line-and-stream-rates.py

###############################################################################
# (c) Copyright 2000-2022 CERN for the benefit of the LHCb Collaboration      #
#                                                                             #
# This software is distributed under the terms of the GNU General Public      #
# Licence version 3 (GPL Version 3), copied verbatim in the file "COPYING".   #
#                                                                             #
# In applying this licence, CERN does not waive the privileges and immunities #
# granted to it by virtue of its status as an Intergovernmental Organization  #
# or submit itself to any jurisdiction.                                       #
###############################################################################

import GaudiPython as GP
from GaudiConf.reading import decoder, unpack_rawevent, hlt_decisions
from Configurables import (ApplicationMgr, LHCbApp, IODataManager,
                           EventSelector, createODIN)
from GaudiConf import IOHelper
from PyConf.application import configured_ann_svc
import operator
from collections import Counter
import json
import argparse
import csv
import os
import yaml
from PRConfig.bandwidth_helpers import FileNameHelper
'''

    When running production-stream config, returns:

        Per line (in form of single HTML table):
            1. Inclusive retention
            2. Inclusive rate
            3. Exclusive retention
            4. Exclusive rate
            5. Average DstData bank size
            6. DstData bandwidth
            7. Average event size (all banks in particular stream)
            8. Bandwidth

        Per stream in Turbo/Full/Turcal
            1. Inclusive retention
            2. Inclusive rate
            3. Average DstData bank size
            4. DstData bandwidth
            5. Average event size (all banks in particular stream)
            6. Bandwidth

    When running wg-stream config, returns same figures as above (both per line and per stream)

'''

LHCb = GP.gbl.LHCb
RAW_BANK_TYPES = [(i, LHCb.RawBank.typeName(i))
                  for i in range(LHCb.RawBank.LastType)]


def parse_yaml(file_path):
    with open(os.path.expandvars(file_path), 'r') as f:
        return yaml.safe_load(f)


def rawbank_sizes(rawevent, lst):
    """Return (name, size) for each raw bank type."""
    if rawevent:

        def size(i):
            return sum(bank.totalSize() for bank in rawevent.banks(i))
    else:

        def size(i):
            return 0

    return [(name, size(i)) for i, name in lst]


def processing_events_per_line_and_stream(evt_max, lines, process):
    '''
        Returns, per line:
        i) How many events triggered on
        ii) How many unique events triggered on
        iii) Average DstData size of all events
        iv) Average size of all events

        Returns, per stream:
        i) How many events triggered on
        ii) Average DstData size of all events
        iii) Average size of all events
    '''

    # Per file (stream) information
    events_file = 0
    raw_size_all = 0
    dst_size_all = 0

    # Per line information
    # Stores how many events each line fired on
    event_stats = {
        line: []
        for line in [line + 'Decision' for line in list(lines)]
    }

    # Stores whole event size size
    raw = {line: 0 for line in [line + 'Decision' for line in list(lines)]}

    # Stores DstData bank size
    dst = {line: 0 for line in [line + 'Decision' for line in list(lines)]}

    exclusive = {}

    # Loop over all events
    analysed = 0
    while analysed < evt_max:
        analysed += 1

        exclusive.update({analysed: 0})

        # Run an event
        appMgr.run(1)
        report = evt[f'/Event/{process.capitalize()}/DecReports']
        rawevent = evt['/Event/DAQ/RawEvent']

        evtsize = sum(
            bank[1] for bank in rawbank_sizes(rawevent, RAW_BANK_TYPES))
        dstsize = sum(
            bank[1] for bank in rawbank_sizes(rawevent, [(60, 'DstData')]))

        # Will quit running if there are no more events in the input file
        if report:
            # Count per file/stream
            events_file += 1
            raw_size_all += evtsize
            dst_size_all += dstsize
            for line in event_stats.keys():
                # Count per line
                if report.decReport(line):
                    if report.decReport(line).decision() == 1:
                        event_stats[line].append(analysed)
                        exclusive[analysed] += 1
                        raw[line] += evtsize
                        dst[line] += dstsize
        else:
            break
    # First three variables per stream/file, last four for lines
    return events_file, raw_size_all, dst_size_all, event_stats, exclusive, raw, dst


def rates_per_line(event_stats, exclusive, raw, dst, input_rate,
                   output_file_path):

    data = []

    # Compute exclusive rate
    sort = dict(
        sorted(
            {k: v
             for (k, v) in exclusive.items() if v > 0}.items(),
            key=operator.itemgetter(1),
            reverse=True))

    unique_events = [key for key, value in sort.items() if value == 1]

    for line, val in event_stats.items():
        events_all = val + unique_events
        num_events = len(event_stats[line])
        row_values = (
            line,
            num_events / LHCbApp().EvtMax * 100
            if num_events else 0,  # Inclusive Retention (expressed as %)
            num_events / LHCbApp().EvtMax * input_rate
            if num_events else 0,  # Inclusive Rate (in kHz)
            len([
                key for key, value in Counter(events_all).items() if value > 1
            ]) / LHCbApp().EvtMax * 100
            if num_events else 0,  # Exclusive retention (expressed as %)
            len([
                key for key, value in Counter(events_all).items() if value > 1
            ]) / LHCbApp().EvtMax * input_rate
            if num_events else 0,  # Exclusive rate (in kHz)
            raw[line] / num_events * 1e-3
            if num_events else 0,  # Average event size (in kB)
            (num_events / LHCbApp().EvtMax * raw[line] / num_events) *
            input_rate / 1e6 if num_events else 0,  # Event bandwidth (in GB/s)
            dst[line] / len(event_stats[line]) * 1e-3
            if num_events else 0,  # Average DstData size (in kB)
            (num_events / LHCbApp().EvtMax * dst[line] / num_events) *
            input_rate / 1e6 if num_events else 0
        )  # DstData Bandwidth (in GB/s)

        data.append(row_values)

    with open(output_file_path, 'w') as f:
        csv_out = csv.writer(f)
        for tup in data:
            csv_out.writerow(tup)

    return


def rates_per_stream(events, raw_size, dst_size, streamname, input_rate,
                     output_file_path):

    data = []

    row_values = (
        streamname,
        events / LHCbApp().EvtMax * 100
        if events else 0,  # Inclusive Retention (expressed as %)
        events / LHCbApp().EvtMax * input_rate
        if events else 0,  # Inclusive Rate (in kHz)
        raw_size / events * 1e-3
        if events else 0,  # Average event size (in kB)
        (events / LHCbApp().EvtMax * raw_size / events) * input_rate / 1e6
        if events else 0,  # Event bandwidth (in GB/s)
        dst_size / events * 1e-3
        if events else 0,  # Average DstData size (in kB)
        (events / LHCbApp().EvtMax * dst_size / events) * input_rate / 1e6
        if events else 0)  # DstData Bandwidth (in GB/s)

    data.append(row_values)

    with open(output_file_path, 'w') as f:
        csv_out = csv.writer(f)
        for tup in data:
            csv_out.writerow(tup)

    return


if __name__ == '__main__':

    parser = argparse.ArgumentParser(description='Inspect Moore output')
    parser.add_argument(
        '-n',
        '--events',
        default=-1,
        type=lambda x: int(round(float(x))),
        help='nb of events to process',
        required=True)
    parser.add_argument(
        '-c',
        '--config',
        type=str,
        required=True,
        help='Path to yaml config file defining the input.')
    parser.add_argument('-s', '--stream', type=str, required=True)
    parser.add_argument(
        '-p',
        '--process',
        type=str,
        help='Compute for Hlt2 or Sprucing lines',
        choices=['hlt2', 'spruce'],
        required=True)
    parser.add_argument(
        '--stream-config',
        type=str,
        help='Choose production or per-WG stream configuration',
        choices=['production', 'wg'],
        required=True)
    args = parser.parse_args()

    fname_helper = FileNameHelper(args.process)

    n_events = args.events

    input_config = parse_yaml(args.config)

    if args.process == "spruce" and args.stream_config == "production":
        raise RuntimeError(
            '"production" stream config not defined for sprucing. Please use "wg".'
        )

    LHCbApp(
        DataType="Upgrade",
        Simulation=True,
        DDDBtag="dddb-20171126",
        CondDBtag="sim-20171127-vc-md100",
        EvtMax=n_events)
    EventSelector().PrintFreq = 10000
    IODataManager(DisablePFNWarning=True)

    # we have to configure the algorithms manually instead of `do_unpacking`
    # because we need to set `input_process='Hlt2'` in `unpack_rawevent`
    # to read MDF output from Sprucing
    algs = []
    unpack = unpack_rawevent(
        bank_types=['ODIN', 'HltDecReports', 'DstData', 'HltRoutingBits'],
        configurables=True)
    hlt2 = [hlt_decisions(source="Hlt2", output_loc="/Event/Hlt2/DecReports")]
    if args.process == 'spruce':
        spruce = [
            hlt_decisions(
                source="Spruce", output_loc="/Event/Spruce/DecReports")
        ]
    else:
        spruce = []
    decoder = decoder(input_process=args.process.capitalize())
    algs = [unpack] + hlt2 + spruce + [decoder] + [createODIN(ODIN='myODIN')]

    appMgr = ApplicationMgr(TopAlg=algs)
    appMgr.ExtSvc += [
        configured_ann_svc(json_file=fname_helper.tck(args.stream_config))
    ]

    IOHelper("MDF").inputFiles(
        [fname_helper.mdf_fname_for_reading(args.stream_config, args.stream)])

    with open(fname_helper.stream_config_json_path(args.stream_config)) as f:
        lines = json.load(f)[args.stream]

    appMgr = GP.AppMgr()
    evt = appMgr.evtsvc()

    # Calculates retention, rate and bandwidth per line and stream (file)
    evts_all, rawbanks_all, dst_all, event_stats, exclusive, raw, dst = processing_events_per_line_and_stream(
        LHCbApp().EvtMax, lines, args.process)
    rates_per_line(
        event_stats, exclusive, raw, dst, input_config['input_rate'],
        fname_helper.tmp_rate_table_per_line_path(args.stream_config,
                                                  args.stream))
    rates_per_stream(
        evts_all, rawbanks_all, dst_all, args.stream,
        input_config['input_rate'],
        fname_helper.tmp_rate_table_per_stream_path(args.stream_config,
                                                    args.stream))