang.surveillance_2019_ncov.ts_2019_ncov_{geo}_{metric}_raw`70 ) a, UNNEST(arr)71 """.format(metric=metric, array_query=array_query, geo=geo)72 73 query_job = qr.run_query(query, destination_table='stanleysfang.surveillance_2019_ncov.ts_2019_ncov_{geo}_{metric}'.format(geo=geo, metric=metric), time_partitioning=True, partition_field='dt')74for job in qr.job_history:75 print_job_result(job, client)76# agg data77ts_2019_ncov_query = \78"""79SELECT80 dt, county, province_state, country_region, combined_key, latitude, longitude,81 population,82 total_confirmed, total_deaths, daily_new_confirmed, daily_new_deaths,83 ROUND(AVG(daily_new_confirmed) OVER(PARTITION BY combined_key ORDER BY dt ROWS BETWEEN 6 PRECEDING AND CURRENT ROW), 1) AS daily_new_confirmed_7d_ma,84 ROUND(AVG(daily_new_deaths) OVER(PARTITION BY combined_key ORDER BY dt ROWS BETWEEN 6 PRECEDING AND CURRENT ROW), 1) AS daily_new_deaths_7d_ma,85 ROUND(AVG(daily_new_confirmed) OVER(PARTITION BY combined_key ORDER BY dt ROWS BETWEEN 27 PRECEDING AND CURRENT ROW), 1) AS daily_new_confirmed_28d_ma,86 ROUND(AVG(daily_new_deaths) OVER(PARTITION BY combined_key ORDER BY dt ROWS BETWEEN 27 PRECEDING AND CURRENT ROW), 1) AS daily_new_deaths_28d_ma,87 IF(population = 0, NULL, ROUND(total_confirmed/population, 4)) AS incident_rate,88 IF(total_confirmed = 0, NULL, ROUND(total_deaths/total_confirmed, 4)) AS case_fatality_rate,89 MAX(dt) OVER() AS last_updated_dt,90 TIMESTAMP(REGEXP_REPLACE(STRING(CURRENT_TIMESTAMP, "America/Los_Angeles"), r'[\+-][0-9]{{2}}$', '')) AS last_updated_ts -- need the double bracket to avoid error with str.format91FROM (92 SELECT93 * EXCEPT(daily_new_confirmed, daily_new_deaths),94 IFNULL(daily_new_confirmed, total_confirmed) AS daily_new_confirmed,95 IFNULL(daily_new_deaths, total_deaths) AS daily_new_deaths96 FROM (97 SELECT98 *,99 total_confirmed - LAG(total_confirmed) OVER(PARTITION BY combined_key ORDER BY dt) AS daily_new_confirmed,100 total_deaths - LAG(total_deaths) OVER(PARTITION BY combined_key ORDER BY dt) AS daily_new_deaths101 FROM (102 SELECT103 a.*, b.population104 FROM (105 SELECT106 COALESCE(a.dt, b.dt) AS dt,107 COALESCE(a.county, b.county) AS county,108 COALESCE(a.province_state, b.province_state) AS province_state,109 COALESCE(a.country_region, b.country_region) AS country_region,110 COALESCE(a.combined_key, b.combined_key) AS combined_key,111 COALESCE(a.latitude, b.latitude) AS latitude,112 COALESCE(a.longitude, b.longitude) AS longitude,113 IFNULL(a.total_confirmed, 0) AS total_confirmed,114 IFNULL(b.total_deaths, 0) AS total_deaths115 FROM `stanleysfang.surveillance_2019_ncov.ts_2019_ncov_{geo}_confirmed` a116 FULL JOIN `stanleysfang.surveillance_2019_ncov.ts_2019_ncov_{geo}_deaths` b117 ON a.dt = b.dt AND a.combined_key = b.combined_key118 ) a119 LEFT JOIN `stanleysfang.surveillance_2019_ncov.ts_2019_ncov_{geo}_deaths_raw` b120 ON a.combined_key = b.combined_key121 )122 )123)124""".format(geo=geo)125query_job = qr.run_query(ts_2019_ncov_query, destination_table='stanleysfang.surveillance_2019_ncov.ts_2019_ncov_{geo}'.format(geo=geo), time_partitioning=True, partition_field='dt')126print_job_result(query_job, client)127# US current table128us_cur_query = \129"""130SELECT *131FROM `stanleysfang.surveillance_2019_ncov.ts_2019_ncov_{geo}`132WHERE dt = last_updated_dt133""".format(geo=geo)134query_job = qr.run_query(us_cur_query, destination_table='stanleysfang.surveillance_2019_ncov.ts_2019_ncov_{geo}_cur'.format(geo=geo))135print_job_result(query_job, client)136#### Extract Table ####137extract_job = extractor.extract(query_job.destination, 'gs://surveillance_2019_ncov/ts_2019_ncov_{geo}_cur.csv'.format(geo=geo))...

Full Screen

Full Screen

master.py

Source:master.py Github

copy

Full Screen

...46 print(f'\nWorker {worker}: {data}')47 client.sendall(bytes("EXECUTE_JOB", "utf-8"))48 client.sendall(bytes(job, "utf-8"))49 result = client.recv(1024).decode('utf-8')50 print_job_result(result, job)51 job_execution.pop(job)52 workers[worker] = 'ACTIVE'53def check_job_execution():54 print('---------------RESULT---------------')55 if not job_execution:56 print()57 print('There is no job can be execute')58 else:59 print('\n')60 for key in job_execution:61 print('Job '+key+' got executed in Worker Node : '+job_execution[key])62 print('------------------------------------')63def check_worker_availability():64 for worker, status in workers.items():65 try:66 client = socket.socket(socket.AF_INET, socket.SOCK_STREAM)67 client.connect((worker, PORT))68 # Print Worker 'IP': Connection accepted69 data = client.recv(1024).decode('utf-8')70 print(f'\nWorker {worker}: {data}')71 # Print Worker 'IP': 'Status'72 client.send(bytes('CHECK_AVAILABILITY','utf-8'))73 data = client.recv(1024).decode('utf-8')74 workers[worker] = data75 print(f'Worker {worker}: {data}')76 except(OSError):77 print(f"Worker {worker}: DEAD")78 workers[worker] = 'DEAD'79 continue80def check_job():81 for worker, status_job in workers.items():82 try:83 client = socket.socket(socket.AF_INET, socket.SOCK_STREAM)84 client.connect((worker, PORT))85 # Print Worker 'IP': Connection accepted86 data = client.recv(1024).decode('utf-8')87 print(f'\nWorker {worker}: {data}')88 # Print Worker 'IP': 'Status'89 client.send(bytes('CHECK_STATUS_JOB','utf-8'))90 data = client.recv(1024).decode('utf-8')91 workers[worker] = data92 print(f'Worker {worker}, Status Job= : {data}')93 except(OSError):94 print(f"Worker {worker}: DEAD")95 workers[worker] = 'DEAD'96 continue97def print_list_of_actions():98 states = "Action"99 action = input('\n{}\n{}\n{}\n{}\n{}\n{}\n{}\n{}\n{}'.format(100 '-------------------- WAITING FOR NEW ACTIONS --------------------',101 '*** List of actions: ***',102 "[1] Execute jobs on server",103 "[2] Check worker node availability",104 "[3] Check worker status job",105 "[4] Check execution place",106 "[0] Exit",107 '************************',108 "Input Action(just number): "))109 return int(action)110def print_list_of_jobs():111 states = "Job"112 # print('\n-------------------- Waiting for new jobs --------------------')113 job = input('\n{}\n{}\n{}\n{}\n{}\n{}'.format(114 "*** List of jobs: ***",115 "[1] List sorting. \nHOW TO USE: sort [list]",116 "[2] Book search. \nHOW TO USE: book_search \"Harry_potter\". \nNote: book title in quotes and separated with underscores (_)",117 "[3] Generate n-th fibonnaci number. \nHOW TO USE: fibonacci n",118 '*********************',119 "Input Job(pay attention to 'HOW TO USE'): "120 ))121 return job122def print_job_result(result, job):123 print('\n\n{}\n{}\n{}\n{}\n{}\n\n'.format(124 '-------------------------------------------- RESULT --------------------------------------------',125 f'RESULT FROM PREVIOUS "{job}" JOB',126 f'{result}',127 '*** To continue using the program, follow the instructions before this result section ***',128 '------------------------------------------ END RESULT ------------------------------------------'))129 print(states)130if __name__ == '__main__':...

Full Screen

Full Screen

example_job.py

Source:example_job.py Github

copy

Full Screen

...38 yield {'index': message.get('index', 0) * 10 + i,39 'job_arg1': job_arg1,40 'job_arg2': job_arg2}41@pipe_processor42def print_job_result(message, logger, job_arg1=None, job_arg2=None):43 logger.info('Job with parameters (%s, %s) result: %s',44 job_arg1, job_arg2, message.index)45job1 = Job('job1', job_arg1=context, job_arg2=context)46job2 = Job('job2', job_arg1=context)47job1_pipeline = job1(Message.log('Job 1 start message') >>48 multiplicate >> multiplicate >> multiplicate >> multiplicate49 ) >> print_job_result50job2_pipeline = job2(Message.log('Job 2 start message') >>51 multiplicate >> multiplicate >> multiplicate >> multiplicate52 ) >> print_job_result53program = Program(name='celery_example',54 pipelines={55 'start': Event.on_start >> Message.log('Program started'),56 'job1': job1_pipeline,...

Full Screen

Full Screen

Automation Testing Tutorials

Learn to execute automation testing from scratch with LambdaTest Learning Hub. Right from setting up the prerequisites to run your first automation test, to following best practices and diving deeper into advanced test scenarios. LambdaTest Learning Hubs compile a list of step-by-step guides to help you be proficient with different test automation frameworks i.e. Selenium, Cypress, TestNG etc.

LambdaTest Learning Hubs:

YouTube

You could also refer to video tutorials over LambdaTest YouTube channel to get step by step demonstration from industry experts.

Run autotest automation tests on LambdaTest cloud grid

Perform automation testing on 3000+ real desktop and mobile devices online.

Try LambdaTest Now !!

Get 100 minutes of automation test minutes FREE!!

Next-Gen App & Browser Testing Cloud

Was this article helpful?

Helpful

NotHelpful