#!/usr/bin/env python
#
# This script is for feeding data to the monitoring system while running CDAQ
# instead of CODA. The data are read from files appearing on volatile.
#
# This takes no arguments but it does expect the following:
#
# RUN_PERIOD    - from RUN_PERIOD environment variable
# run number    - from HD:coda:daq:run_number EPICS variable
#
# In addition, the ET_FILENAME and host are hard-coded here. In particular, if
# the ET line in $DAQ_HOME/config/monitoring/nodes.conf changes, then it should
# be changed here as well.
#
# This will continuously scan for files on all gluonraid partitions for whatever
# the current run is (by way of the HD:coda:daq:run_number EPICS variable).
# When a new one is found, it runs file2et to feed events from it into the
# secondary ET system. If a newer file appears, the existing file2et process
# is killed and a new one started for the new file.
#

import os
import sys
import glob
import time
import psutil
import socket
import datetime
import subprocess
from epics import caget

RUNPERIOD = 'RunPeriod-2019-11'

proc_file2et    = None
CURRENT_FILE    = ''
LAST_RUN_NUMBER = 0
SLEEP_TIME      = 10  # seconds to sleep between iterations
ET_FILENAME     = '/tmp/et_sys_monitoring'  # This is the default in start_monitoring
ET_HOST         = 'gluon102'            # This is specified in $DAQ_HOME/config/monitoring/nodes.conf

hostname = socket.gethostname()
hostname_short = hostname.split('.jlab.org')[0].split('-daq')[0]

#----------------------------
#  StartFile2ET
#
# Find age of most recently modified file in seconds in given directory.
#----------------------------
def StartFile2ET(fname):
	global proc_file2et, CURRENT_FILE, ET_FILENAME, ET_HOST

	logfile = '/gluex/log/file2et_%s.log' % hostname_short
	cmd = ['hdlog', '-c', '-o', logfile, 'file2et', '-f', ET_FILENAME, '-H', ET_HOST, fname]
	print('[%s] Running: %s' % (datetime.datetime.now(), ' '.join(cmd)))
	proc_file2et = subprocess.Popen( cmd, stdout=subprocess.PIPE, stderr=subprocess.PIPE )
	CURRENT_FILE = fname

#----------------------------
#  StopFile2ET
#
# Find age of most recently modified file in seconds in given directory.
#----------------------------
def StopFile2ET():
	global proc_file2et, CURRENT_FILE

	if not proc_file2et: return  # no process to stop!

	print('[%s] Stopping process feeding file: %s' %(datetime.datetime.now(), CURRENT_FILE))

	# Send SIGTERM and wait up to 3 seconds for process to terminate.
	# If it doesn't, then send a SIGKILL.
	proc_file2et.terminate()
	gone,alive = psutil.wait_procs( [proc_file2et], timeout=3 )
	if proc_file2et in alive:
		proc_file2et.kill()

	# Reset globals
	proc_file2et    = None
	CURRENT_FILE    = ''

#===========================================================================

# Loop over command line arguments
for arg in sys.argv[1:]:
	pass

RUN_PERIOD = os.getenv('RUN_PERIOD')
if not RUN_PERIOD:
	print('RUN_PERIOD environment variable not set!')
	os.exit(-1)

# Loop forever, checking the run number and looking for new files
while True:

	# Get run number from EPICS
	try:
		run_number = int(caget('HD:coda:daq:run_number'))
		if run_number != LAST_RUN_NUMBER:
			print('[%s] Run number changed from %d to %d' % (datetime.datetime.now(), LAST_RUN_NUMBER, run_number))
			StopFile2ET()
			LAST_RUN_NUMBER = run_number
	except:
		print('[%s] ERROR getting run number from EPICS!' % (datetime.datetime.now()))
		print(sys.exc_info())
		run_number = -1

	# Search for most recent file on volatile from current run/runperiod
	filepat = '/gluonraid*/data*/rawdata/volatile/%s/rawdata/Run%06d/hd_rawdata_*.evio' %( RUN_PERIOD, run_number)
	files = glob.glob( filepat )
	if files:
		latest_file = max( files, key=os.path.getctime)
		if latest_file != CURRENT_FILE:
			StopFile2ET()   # This will silently do nothing if the run number changed and StopFile2ET() was already called above
			StartFile2ET(latest_file)
		else:
			if not proc_file2et:
				print('[%s] No new files to process for run %d' % (datetime.datetime.now(), run_number))
	else:
		print('[%s] No files to process for run %d' % (datetime.datetime.now(), run_number))

	# Check if existing process is stopped
	if proc_file2et:
		if proc_file2et.poll() is not None:
			print('[%s] Process ended for file: %s' % (datetime.datetime.now(), CURRENT_FILE))
			proc_file2et = None
			# Note: do not reset CURRENT_FILE here since that would cause the above to re-run it.
		else:
			print('[%s] Continuing to feed %s' % (datetime.datetime.now(), CURRENT_FILE))

	time.sleep( SLEEP_TIME )