Estimate processing times for various API tasks
#
This example shows how to use information from api responses
to estimate the amount of time taken by each api call.
Each api call tracks the time when the api payload is received at the backend,
and each api call produces an output file.
Hence, we can estimate the processing time as the difference between the time api payload was received,
and the time the output file was generated.
import time
import datetime
import pandas as pd
from utils.logging import logger
from utils.byte_genie import ByteGenie
Initialise ByteGenie
Initialise ByteGenie in async mode (tasks will run in the background, i.e. API call will return a response before completing the task)
bg_async = ByteGenie(
secrets_file='secrets.json',
task_mode='async',
verbose=1,
)
Initialise ByteGenie in sync mode (tasks will run in the foreground, i.e. API call will keep running until the task is finished)
bg_sync = ByteGenie(
secrets_file='secrets.json',
task_mode='sync',
verbose=1,
)
Get usage summary
To track api usage, we can use `/get_usage_summary` endpoint.
This endpoint provides daily usage between a `start_time` and an `end_time`
call /get_usage_summary endpoint
usage_resp = bg_async.get_usage_summary(
start_time='2023-09-21',
end_time='2023-10-21',
# agg_lvl='file',
route='api',
)
get response data
df_usage = usage_resp.get_output()
df_usage = pd.DataFrame(df_usage)
df_usage = df_usage[df_usage['start_time'] != ''].reset_index(drop=True)
Check processing time for each task
logger.info(df_usage[['usage_date', 'start_time', 'processing_time', 'file']].to_dict('records'))
"""
A sample of processing times for usage: `df_usage[['usage_date', 'start_time', 'processing_time', 'file']].head().to_dict('records')`
<p>
[
{'usage_date': '2023-09-29', 'start_time': 1695986266.7920444, 'processing_time': 13.095955610275269,
'file': 'gs://db-genie/entity_type=api-tasks/entity=556938a63276f9b933241de19431d3cd/data_type=structured/format=pickle/variable_desc=write_and_execute_code/source=demo-genie/556938a63276f9b933241de19431d3cd.pickle'},
{'usage_date': '2023-09-29', 'start_time': 1695993193.0123582, 'processing_time': 10.197641849517822,
'file': 'gs://db-genie/entity_type=api-tasks/entity=5b4889607efeaf69c7d2c738c0a003f7/data_type=structured/format=pickle/variable_desc=write_and_execute_code/source=demo-genie/5b4889607efeaf69c7d2c738c0a003f7.pickle'},
{'usage_date': '2023-09-29', 'start_time': 1695992131.7515397, 'processing_time': 19.38746023178101,
'file': 'gs://db-genie/entity_type=api-tasks/entity=6341de8820fd9bd16349dfb60d03a285/data_type=structured/format=pickle/variable_desc=write_and_execute_code/source=demo-genie/6341de8820fd9bd16349dfb60d03a285.pickle'},
{'usage_date': '2023-09-29', 'start_time': 1695983936.4336717, 'processing_time': 27.400328636169437,
'file': 'gs://db-genie/entity_type=api-tasks/entity=f39ca218084d5b33b3354fd95ffb3188/data_type=structured/format=pickle/variable_desc=write_and_execute_code/source=demo-genie/f39ca218084d5b33b3354fd95ffb3188.pickle'},
{'usage_date': '2023-10-14', 'start_time': 1697254511.178348, 'processing_time': 2.172652006149292,
'file': 'gs://db-genie/entity_type=api-tasks/entity=98f425ed1e7e18ffee3e279a2f7c7cd2/data_type=structured/format=pickle/variable_desc=extract_text_years/source=demo-genie/98f425ed1e7e18ffee3e279a2f7c7cd2.pickle'}
]
The processing time is in seconds.
</p>
"""
Check average processing time by API endpoints
logger.info(df_usage.groupby(['func'])[['processing_time', 'usage']].mean().reset_index().to_dict('records'))
"""
Average processing time and usage by API endpoint, `df_usage.groupby(['func'])[['processing_time', 'usage']].mean().reset_index().to_dict('records')`
<p>
[
{'func': 'classify_text', 'processing_time': 5.691191911697388, 'usage': 0.0006484985351562},
{'func': 'classify_texts', 'processing_time': 3.966008901596069, 'usage': 0.0010223388671875},
{'func': 'create_dataset', 'processing_time': 23.938680410385132, 'usage': 2.051443576812744},
{'func': 'download_documents', 'processing_time': 558.6726469993591, 'usage': 0.109130859375},
{'func': 'extract_text_years', 'processing_time': 2.297140955924988, 'usage': 0.0004730224609375},
{'func': 'filter_pages_pipeline', 'processing_time': 765.0258118510246, 'usage': 340.1848633289337},
{'func': 'standardise_years', 'processing_time': 4.605401039123535, 'usage': 0.0048713684082031},
{'func': 'structure_quants_pipeline', 'processing_time': 842.2113270759583, 'usage': 88.72238190968831},
{'func': 'write_and_execute_code', 'processing_time': 17.520346581935883, 'usage': 0.002090454101562475}
]
`processing_time` is the time between when the task is received at the backend, to when the output file is generated.
`usage` is the volume of output generated (in MegaBytes).
</p>
<p>
Pipeline tasks such as, `filter_pages_pipeline` and `structure_quants_pipeline`,
which combine multiple steps in one endpoint, typically take the longest, and product a larger volume of output.
</p>
"""
Check total usage by date
logger.info(df_usage.groupby(['usage_date'])[['n_tasks', 'processing_time', 'usage']].sum().reset_index().to_dict('records'))
"""
Total usage by date, `df_usage.groupby(['usage_date'])[['n_tasks', 'processing_time', 'usage']].sum().reset_index().to_dict('records')`
<p>
[
{'usage_date': '2023-09-29', 'n_tasks': 4, 'processing_time': 70.08138632774353, 'usage': 0.0083618164062499},
{'usage_date': '2023-10-14', 'n_tasks': 4, 'processing_time': 13.805083990097046, 'usage': 0.0106887817382812},
{'usage_date': '2023-10-15', 'n_tasks': 2, 'processing_time': 9.657200813293457, 'usage': 0.0016708374023437001},
{'usage_date': '2023-10-17', 'n_tasks': 1, 'processing_time': 822.0982067584991, 'usage': 623.9474687576294},
{'usage_date': '2023-10-18', 'n_tasks': 4, 'processing_time': 2743.1195018291473, 'usage': 189.51886749267578},
{'usage_date': '2023-10-19', 'n_tasks': 3, 'processing_time': 2549.1925797462463, 'usage': 4.803284645080566},
{'usage_date': '2023-10-20', 'n_tasks': 2, 'processing_time': 2118.3646898269653, 'usage': 549.5814867019653},
{'usage_date': '2023-10-21', 'n_tasks': 4, 'processing_time': 1045.8188865184784, 'usage': 529.5437860488892}
]
</p>
"""