def __init__(self, alert_message_queue, event_loop):
self.predictors = {}
self.alert_message_queue = alert_message_queue
self.predict_task_futures = {}
self.loop = event_loop
self.cloudwatch_logger = Cloudwatch()
self.model = os.getenv('suppression_model', 'trews')
# Start monitoring task
self.loop.create_task(self.monitor_predictors())
def __init__(self, hospital, lookback_hours, jhapi_server, jhapi_id,
jhapi_secret, lookback_days=None, op_lookback_days=None):
if jhapi_server not in servers:
raise ValueError("Incorrect server provided")
if int(lookback_hours) > 72:
raise ValueError("Lookback hours must be less than 72 hours")
self.jhapi_server = jhapi_server
self.server = servers[jhapi_server]
self.hospital = hospital
self.lookback_hours = int(lookback_hours)
self.lookback_days = int(lookback_days) if lookback_days else int(int(lookback_hours)/24.0 + 1)
self.op_lookback_days = op_lookback_days
self.from_date = (dt.datetime.now() + dt.timedelta(days=1)).strftime('%Y-%m-%d')
tomorrow = dt.datetime.now() + dt.timedelta(days=1)
self.dateFrom = (tomorrow - dt.timedelta(days=self.lookback_days)).strftime('%Y-%m-%d')
self.dateTo = tomorrow.strftime('%Y-%m-%d')
self.headers = {
'client_id': jhapi_id,
'client_secret': jhapi_secret,
'User-Agent': ''
}
self.cloudwatch_logger = Cloudwatch()
def __init__(self,
event_loop,
alert_server_port=31000,
alert_dns='0.0.0.0'):
self.db = Database()
self.loop = event_loop
self.alert_message_queue = asyncio.Queue(loop=event_loop)
self.predictor_manager = PredictorManager(self.alert_message_queue,
self.loop)
self.alert_server_port = alert_server_port
self.alert_dns = alert_dns
self.channel = os.getenv('etl_channel', 'on_opsdx_dev_etl')
self.suppression_tasks = {}
self.model = os.getenv('suppression_model', 'trews')
self.TREWS_ETL_SUPPRESSION = int(os.getenv('TREWS_ETL_SUPPRESSION', 0))
self.notify_web = int(os.getenv('notify_web', 0))
self.lookbackhours = int(os.getenv('TREWS_ETL_HOURS', 24))
self.nprocs = int(os.getenv('nprocs', 2))
self.hospital_to_predict = os.getenv('hospital_to_predict', 'HCGH')
self.push_based = bool(os.getenv('push_based', 0))
self.workspace = os.getenv('workspace', 'workspace')
self.cloudwatch_logger = Cloudwatch()
self.job_status = {}
from aiohttp import web
from aiohttp.web import Response, json_response
from aiocache import LRUMemoryCache
from aiocache.plugins import HitMissRatioPlugin
from etllib import extractor
from etl.mappings.flowsheet_ids import flowsheet_ids
import etl.io_config.core as core
import json
import pandas as pd
from etl.io_config.cloudwatch import Cloudwatch
EPIC_WEB_REQUEST_INTERVAL_SECS = int(core.get_environment_var('EPIC_WEB_REQUEST_INTERVAL_SECS', 10))
SWITCH_WEB_REQUEST = int(core.get_environment_var('SWITCH_WEB_REQUEST', 1))
CLOUDWATCH_ON = int(core.get_environment_var('CLOUDWATCH_ON', 0))
cloudwatch_logger = Cloudwatch()
order_extraction = {
extractor.extract_active_procedures,
extractor.extract_lab_orders,
extractor.extract_med_orders
}
med_order_extraction = {
extractor.extract_med_orders
}
med_admin_extraction = {
extractor.extract_med_admin
}
class AlertServer:
def __init__(self,
event_loop,
alert_server_port=31000,
alert_dns='0.0.0.0'):
self.db = Database()
self.loop = event_loop
self.alert_message_queue = asyncio.Queue(loop=event_loop)
self.predictor_manager = PredictorManager(self.alert_message_queue,
self.loop)
self.alert_server_port = alert_server_port
self.alert_dns = alert_dns
self.channel = os.getenv('etl_channel', 'on_opsdx_dev_etl')
self.suppression_tasks = {}
self.model = os.getenv('suppression_model', 'trews')
self.TREWS_ETL_SUPPRESSION = int(os.getenv('TREWS_ETL_SUPPRESSION', 0))
self.notify_web = int(os.getenv('notify_web', 0))
self.lookbackhours = int(os.getenv('TREWS_ETL_HOURS', 24))
self.nprocs = int(os.getenv('nprocs', 2))
self.hospital_to_predict = os.getenv('hospital_to_predict', 'HCGH')
self.push_based = bool(os.getenv('push_based', 0))
self.workspace = os.getenv('workspace', 'workspace')
self.cloudwatch_logger = Cloudwatch()
self.job_status = {}
async def async_init(self):
self.db_pool = await self.db.get_connection_pool()
async def convert_enc_ids_to_pat_ids(self, enc_ids):
''' Return a list of pat_ids from their corresponding enc_ids '''
async with self.db_pool.acquire() as conn:
sql = '''
SELECT distinct pat_id FROM pat_enc where enc_id
in ({})
'''.format(','.join([str(i) for i in enc_ids]))
pat_ids = await conn.fetch(sql)
return pat_ids
async def suppression(self, pat_id, tsp):
''' Alert suppression task for a single patient
and notify frontend that the patient has updated'''
async def criteria_ready(conn, pat_id, tsp):
'''
criteria is ready when
1. criteria is updated after tsp
2. no new data in criteria_meas within lookbackhours (ETL will not update criteria)
'''
sql = '''
SELECT count(*) > 0
or (select count(*) = 0 from criteria_meas m
where m.pat_id = '{pat_id}' and now() - tsp < (select value::interval from parameters where name = 'lookbackhours')) ready
FROM criteria where pat_id = '{pat_id}'
and update_date > '{tsp}'::timestamptz
'''.format(pat_id=pat_id, tsp=tsp)
cnt = await conn.fetch(sql)
return cnt[0]['ready']
async with self.db_pool.acquire() as conn:
n = 0
N = 60
logging.info("enter suppression task for {} - {}".format(
pat_id, tsp))
while not await criteria_ready(conn, pat_id, tsp):
await asyncio.sleep(10)
n += 1
logging.info("retry criteria_ready {} times for {}".format(
n, pat_id))
if n >= 60:
break
if n < 60:
logging.info("criteria is ready for {}".format(pat_id))
sql = '''
select update_suppression_alert('{pat_id}', '{channel}', '{model}', '{notify}');
'''.format(pat_id=pat_id,
channel=self.channel,
model=self.model,
notify=self.notify_web)
logging.info("suppression sql: {}".format(sql))
await conn.fetch(sql)
logging.info(
"generate suppression alert for {}".format(pat_id))
else:
logging.info("criteria is not ready for {}".format(pat_id))
def garbage_collect_suppression_tasks(self, hosp):
for task in self.suppression_tasks.get(hosp, []):
task.cancel()
self.suppression_tasks[hosp] = []
async def alert_queue_consumer(self):
'''
Check message queue and process messages
'''
logging.info("alert_queue_consumer started")
while True:
msg = await self.alert_message_queue.get()
logging.info("alert_message_queue recv msg: {}".format(msg))
# Predictor finished
if msg.get('type') == 'FIN':
if self.model == 'lmc' or self.model == 'trews-jit':
if self.TREWS_ETL_SUPPRESSION == 1:
suppression_future = asyncio.ensure_future(
self.run_suppression(msg), loop=self.loop)
elif self.TREWS_ETL_SUPPRESSION == 2:
suppression_future = asyncio.ensure_future(
self.run_suppression_mode_2(msg), loop=self.loop)
else:
logging.error("Unknown model: {}".format(self.model))
# self.suppression_tasks[msg['hosp']].append(suppression_future)
# logging.info("create {model} suppression task for {}".format(self.model,msg['hosp']))
logging.info("alert_queue_consumer quit")
async def suppression(self, pat_id, tsp):
''' Alert suppression task for a single patient
and notify frontend that the patient has updated'''
async def run_suppression_mode_2(self, msg):
t_fin = dt.datetime.now()
# if msg['hosp']+msg['time'] in self.job_status:
if msg['job_id'] in self.job_status:
t_start = self.job_status[msg['job_id']]['t_start']
self.cloudwatch_logger.push_many(
dimension_name='AlertServer',
metric_names=[
'prediction_time_{}{}'.format(
msg['hosp'], '_push' if self.push_based else ''),
'prediction_enc_cnt_in_{}{}'.format(
msg['hosp'], '_push' if self.push_based else ''),
'prediction_enc_cnt_out_{}{}'.format(
msg['hosp'], '_push' if self.push_based else '')
],
metric_values=[(t_fin - t_start).total_seconds(),
len(msg['enc_ids']),
len(msg['predicted_enc_ids'])],
metric_units=['Seconds', 'Count', 'Count'])
logging.info("start to run suppression mode 2 for msg {}".format(msg))
tsp = msg['time']
enc_id_str = ','.join([str(i) for i in msg['enc_ids'] if i])
hospital = msg['hosp']
logging.info("received FIN for enc_ids: {}".format(enc_id_str))
# calculate criteria here
# NOTE: I turst the enc_ids from FIN msg
async with self.db_pool.acquire() as conn:
if self.notify_web:
if self.push_based:
job_id = msg['job_id']
await self.calculate_criteria_enc(conn, msg['enc_ids'])
sql = '''
with pats as (
select p.enc_id, p.pat_id from pat_enc p
where p.enc_id in ({enc_ids})
),
refreshed as (
insert into refreshed_pats (refreshed_tsp, pats)
select now(), jsonb_agg(pat_id) from pats
returning id
)
select pg_notify('{channel}', 'invalidate_cache_batch:' || id || ':' || '{model}') from refreshed;
'''.format(channel=self.channel,
model=self.model,
enc_ids=enc_id_str)
else:
await self.calculate_criteria_hospital(conn, hospital)
sql = '''
with pats as (
select p.enc_id, p.pat_id from pat_enc p
inner join get_latest_enc_ids('{hosp}') e on p.enc_id = e.enc_id
),
refreshed as (
insert into refreshed_pats (refreshed_tsp, pats)
select now(), jsonb_agg(pat_id) from pats
returning id
)
select pg_notify('{channel}', 'invalidate_cache_batch:' || id || ':' || '{model}') from refreshed;
'''.format(channel=self.channel,
model=self.model,
enc_id_str=enc_id_str,
hosp=msg['hosp'])
logging.info("trews alert sql: {}".format(sql))
await conn.fetch(sql)
logging.info("generated trews alert for {}".format(hospital))
logging.info(
"complete to run suppression mode 2 for msg {}".format(msg))
t_end = dt.datetime.now()
if msg['job_id'] in self.job_status:
t_start = self.job_status[msg['job_id']]['t_start']
self.cloudwatch_logger.push_many(
dimension_name='AlertServer',
metric_names=[
'e2e_time_{}{}'.format(msg['hosp'],
'_push' if self.push_based else ''),
'criteria_time_{}{}'.format(
msg['hosp'], '_push' if self.push_based else ''),
],
metric_values=[
(t_end - t_start).total_seconds(),
(t_end - t_fin).total_seconds(),
],
metric_units=['Seconds', 'Seconds'])
self.job_status.pop(msg['job_id'], None)
async def run_suppression(self, msg):
# Wait for Advance Criteria Snapshot to finish and then start generating notifications
logging.info("start to run suppression for msg {}".format(msg))
tsp = msg['time']
pat_ids = await self.convert_enc_ids_to_pat_ids(msg['enc_ids'])
pats_str = ','.join([str(i) for i in pat_ids])
hospital = msg['hosp']
logging.info("received FIN for enc_ids: {}".format(pats_str))
async def criteria_ready(conn, enc_ids, tsp):
'''
criteria is ready when
1. criteria is updated after tsp
2. no new data in criteria_meas within lookbackhours (ETL will not update criteria)
'''
sql = '''
with pats as (
select distinct enc_id from criteria where enc_id in ({enc_ids})
),
updated_pats as (
select distinct enc_id from criteria where enc_id in ({enc_ids}) and update_date >= '{tsp}'::timestamptz
)
SELECT * from pats except select * from updated_pats
'''.format(enc_ids=enc_ids, tsp=tsp)
cnt = await conn.fetch(sql)
if cnt is None or len(cnt) == 0:
logging.info("criteria is ready")
return True
else:
logging.info("criteria is not ready ({})".format(len(cnt)))
return False
async with self.db_pool.acquire() as conn:
n = 0
N = 60
logging.info("enter suppression task for {}".format(msg))
while not await criteria_ready(conn, pats_str, tsp):
await asyncio.sleep(10)
n += 1
logging.info("retry criteria_ready {} times for {}".format(
n, pats_str))
if n >= 60:
break
if n < 60:
if self.notify_web:
sql = '''
with pats as (
select enc_id, pat_id from pat_enc where enc_id in ({pats})
),
alerts as (
select update_suppression_alert(enc_id, '{channel}', '{model}', 'false') from pats),
refreshed as (
insert into refreshed_pats (refreshed_tsp, pats)
select now(), jsonb_agg(pat_id) from pats
returning id
)
select pg_notify('{channel}', 'invalidate_cache_batch:' || id || ':' || '{model}') from refreshed;
'''.format(channel=self.channel, model=self.model, pats=pats_str)
else:
sql = '''
with pats as (
select enc_id from pat_enc where enc_id in ({pats})
)
select update_suppression_alert(enc_id, '{channel}', '{model}', 'false') from pats)
'''.format(channel=self.channel, model=self.model, pats=pats_str)
logging.info("lmc suppression sql: {}".format(sql))
await conn.fetch(sql)
logging.info(
"generate suppression alert for {}".format(hospital))
else:
logging.info("criteria is not ready for {}".format(pats_str))
async def distribute_calculate_criteria(self, conn, job_id):
server = 'dev_db' if 'dev' in self.channel else 'prod_db'
hospital = None
if 'hcgh' in job_id:
hospital = 'HCGH'
elif 'bmc' in job_id:
hospital = 'BMC'
elif 'jhh' in job_id:
hospital = 'JHH'
else:
logging.error("Invalid job id: {}".format(job_id))
if hospital:
sql = "select garbage_collection('{}');".format(hospital)
logging.info("calculate_criteria sql: {}".format(sql))
await conn.fetch(sql)
sql = '''
select distribute_advance_criteria_snapshot_for_job('{server}', {hours}, '{job_id}', {nprocs});
'''.format(server=server,
hours=self.lookbackhours,
job_id=job_id,
nprocs=self.nprocs)
logging.info("calculate_criteria sql: {}".format(sql))
await conn.fetch(sql)
async def distribute_calculate_criteria_hospital(self, conn, hospital):
server = 'dev_db' if 'dev' in self.channel else 'prod_db'
sql = "select garbage_collection('{}');".format(hospital)
logging.info("calculate_criteria sql: {}".format(sql))
await conn.fetch(sql)
sql = '''
select distribute_advance_criteria_snapshot_for_online_hospital('{server}', '{hospital}', {nprocs});
'''.format(server=server, hospital=hospital, nprocs=self.nprocs)
logging.info("calculate_criteria sql: {}".format(sql))
await conn.fetch(sql)
async def calculate_criteria_hospital(self, conn, hospital):
sql = "select garbage_collection('{}', '{}');".format(
hospital, self.workspace)
logging.info("calculate_criteria sql: {}".format(sql))
await conn.fetch(sql)
sql = '''
select advance_criteria_snapshot(enc_id) from get_latest_enc_ids('{hospital}');
'''.format(hospital=hospital)
logging.info("calculate_criteria sql: {}".format(sql))
await conn.fetch(sql)
async def calculate_criteria_enc(self, conn, enc_ids):
sql = "select garbage_collection_enc_ids(array[{}],'{}')".format(
','.join([str(enc_id) for enc_id in enc_ids]), self.workspace)
logging.info("calculate_criteria sql: {}".format(sql))
await conn.fetch(sql)
sql = 'select advance_criteria_snapshot_enc_ids(array[{}])'.format(
','.join([str(enc_id) for enc_id in enc_ids]))
logging.info("calculate_criteria sql: {}".format(sql))
await conn.fetch(sql)
logging.info("complete calculate_criteria_enc")
async def calculate_criteria_push(self, conn, job_id, excluded=None):
sql = '''
select garbage_collection(enc_id, '{workspace}')
from (select distinct enc_id from {workspace}.cdm_t
where job_id = '{job_id}' {where}) e;
'''.format(workspace=self.workspace, job_id=job_id, where=excluded)
logging.info("calculate_criteria sql: {}".format(sql))
await conn.fetch(sql)
sql = '''
select advance_criteria_snapshot_batch(
'select distinct enc_id from {workspace}.cdm_t where job_id = ''{job_id}'' {where}'
);
'''.format(workspace=self.workspace, job_id=job_id, where=excluded)
# sql = '''
# select advance_criteria_snapshot(enc_id)
# from (select distinct enc_id from {workspace}.cdm_t
# where job_id = '{job_id}' {where}) e;
# '''.format(workspace=self.workspace, job_id=job_id, where=excluded)
logging.info("calculate_criteria sql: {}".format(sql))
await conn.fetch(sql)
logging.info("complete calculate_criteria_enc")
async def get_enc_ids_to_predict(self, job_id):
async with self.db_pool.acquire() as conn:
# rule to select predictable enc_ids:
# 1. has changes delta twf > 0
# 2. adult
# 3. HCGH patients only
sql = '''
select distinct t.enc_id
from {workspace}.cdm_t t
inner join cdm_twf twf on t.enc_id = twf.enc_id
left join {workspace}.{job_id}_cdm_twf twf_delta on twf_delta.enc_id = t.enc_id
left join {workspace}.{job_id}_cdm_twf_del twf_del on twf_del.enc_id = t.enc_id
inner join cdm_s s on twf.enc_id = s.enc_id
inner join cdm_s s2 on twf.enc_id = s2.enc_id
where s.fid = 'age' and s.value::float >= 18.0
and s2.fid = 'hospital' and s2.value = 'HCGH'
and job_id = '{job_id}' and (twf_delta.enc_id is not null or twf_del.enc_id is not null)
'''.format(workspace=self.workspace, job_id=job_id)
res = await conn.fetch(sql)
predict_enc_ids = [row[0] for row in res]
return predict_enc_ids
async def run_trews_alert(self, job_id, hospital, excluded_enc_ids=None):
async with self.db_pool.acquire() as conn:
if self.push_based and hospital == 'PUSH':
# calculate criteria here
excluded = ''
if excluded_enc_ids:
excluded = 'and enc_id not in ({})'.format(','.join(
[str(id) for id in excluded_enc_ids]))
await self.calculate_criteria_push(conn,
job_id,
excluded=excluded)
if self.notify_web:
sql = '''
with pats as (
select e.enc_id, p.pat_id from (
select distinct enc_id from {workspace}.cdm_t
where job_id = '{job_id}'
{where}
) e
inner join pat_enc p on e.enc_id = p.enc_id
),
refreshed as (
insert into refreshed_pats (refreshed_tsp, pats)
select now(), jsonb_agg(pat_id) from pats
returning id
)
select pg_notify('{channel}', 'invalidate_cache_batch:' || id || ':' || '{model}') from refreshed;
'''.format(channel=self.channel,
model=self.model,
where=excluded,
workspace=self.workspace,
job_id=job_id)
logging.info("trews alert sql: {}".format(sql))
await conn.fetch(sql)
logging.info(
"generated trews alert for {} without prediction".
format(hospital))
elif self.TREWS_ETL_SUPPRESSION == 2:
# calculate criteria here
await self.calculate_criteria_hospital(conn, hospital)
if self.notify_web:
sql = '''
with pats as (
select e.enc_id, p.pat_id from get_latest_enc_ids('{hospital}') e inner join pat_enc p on e.enc_id = p.enc_id
),
refreshed as (
insert into refreshed_pats (refreshed_tsp, pats)
select now(), jsonb_agg(pat_id) from pats
returning id
)
select pg_notify('{channel}', 'invalidate_cache_batch:' || id || ':' || '{model}') from refreshed;
'''.format(channel=self.channel, model=self.model, hospital=hospital)
logging.info("trews alert sql: {}".format(sql))
await conn.fetch(sql)
logging.info(
"generated trews alert for {}".format(hospital))
elif self.TREWS_ETL_SUPPRESSION == 1:
if self.notify_web:
sql = '''
with pats as (
select e.enc_id, p.pat_id from get_latest_enc_ids('{hospital}') e inner join pat_enc p on e.enc_id = p.enc_id
),
alerts as (
select update_suppression_alert(enc_id, '{channel}', '{model}', 'false') from pats),
refreshed as (
insert into refreshed_pats (refreshed_tsp, pats)
select now(), jsonb_agg(pat_id) from pats
returning id
)
select pg_notify('{channel}', 'invalidate_cache_batch:' || id || ':' || '{model}') from refreshed;
'''.format(channel=self.channel,
model=self.model,
hospital=hospital)
else:
sql = '''
select update_suppression_alert(enc_id, '{channel}', '{model}', 'false') from
(select distinct t.enc_id from cdm_t t
inner join get_latest_enc_ids('{hospital}') h on h.enc_id = t.enc_id
where now() - tsp < (select value::interval from parameters where name = 'lookbackhours')) sub;
'''.format(channel=self.channel,
model=self.model,
hospital=hospital)
logging.info("trews suppression sql: {}".format(sql))
await conn.fetch(sql)
logging.info(
"generate trews suppression alert for {}".format(hospital))
async def connection_handler(self, reader, writer):
''' Alert server connection handler '''
addr = writer.transport.get_extra_info('peername')
sock = writer.transport.get_extra_info('socket')
if not addr:
logging.error(
'Connection made without a valid remote address, (Timeout %s)'
% str(sock.gettimeout()))
return
else:
logging.debug('Connection from %s (Timeout %s)' %
(str(addr), str(sock.gettimeout())))
# Get the message that started this callback function
message = await protocol.read_message(reader, writer)
logging.info("connection_handler: recv msg from {} type {}".format(
message.get('from'), message.get('type')))
if message.get('from') == 'predictor':
return await self.predictor_manager.register(
reader, writer, message)
elif message.get('type') == 'ETL':
self.cloudwatch_logger.push_many(
dimension_name='AlertServer',
metric_names=['etl_done_{}'.format(message['hosp'])],
metric_values=[1],
metric_units=['Count'])
# self.job_status[message['hosp'] + message['time']] = {
# 'msg': message, 't_start': dt.datetime.now()
# }
if self.model == 'lmc' or self.model == 'trews-jit':
job_id_items = message['job_id'].split('_')
t_start = parser.parse(job_id_items[-1] if len(job_id_items) ==
4 else job_id_items[-2])
if self.push_based:
# create predict task for predictor
predict_enc_ids = await self.get_enc_ids_to_predict(
message['job_id'])
if predict_enc_ids:
self.job_status[message['job_id']] = {
't_start': t_start
}
self.predictor_manager.cancel_predict_tasks(
job_id=message['job_id'])
self.predictor_manager.create_predict_tasks(
hosp=message['hosp'],
time=message['time'],
job_id=message['job_id'],
active_encids=predict_enc_ids)
else:
logging.info("predict_enc_ids is None or empty")
# create criteria update task for patients who do not need to predict
t_fin = dt.datetime.now()
await self.run_trews_alert(
message['job_id'],
message['hosp'],
excluded_enc_ids=predict_enc_ids)
t_end = dt.datetime.now()
self.cloudwatch_logger.push_many(
dimension_name='AlertServer',
metric_names=[
'e2e_time_{}{}'.format(
message['hosp'],
'_short' if self.push_based else ''),
'criteria_time_{}{}'.format(
message['hosp'],
'_short' if self.push_based else ''),
],
metric_values=[
(t_end - t_start).total_seconds(),
(t_end - t_fin).total_seconds(),
],
metric_units=['Seconds', 'Seconds'])
elif message.get('hosp') in self.hospital_to_predict:
if self.model == 'lmc':
self.garbage_collect_suppression_tasks(message['hosp'])
self.job_status[message['job_id']] = {'t_start': t_start}
self.predictor_manager.cancel_predict_tasks(
job_id=message['hosp'])
self.predictor_manager.create_predict_tasks(
hosp=message['hosp'],
time=message['time'],
job_id=message['job_id'])
else:
logging.info("skip prediction for msg: {}".format(message))
t_fin = dt.datetime.now()
await self.run_trews_alert(message['job_id'],
message['hosp'])
t_end = dt.datetime.now()
self.cloudwatch_logger.push_many(
dimension_name='AlertServer',
metric_names=[
'e2e_time_{}'.format(message['hosp']),
'criteria_time_{}'.format(message['hosp']),
],
metric_values=[
(t_end - t_start).total_seconds(),
(t_end - t_fin).total_seconds(),
],
metric_units=['Seconds', 'Seconds'])
elif self.model == 'trews':
await self.run_trews_alert(message['job_id'], message['hosp'])
else:
logging.error("Unknown suppression model {}".format(
self.model))
else:
logging.error("Don't know how to process this message")
def start(self):
''' Start the alert server and queue consumer '''
self.loop.run_until_complete(self.async_init())
consumer_future = asyncio.ensure_future(self.alert_queue_consumer())
server_future = self.loop.run_until_complete(
asyncio.start_server(self.connection_handler,
self.alert_dns,
self.alert_server_port,
loop=self.loop))
logging.info('Serving on {}'.format(
server_future.sockets[0].getsockname()))
# Run server until Ctrl+C is pressed
try:
self.loop.run_forever()
except KeyboardInterrupt:
print("Exiting")
consumer_future.cancel()
# Close the server
logging.info('received stop signal, cancelling tasks...')
for task in asyncio.Task.all_tasks():
task.cancel()
logging.info('bye, exiting in a minute...')
server_future.close()
self.loop.run_until_complete(server_future.wait_closed())
self.loop.stop()
finally:
self.loop.close()
class PredictorManager:
def __init__(self, alert_message_queue, event_loop):
self.predictors = {}
self.alert_message_queue = alert_message_queue
self.predict_task_futures = {}
self.loop = event_loop
self.cloudwatch_logger = Cloudwatch()
self.model = os.getenv('suppression_model', 'trews')
# Start monitoring task
self.loop.create_task(self.monitor_predictors())
async def monitor_predictors(self):
''' Monitor predictors and log current status in cloudwatch every 30 seconds '''
while True:
# Build list of tuples of cloudwatch info (name, value, unit)
metric_tuples = []
# Get overall predictor info
metric_tuples.append(
('push_num_predictors', int(len(self.predictors)), 'Count'))
# Send individual predictor info to cloudwatch
for pred_id, pred in self.predictors.items():
metric_tuples += [
('push_predictor_{}_{}_{}_status'.format(*pred_id),
STATUS_DICT[pred.status], 'None'),
]
logging.info("cloudwatch metrics: {}".format(metric_tuples))
# Send all info to cloudwatch
self.cloudwatch_logger.push_many(
dimension_name='LMCPredictors',
metric_names=[metric[0] for metric in metric_tuples],
metric_values=[metric[1] for metric in metric_tuples],
metric_units=[metric[2] for metric in metric_tuples])
await asyncio.sleep(1)
async def register(self, reader, writer, msg):
''' Register connection from a predictor '''
# Create predictor object
pred = Predictor(reader, writer, msg['status'], msg['node_index'],
msg['partition_index'], msg['model_type'],
msg['is_active'], msg['ip_address'], self)
# Cancel any existing predictor with same id
if pred.id in self.predictors:
self.predictors[pred.id].shutdown = True
# Save predictor in data structure
self.predictors[pred.id] = pred
logging.info("Registered {}".format(pred))
# Start listener loop
return await pred.listen(self.alert_message_queue)
def get_partition_ids(self):
return set([p.partition_index for p in self.predictors.values()])
def get_model_types(self):
return set([p.model_type for p in self.predictors.values()])
def cancel_predict_tasks(self, job_id):
''' Cancel the existing tasks for previous ETL '''
logging.info("cancel the existing tasks for previous ETL")
for future in self.predict_task_futures.get(job_id, []):
future.cancel()
logging.info("{} cancelled".format(future))
def create_predict_tasks(self, hosp, time, job_id, active_encids=None):
''' Start all predictors '''
logging.info("Starting all predictors for ETL {} {} {} {}".format(
hosp, time, job_id, active_encids))
self.predict_task_futures[job_id] = []
for pid in self.get_partition_ids():
for model in self.get_model_types():
future = asyncio.ensure_future(self.run_predict(
pid, model, hosp, time, job_id, active_encids),
loop=self.loop)
self.predict_task_futures[job_id].append(future)
logging.info("Started {} predictors".format(
len(self.predict_task_futures[job_id])))
async def run_predict(self,
partition_id,
model_type,
hosp,
time,
job_id,
active_encids,
active=True):
''' Start a predictor for a given partition id and model '''
backoff = 1
# Start the predictor
logging.info(
"start the predictors for {} at {}: partition_id {} model_type {}".
format(hosp, time, partition_id, model_type))
while True:
pred = self.predictors.get((partition_id, model_type, active))
if pred and pred.status != 'DEAD':
try:
predictor_started = await pred.start_predictor(
hosp, time, job_id, active_encids)
break
except (ConnectionRefusedError) as e:
err = e
else:
err = '{} dead'.format(
predictor_str(partition_id, model_type, active))
if self.model == 'trews-jit':
return
else:
# Log reason for error
logging.error("{} -- trying {} predictor {}".format(
err, 'backup' if active else 'active',
humanize.naturaltime(dt.datetime.now() +
dt.timedelta(seconds=backoff))))
# Switch to backup if active (and vice versa)
active = not active
await asyncio.sleep(backoff)
backoff = backoff * 2 if backoff < 64 else backoff
# Check status
if predictor_started == False:
return None
class JHAPIConfig:
def __init__(self, hospital, lookback_hours, jhapi_server, jhapi_id,
jhapi_secret, lookback_days=None, op_lookback_days=None):
if jhapi_server not in servers:
raise ValueError("Incorrect server provided")
if int(lookback_hours) > 72:
raise ValueError("Lookback hours must be less than 72 hours")
self.jhapi_server = jhapi_server
self.server = servers[jhapi_server]
self.hospital = hospital
self.lookback_hours = int(lookback_hours)
self.lookback_days = int(lookback_days) if lookback_days else int(int(lookback_hours)/24.0 + 1)
self.op_lookback_days = op_lookback_days
self.from_date = (dt.datetime.now() + dt.timedelta(days=1)).strftime('%Y-%m-%d')
tomorrow = dt.datetime.now() + dt.timedelta(days=1)
self.dateFrom = (tomorrow - dt.timedelta(days=self.lookback_days)).strftime('%Y-%m-%d')
self.dateTo = tomorrow.strftime('%Y-%m-%d')
self.headers = {
'client_id': jhapi_id,
'client_secret': jhapi_secret,
'User-Agent': ''
}
self.cloudwatch_logger = Cloudwatch()
def make_requests(self, ctxt, endpoint, payloads, http_method='GET', url_type=None, server_type='internal'):
# Define variables
server = self.server if server_type == 'internal' else servers['{}-{}'.format(self.jhapi_server, server_type)]
url = "{}{}".format(server, endpoint)
request_settings = self.generate_request_settings(http_method, url, payloads, url_type)
semaphore = asyncio.Semaphore(ctxt.flags.JHAPI_SEMAPHORE, loop=ctxt.loop)
base = ctxt.flags.JHAPI_BACKOFF_BASE
max_backoff = ctxt.flags.JHAPI_BACKOFF_MAX
session_attempts = ctxt.flags.JHAPI_ATTEMPTS_SESSION
request_attempts = ctxt.flags.JHAPI_ATTEMPTS_REQUEST
# Asyncronous task to make a request
async def fetch(session, sem, setting):
success = 0
error = 0
for i in range(request_attempts):
try:
async with sem:
async with session.request(**setting) as response:
if response.status != 200:
body = await response.text()
logging.error("Status={}\tMessage={}\tRequest={}".format(response.status, body, setting))
response = None
error += 1
else:
response = await response.json()
success += 1
break
except IOError as e:
if i < request_attempts - 1 and not e.errno in (104): # Connection reset by peer
logging.error(e)
wait_time = min(((base**i) + random.uniform(0, 1)), max_backoff)
sleep(wait_time)
else:
logging.error("Request IOError: Request={}".format(setting))
raise Exception("Fail to request URL {}".format(url))
except Exception as e:
logging.error("Request exception: Request={}".format(setting))
if i < request_attempts - 1 and str(e) != 'Session is closed':
logging.error(e)
wait_time = min(((base**i) + random.uniform(0, 1)), max_backoff)
sleep(wait_time)
else:
raise Exception("Fail to request URL {}".format(url))
return response, i+1, success, error
# Get the client session and create a task for each request
async def run(request_settings, semaphore, loop):
async with ClientSession(headers=self.headers, loop=ctxt.loop) as session:
tasks = [asyncio.ensure_future(fetch(session, semaphore, setting),
loop=loop) for setting in request_settings]
return await asyncio.gather(*tasks)
# Start the run task to make all requests
for attempt in range(session_attempts):
try:
task = run(request_settings, semaphore, ctxt.loop)
future = asyncio.ensure_future(task, loop=ctxt.loop)
ctxt.loop.run_until_complete(future)
break
except Exception as e:
if attempt < session_attempts - 1:
logging.error("Session Error Caught for URL {}, retrying... {} times".format(url, attempt+1))
logging.exception(e)
wait_time = min(((base**attempt) + random.uniform(0, 1)), max_backoff)
sleep(wait_time)
else:
raise Exception("Session failed for URL {}".format(url))
# Push number of requests to cloudwatch
logging.info("Made {} requests".format(sum(x[1] for x in future.result())))
self.cloudwatch_logger.push(
dimension_name = 'ETL',
metric_name = 'requests_made',
value = sum(x[1] for x in future.result()),
unit = 'Count'
)
label = self.hospital + '_' + endpoint.replace('/', '_') + '_' + http_method
self.cloudwatch_logger.push_many(
dimension_name = 'ETL',
metric_names = ['{}_success'.format(label), '{}_error'.format(label), 'jh_api_request_success', 'jh_api_request_error'],
metric_values = [sum(x[2] for x in future.result()), sum(x[3] for x in future.result()), sum(x[2] for x in future.result()), sum(x[3] for x in future.result())],
metric_units = ['Count','Count','Count','Count']
)
# Return responses
return [x[0] for x in future.result()]
def generate_request_settings(self, http_method, url, payloads=None, url_type=None):
request_settings = []
if url_type == 'rest' and http_method == 'GET':
for payload in payloads:
u = url + payload
request_settings.append({'method': http_method,'url': u})
else:
for payload in payloads:
setting = {
'method': http_method,
'url': url
}
if payload is not None:
key = 'params' if http_method == 'GET' else 'json'
setting[key] = payload
request_settings.append(setting)
return request_settings
def extract_bedded_patients(self, ctxt, hospital, limit=None):
resource = '/facilities/hospital/' + self.hospital + '/beddedpatients'
responses = self.make_requests(ctxt, resource, [None], 'GET')
if limit:
ctxt.log.info("max_num_pats = {}".format(limit))
df = pd.DataFrame(responses[0]).head(limit) if limit else pd.DataFrame(responses[0])
if df.empty:
ctxt.log.error("No beddedpatients.")
sys.exit()
return df.assign(hospital = hospital)
def extract_ed_patients(self, ctxt, hospital, limit=None):
resource = '/facilities/hospital/' + self.hospital + '/edptntlist?eddept=ADULT'
responses = self.make_requests(ctxt, resource, [None], 'GET')
if limit:
ctxt.log.info("max_num_pats = {}".format(limit))
df = pd.DataFrame(responses[0]).head(limit) if limit else pd.DataFrame(responses[0])
return df.assign(hospital = hospital)
def combine(self, response_list, to_merge):
if type(response_list) != list:
raise TypeError("First argument must be a list of responses")
dfs = pd.DataFrame()
for idx, df in enumerate(response_list):
if not df.empty:
dfs = pd.concat([dfs, df.assign(index_col=idx)])
if dfs.empty:
return dfs
return pd.merge(dfs, to_merge, how='inner', left_on='index_col',
right_index=True, sort=False).drop('index_col', axis=1)
def extract_ed_patients_mrn(self, ctxt, ed_patients):
resource = '/patients/mrn/'
payloads = [row['pat_id'] for i, row in ed_patients.iterrows()]
responses = self.make_requests(ctxt, resource, payloads, 'GET', url_type='rest')
def calculate_age(born):
today = date.today()
return today.year - born.year - ((today.month, today.day) < (born.month, born.day))
for r in responses:
pat_id = [pid["ID"] for pid in r[0]['IDs'] if pid['Type'] == 'EMRN'][0]
sex = r[0]['Sex']
gender = 0 if sex == 'Female' else 1
dob = parse(r[0]["DateOfBirth"])
age = calculate_age(dob)
ed_patients.loc[ed_patients.pat_id == pat_id,'age'] = age
ed_patients.loc[ed_patients.pat_id == pat_id,'gender'] = gender
return ed_patients
def extract_chiefcomplaint(self, ctxt, beddedpatients):
resource = '/patients/getdata/chiefcomplaint'
payloads = [{
"ContactID": {
"ID": pat['visit_id'],
"Type": "CSN"
},
"DataFormat": None,
"Items": [
{
"ItemNumber": "18100",
"LineRange": {
"From": 1,
"To": 10
}
}
],
"RecordID": {
"ID": pat['pat_id'],
"Type":"EMRN"
}
} for _, pat in beddedpatients.iterrows()]
responses = self.make_requests(ctxt, resource, payloads, 'POST', server_type='epic')
for r in responses:
if r:
raw_items = r['Items'][0]
new_items = '[' + ','.join(["{{\"reason\" : \"{}\"}}".format(reason) for reason in [item['Value'] for item in raw_items['Lines'] if item['LineNumber'] > 0]]) + ']'
r['Items'] = new_items
r['RecordIDs'] = None
r['ContactIDs'] = [id for id in r['ContactIDs'] if id['Type'] == 'CSN']
dfs = [pd.DataFrame(r) for r in responses]
df = self.combine(dfs, beddedpatients[['pat_id', 'visit_id']])
return df
def extract_flowsheets(self, ctxt, bedded_patients):
resource = '/patients/flowsheetrows'
flowsheet_row_ids = []
for fid, internal_id_list in flowsheet_ids:
for internal_id in internal_id_list:
flowsheet_row_ids.append({'ID': str(internal_id),
'Type': 'Internal'})
payloads = [{
'ContactID': pat['visit_id'],
'ContactIDType': 'CSN',
'FlowsheetRowIDs': flowsheet_row_ids,
'LookbackHours': self.lookback_hours,
'PatientID': pat['pat_id'],
'PatientIDType': 'EMRN'
} for _, pat in bedded_patients.iterrows()]
responses = self.make_requests(ctxt, resource, payloads, 'POST')
dfs = [pd.DataFrame(r) for r in responses]
return self.combine(dfs, bedded_patients[['pat_id', 'visit_id']])
def extract_active_procedures(self, ctxt, bedded_patients):
resource = '/facilities/hospital/' + self.hospital + '/orders/activeprocedures'
payloads = [{'csn': pat['visit_id']} for _, pat in bedded_patients.iterrows()]
responses = self.make_requests(ctxt, resource, payloads, 'GET')
dfs = [pd.DataFrame(r) for r in responses]
return self.combine(dfs, bedded_patients[['pat_id', 'visit_id']])
def extract_treatmentteam(self, ctxt, bedded_patients):
resource = '/patients/treatmentteam'
payloads = [{
'id': pat['visit_id'],
'idtype': 'csn'
} for _, pat in bedded_patients.iterrows()]
responses = self.make_requests(ctxt, resource, payloads, 'GET')
dfs = [pd.DataFrame(r['TreatmentTeam'] if r else None) for r in responses]
return self.combine(dfs, bedded_patients[['pat_id', 'visit_id']])
def extract_lab_orders(self, ctxt, bedded_patients):
resource = '/patients/labs/procedure'
procedure_types = []
for _, ids in lab_procedure_ids:
procedure_types += ({'Type': 'INTERNAL', 'ID': str(x)} for x in ids)
payloads = [{
'Id': pat['pat_id'],
'IdType': 'patient',
'FromDate': self.from_date,
'MaxNumberOfResults': 200,
'NumberDaysToLookBack': self.lookback_days,
'ProcedureTypes': procedure_types
} for _, pat in bedded_patients.iterrows()]
responses = self.make_requests(ctxt, resource, payloads, 'POST')
dfs = [pd.DataFrame(r['ProcedureResults'] if r else None) for r in responses]
return self.combine(dfs, bedded_patients[['pat_id', 'visit_id']])
def extract_lab_results(self, ctxt, bedded_patients):
resource = '/patients/labs/component'
component_types = []
for _, cidl in component_ids:
component_types += ({'Type': 'INTERNAL', 'Value': str(x)} for x in cidl)
payloads = [{
'Id': pat['pat_id'],
'IdType': 'patient',
'FromDate': self.from_date,
'MaxNumberOfResults': 200,
'NumberDaysToLookBack': self.lookback_days,
'ComponentTypes': component_types
} for _, pat in bedded_patients.iterrows()]
responses = self.make_requests(ctxt, resource, payloads, 'POST')
dfs = [pd.DataFrame(r['ResultComponents'] if r else None) for r in responses]
return self.combine(dfs, bedded_patients[['pat_id', 'visit_id']])
def extract_loc_history(self, ctxt, bedded_patients):
resource = '/patients/adtlocationhistory'
payloads = [{
'id': pat['visit_id'],
'type': 'CSN'
} for _, pat in bedded_patients.iterrows()]
responses = self.make_requests(ctxt, resource, payloads, 'GET')
dfs = [pd.DataFrame(r) for r in responses]
return self.combine(dfs, bedded_patients[['pat_id', 'visit_id']])
def extract_med_orders(self, ctxt, bedded_patients):
resource = '/patients/medications'
payloads = [{
'id': pat['pat_id'],
'dayslookback': str(self.lookback_days),
'searchtype': 'IP'
} for _, pat in bedded_patients.iterrows()] + \
[{
'id': pat['pat_id'],
'dayslookback': str(self.op_lookback_days),
'searchtype': 'OP'
} for _, pat in bedded_patients.iterrows()]
responses = self.make_requests(ctxt, resource, payloads, 'GET')
dfs = [pd.DataFrame(r) for r in responses]
half = len(dfs)//2
med_ip = self.combine(dfs[:half], bedded_patients[['pat_id', 'visit_id']])
med_op = self.combine(dfs[half:], bedded_patients[['pat_id', 'visit_id']])
def remove_inpatient(val):
if val is None or str(val) == 'nan':
return val
for item in val:
if item == 'OrderMode' and val[item] == 'Outpatient':
return val
return {}
med_op['MedicationOrders'] = med_op['MedicationOrders'].apply(remove_inpatient)
return pd.concat([med_ip, med_op]).reset_index(drop=True)
def extract_med_admin(self, ctxt, med_orders):
if med_orders is None or med_orders.empty:
return None
else:
resource = '/patients/medicationadministrationhistory'
payloads = [{
'ContactID': order['visit_id'],
'ContactIDType': 'CSN',
'OrderIDs': list(itertools.chain.from_iterable(order['ids'])),
'PatientID': order['pat_id']
} for _, order in med_orders.iterrows()]
responses = self.make_requests(ctxt, resource, payloads, 'POST')
dfs = [pd.DataFrame(r) for r in responses]
return self.combine(dfs, med_orders[['pat_id', 'visit_id']])
def extract_notes(self, ctxt, bedded_patients):
resource = '/patients/documents/list'
payloads = [{
'id' : pat['pat_id'],
'dateFrom' : self.dateFrom,
'dateTo' : self.dateTo
} for _, pat in bedded_patients.iterrows()]
responses = self.make_requests(ctxt, resource, payloads, 'GET')
logging.info('#NOTES PAYLOADS: %s' % len(payloads))
logging.info('#NOTES RESPONSES: %s' % len(responses))
dfs = [pd.DataFrame(r['DocumentListData'] if r else None) for r in responses]
df = self.combine(dfs, bedded_patients[['pat_id']])
if not df.empty:
not_empty_idx = df.Key.str.len() > 0
df = df[not_empty_idx].reset_index()
return df
def extract_note_texts(self, ctxt, notes):
if not notes.empty:
resource = '/patients/documents/text'
payloads = [{ 'key' : note['Key'] } for _, note in notes.iterrows()]
responses = self.make_requests(ctxt, resource, payloads, 'GET')
logging.info('#NOTE TEXTS PAYLOADS: %s' % len(payloads))
logging.info('#NOTE TEXTS RESPONSES: %s' % len(responses))
dfs = [
pd.DataFrame([{'DocumentText': r['DocumentText']}] if r else None)
for r in responses
]
return self.combine(dfs, notes[['Key']])
return pd.DataFrame()
def extract_contacts(self, ctxt, pat_id_list):
if not pat_id_list:
return pd.DataFrame()
resource = '/patients/contacts'
pat_id_df = pd.DataFrame(pat_id_list)
# Get rid of fake patients by filtering out incorrect pat_ids
pat_id_df = pat_id_df[pat_id_df['pat_id'].str.contains('E.*')]
payloads = [{
'id' : pat['visit_id'],
'idtype' : 'csn',
'dateFrom' : self.dateFrom,
'dateTo' : self.dateTo,
} for _, pat in pat_id_df.iterrows()]
responses = self.make_requests(ctxt, resource, payloads, 'GET')
response_dfs = [pd.DataFrame(r['Contacts'] if r else None) for r in responses]
dfs = pd.concat(response_dfs)
if dfs.empty:
return None
else:
return pd.merge(pat_id_df, dfs, left_on='visit_id', right_on='CSN')
def push_notifications(self, ctxt, notifications):
if ctxt.flags.TREWS_ETL_EPIC_NOTIFICATIONS:
logging.info("pushing notifications to epic")
resource = '/patients/addflowsheetvalue'
load_tz='US/Eastern'
t_utc = dt.datetime.utcnow().replace(tzinfo=pytz.utc)
current_time = str(t_utc.astimezone(pytz.timezone(load_tz)))
payloads = [{
'PatientID': n['pat_id'],
'ContactID': n['visit_id'],
'UserID': 'WSEPSIS',
'FlowsheetID': '9490',
'Value': n['count'],
'InstantValueTaken': current_time,
'FlowsheetTemplateID': '304700006',
} for n in notifications]
for payload in payloads:
logging.info('%s NOTIFY %s %s %s' % (payload['InstantValueTaken'],
payload['PatientID'],
payload['ContactID'],
payload['Value']))
self.make_requests(ctxt, resource, payloads, 'POST')
logging.info("pushed notifications to epic")
else:
logging.info("not pushing notifications to epic")