File: //bin/pcp2spark
#!/usr/bin/pmpython
#
# Copyright (C) 2015-2019 Marko Myllynen <myllynen@redhat.com>
# Copyright (C) 2018 Red Hat.
#
# This program is free software; you can redistribute it and/or modify it
# under the terms of the GNU General Public License as published by the
# Free Software Foundation; either version 2 of the License, or (at your
# option) any later version.
#
# This program is distributed in the hope that it will be useful, but
# WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
# or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
# for more details.
# pylint: disable=superfluous-parens
# pylint: disable=invalid-name, line-too-long, no-self-use
# pylint: disable=too-many-boolean-expressions, too-many-statements
# pylint: disable=too-many-instance-attributes, too-many-locals
# pylint: disable=too-many-branches, too-many-nested-blocks
# pylint: disable=broad-except, consider-using-enumerate
""" PCP to Spark Bridge """
# Common imports
from collections import OrderedDict
import errno
import time
import sys
# Our imports
import socket
import json
# PCP Python PMAPI
from pcp import pmapi, pmconfig
from cpmapi import PM_CONTEXT_ARCHIVE, PM_IN_NULL, PM_DEBUG_APPL0, PM_DEBUG_APPL1
from cpmapi import PM_TIME_MSEC
if sys.version_info[0] >= 3:
long = int # pylint: disable=redefined-builtin
# Default config
DEFAULT_CONFIG = ["./pcp2spark.conf", "$HOME/.pcp2spark.conf", "$HOME/.pcp/pcp2spark.conf", "$PCP_SYSCONF_DIR/pcp2spark.conf"]
# Defaults
CONFVER = 1
SPARK_SERVER = "127.0.0.1"
SPARK_PORT = 44325
class PCP2Spark(object):
""" PCP to Spark """
def __init__(self):
""" Construct object, prepare for command line handling """
self.context = None
self.daemonize = 0
self.pmconfig = pmconfig.pmConfig(self)
self.opts = self.options()
# Configuration directives
self.keys = ('source', 'output', 'derived', 'header', 'globals',
'samples', 'interval', 'type', 'precision', 'daemonize',
'spark_server', 'spark_port',
'count_scale', 'space_scale', 'time_scale', 'version',
'count_scale_force', 'space_scale_force', 'time_scale_force',
'type_prefer', 'precision_force', 'limit_filter', 'limit_filter_force',
'live_filter', 'rank', 'invert_filter', 'predicate', 'names_change',
'speclocal', 'instances', 'ignore_incompat', 'ignore_unknown',
'omit_flat', 'include_labels')
# The order of preference for options (as present):
# 1 - command line options
# 2 - options from configuration file(s)
# 3 - built-in defaults defined below
self.check = 0
self.version = CONFVER
self.source = "local:"
self.output = None # For pmrep conf file compat only
self.speclocal = None
self.derived = None
self.header = 1
self.globals = 1
self.samples = None # forever
self.interval = pmapi.timeval(60) # 60 sec
self.opts.pmSetOptionInterval(str(60)) # 60 sec
self.delay = 0
self.type = 0
self.type_prefer = self.type
self.ignore_incompat = 0
self.ignore_unknown = 0
self.names_change = 0 # ignore
self.instances = []
self.live_filter = 0
self.rank = 0
self.limit_filter = 0
self.limit_filter_force = 0
self.invert_filter = 0
self.predicate = None
self.omit_flat = 0
self.include_labels = 0
self.precision = 3 # .3f
self.precision_force = None
self.timefmt = "%H:%M:%S" # For compat only
self.interpol = 0
self.count_scale = None
self.count_scale_force = None
self.space_scale = None
self.space_scale_force = None
self.time_scale = None
self.time_scale_force = None
self.spark_server = SPARK_SERVER
self.spark_port = SPARK_PORT
# Internal
self.runtime = -1
self.client_conn = None
# Performance metrics store
# key - metric name
# values - 0:txt label, 1:instance(s), 2:unit/scale, 3:type,
# 4:width, 5:pmfg item, 6:precision, 7:limit
self.metrics = OrderedDict()
self.pmfg = None
self.pmfg_ts = None
# Read configuration and prepare to connect
self.config = self.pmconfig.set_config_path(DEFAULT_CONFIG)
self.pmconfig.read_options()
self.pmconfig.read_cmd_line()
self.pmconfig.prepare_metrics()
self.pmconfig.set_signal_handler()
def options(self):
""" Setup default command line argument option handling """
opts = pmapi.pmOptions()
opts.pmSetOptionCallback(self.option)
opts.pmSetOverrideCallback(self.option_override)
opts.pmSetShortOptions("a:h:LK:c:Ce:D:V?HGA:S:T:O:s:t:rRIi:jJ:4:58:9:nN:vmP:0:q:b:y:Q:B:Y:g:p:")
opts.pmSetShortUsage("[option...] metricspec [...]")
opts.pmSetLongOptionHeader("General options")
opts.pmSetLongOptionArchive() # -a/--archive
opts.pmSetLongOptionArchiveFolio() # --archive-folio
opts.pmSetLongOptionContainer() # --container
opts.pmSetLongOptionHost() # -h/--host
opts.pmSetLongOptionLocalPMDA() # -L/--local-PMDA
opts.pmSetLongOptionSpecLocal() # -K/--spec-local
opts.pmSetLongOption("config", 1, "c", "FILE", "config file path")
opts.pmSetLongOption("check", 0, "C", "", "check config and metrics and exit")
opts.pmSetLongOption("derived", 1, "e", "FILE|DFNT", "derived metrics definitions")
opts.pmSetLongOption("daemonize", 0, "", "", "daemonize on startup")
opts.pmSetLongOptionDebug() # -D/--debug
opts.pmSetLongOptionVersion() # -V/--version
opts.pmSetLongOptionHelp() # -?/--help
opts.pmSetLongOptionHeader("Reporting options")
opts.pmSetLongOption("no-header", 0, "H", "", "omit headers")
opts.pmSetLongOption("no-globals", 0, "G", "", "omit global metrics")
opts.pmSetLongOptionAlign() # -A/--align
opts.pmSetLongOptionStart() # -S/--start
opts.pmSetLongOptionFinish() # -T/--finish
opts.pmSetLongOptionOrigin() # -O/--origin
opts.pmSetLongOptionSamples() # -s/--samples
opts.pmSetLongOptionInterval() # -t/--interval
opts.pmSetLongOption("raw", 0, "r", "", "output raw counter values (no rate conversion)")
opts.pmSetLongOption("raw-prefer", 0, "R", "", "prefer output raw counter values (no rate conversion)")
opts.pmSetLongOption("ignore-incompat", 0, "I", "", "ignore incompatible instances (default: abort)")
opts.pmSetLongOption("ignore-unknown", 0, "5", "", "ignore unknown metrics (default: abort)")
opts.pmSetLongOption("names-change", 1, "4", "ACTION", "update/ignore/abort on PMNS change (default: ignore)")
opts.pmSetLongOption("instances", 1, "i", "STR", "instances to report (default: all current)")
opts.pmSetLongOption("live-filter", 0, "j", "", "perform instance live filtering")
opts.pmSetLongOption("rank", 1, "J", "COUNT", "limit results to COUNT highest/lowest valued instances")
opts.pmSetLongOption("limit-filter", 1, "8", "LIMIT", "default limit for value filtering")
opts.pmSetLongOption("limit-filter-force", 1, "9", "LIMIT", "forced limit for value filtering")
opts.pmSetLongOption("invert-filter", 0, "n", "", "perform ranking before live filtering")
opts.pmSetLongOption("predicate", 1, "N", "METRIC", "set predicate filter reference metric")
opts.pmSetLongOption("omit-flat", 0, "v", "", "omit single-valued metrics")
opts.pmSetLongOption("include-labels", 0, "m", "", "incldue metric label info")
opts.pmSetLongOption("precision", 1, "P", "N", "prefer N digits after decimal separator (default: 3)")
opts.pmSetLongOption("precision-force", 1, "0", "N", "force N digits after decimal separator")
opts.pmSetLongOption("count-scale", 1, "q", "SCALE", "default count unit")
opts.pmSetLongOption("count-scale-force", 1, "Q", "SCALE", "forced count unit")
opts.pmSetLongOption("space-scale", 1, "b", "SCALE", "default space unit")
opts.pmSetLongOption("space-scale-force", 1, "B", "SCALE", "forced space unit")
opts.pmSetLongOption("time-scale", 1, "y", "SCALE", "default time unit")
opts.pmSetLongOption("time-scale-force", 1, "Y", "SCALE", "forced time unit")
opts.pmSetLongOption("spark-server", 1, "g", "SERVER", "Spark server (default: " + SPARK_SERVER + ")")
opts.pmSetLongOption("spark-port", 1, "p", "PORT", "Spark port (default: " + str(SPARK_PORT) + ")")
return opts
def option_override(self, opt):
""" Override standard PCP options """
if opt in ('g', 'H', 'K', 'n', 'N', 'p'):
return 1
return 0
def option(self, opt, optarg, _index):
""" Perform setup for individual command line option """
if opt == 'daemonize':
self.daemonize = 1
elif opt == 'K':
if not self.speclocal or not self.speclocal.startswith(";"):
self.speclocal = ";" + optarg
else:
self.speclocal = self.speclocal + ";" + optarg
elif opt == 'c':
self.config = optarg
elif opt == 'C':
self.check = 1
elif opt == 'e':
if not self.derived or not self.derived.startswith(";"):
self.derived = ";" + optarg
else:
self.derived = self.derived + ";" + optarg
elif opt == 'H':
self.header = 0
elif opt == 'G':
self.globals = 0
elif opt == 'r':
self.type = 1
elif opt == 'R':
self.type_prefer = 1
elif opt == 'I':
self.ignore_incompat = 1
elif opt == '5':
self.ignore_unknown = 1
elif opt == '4':
if optarg == 'ignore':
self.names_change = 0
elif optarg == 'abort':
self.names_change = 1
elif optarg == 'update':
self.names_change = 2
else:
sys.stderr.write("Unknown names-change action '%s' specified.\n" % optarg)
sys.exit(1)
elif opt == 'i':
self.instances = self.instances + self.pmconfig.parse_instances(optarg)
elif opt == 'j':
self.live_filter = 1
elif opt == 'J':
self.rank = optarg
elif opt == '8':
self.limit_filter = optarg
elif opt == '9':
self.limit_filter_force = optarg
elif opt == 'n':
self.invert_filter = 1
elif opt == 'N':
self.predicate = optarg
elif opt == 'v':
self.omit_flat = 1
elif opt == 'm':
self.include_labels = 1
elif opt == 'P':
self.precision = optarg
elif opt == '0':
self.precision_force = optarg
elif opt == 'q':
self.count_scale = optarg
elif opt == 'Q':
self.count_scale_force = optarg
elif opt == 'b':
self.space_scale = optarg
elif opt == 'B':
self.space_scale_force = optarg
elif opt == 'y':
self.time_scale = optarg
elif opt == 'Y':
self.time_scale_force = optarg
elif opt == 'g':
self.spark_server = optarg
elif opt == 'p':
self.spark_port = int(optarg)
else:
raise pmapi.pmUsageErr()
def connect(self):
""" Establish PMAPI context """
context, self.source = pmapi.pmContext.set_connect_options(self.opts, self.source, self.speclocal)
self.pmfg = pmapi.fetchgroup(context, self.source)
self.pmfg_ts = self.pmfg.extend_timestamp()
self.context = self.pmfg.get_context()
if pmapi.c_api.pmSetContextOptions(self.context.ctx, self.opts.mode, self.opts.delta):
raise pmapi.pmUsageErr()
def validate_config(self):
""" Validate configuration options """
if self.version != CONFVER:
sys.stderr.write("Incompatible configuration file version (read v%s, need v%d).\n" % (self.version, CONFVER))
sys.exit(1)
self.pmconfig.validate_common_options()
self.pmconfig.validate_metrics(curr_insts=not self.live_filter)
self.pmconfig.finalize_options()
def execute(self):
""" Fetch and report """
# Debug
if self.context.pmDebug(PM_DEBUG_APPL1):
sys.stdout.write("Known config file keywords: " + str(self.keys) + "\n")
sys.stdout.write("Known metric spec keywords: " + str(self.pmconfig.metricspec) + "\n")
# Set delay mode, interpolation
if self.context.type != PM_CONTEXT_ARCHIVE:
self.delay = 1
self.interpol = 1
# Common preparations
self.context.prepare_execute(self.opts, False, self.interpol, self.interval)
# Setup a socket to send data to Spark
spark_socket = socket.socket(socket.AF_INET, socket.SOCK_STREAM)
spark_socket.bind((self.spark_server, self.spark_port))
spark_socket.listen(0) # Don't handle listen backlog, only allow single connection
# Wait for a connection from the Spark worker
# job and when we see it accept the connection
sys.stdout.write("Awaiting connection...\n")
self.client_conn, _ = spark_socket.accept()
# Headers
if self.header == 1:
self.header = 0
self.write_header()
# Just checking
if self.check == 1:
return
# Daemonize when requested
if self.daemonize == 1:
self.opts.daemonize()
# Align poll interval to host clock
if self.context.type != PM_CONTEXT_ARCHIVE and self.opts.pmGetOptionAlignment():
align = float(self.opts.pmGetOptionAlignment()) - (time.time() % float(self.opts.pmGetOptionAlignment()))
time.sleep(align)
# Main loop
refresh_metrics = 0
while self.samples != 0:
# Refresh metrics as needed
if refresh_metrics:
refresh_metrics = 0
self.pmconfig.update_metrics(curr_insts=not self.live_filter)
# Fetch values
refresh_metrics = self.pmconfig.fetch()
if refresh_metrics < 0:
break
# Report and prepare for the next round
self.report(self.pmfg_ts())
if self.samples and self.samples > 0:
self.samples -= 1
if self.delay and self.interpol and self.samples != 0:
self.pmconfig.pause()
# Allow to flush buffered values / say goodbye
self.report(None)
def report(self, tstamp):
""" Report metric values """
if tstamp is not None:
tstamp = tstamp.strftime(self.timefmt)
self.write_spark(tstamp)
def write_header(self):
""" Write info header """
if self.context.type == PM_CONTEXT_ARCHIVE:
sys.stdout.write("Sending %d archived metrics to Spark %s:%d...\n(Ctrl-C to stop)\n" % (len(self.metrics), self.spark_server, self.spark_port))
return
sys.stdout.write("Sending %d metrics to Spark %s:%d every %.1f sec" % (len(self.metrics), self.spark_server, self.spark_port, float(self.interval)))
if self.runtime != -1:
sys.stdout.write(":\n%s samples(s) with %.1f sec interval ~ %d sec runtime.\n" % (self.samples, float(self.interval), self.runtime))
elif self.samples:
duration = (self.samples - 1) * float(self.interval)
sys.stdout.write(":\n%s samples(s) with %.1f sec interval ~ %d sec runtime.\n" % (self.samples, float(self.interval), duration))
else:
sys.stdout.write("...\n(Ctrl-C to stop)\n")
def write_spark(self, timestamp):
""" Write (send) metrics to socket for Spark server to pick up """
if timestamp is None:
# Silent goodbye, close in finalize()
return
ts = self.context.datetime_to_secs(self.pmfg_ts(), PM_TIME_MSEC)
# Assemble all metrics into a single document
# Use @-prefixed keys for metadata not coming in from PCP metrics
host = self.context.pmGetContextHostName()
data = {'@host-id': host, '@timestamp': long(ts)}
insts_key = "@instances"
inst_key = "@id"
labels_key = "@labels"
results = self.pmconfig.get_ranked_results(valid_only=True)
for metric in results:
# Install value into outgoing json/dict in key1{key2{key3=value}} style:
# foo.bar.baz=value => foo: { bar: { baz: value ...} }
# foo.bar.noo[0]=value => foo: { bar: { @instances:[{@id: 0, noo: value ...} ... ]}
# If include_labels is selected:
# foo.bar.baz=value labels => foo: { bar: { baz: value ... labels ...} }
# foo.bar.noo[0]=value labels=labels => foo: { bar: { @instances:[{@id: 0, noo: value ...., @labels: labels ...]}
pmns_parts = metric.split(".")
# Find/create the parent dictionary into which to insert the final component
for inst, name, value in results[metric]:
labels = None
if self.include_labels:
labels = self.pmconfig.get_labels_str(metric, inst)
value = round(value, self.metrics[metric][6]) if isinstance(value, float) else value
pmns_leaf_dict = data
for pmns_part in pmns_parts[:-1]:
if pmns_part not in pmns_leaf_dict:
pmns_leaf_dict[pmns_part] = {}
pmns_leaf_dict = pmns_leaf_dict[pmns_part]
last_part = pmns_parts[-1]
if inst == PM_IN_NULL:
if self.include_labels:
value = value, labels
pmns_leaf_dict[last_part] = value
else:
if insts_key not in pmns_leaf_dict:
pmns_leaf_dict[insts_key] = []
insts = pmns_leaf_dict[insts_key]
# Find a preexisting {@id: name} object in there, if any
found = False
for j in range(0, len(insts)):
if insts[j][inst_key] == name:
insts[j][last_part] = value
if self.include_labels:
insts[j][labels_key] = labels
found = True
if not found:
if self.include_labels:
insts.append({inst_key: name, last_part: value, labels_key: labels})
else:
insts.append({inst_key: name, last_part: value})
# Python 3 requires us to encode our data before we can send
# Also send a trailing newline between records
message = json.dumps(data,
sort_keys=True,
ensure_ascii=False,
separators=(',', ': '))
if self.context.pmDebug(PM_DEBUG_APPL0):
print(message)
try:
self.client_conn.sendall(message.encode('utf-8'))
self.client_conn.sendall("\n".encode('utf-8'))
except socket.timeout as err:
sys.stderr.write("Spark connection timed out: %s\n" % str(err))
sys.exit(1)
except socket.error as err:
sys.stderr.write("Socket error: %s\n" % str(err))
sys.exit(1)
def finalize(self):
""" Finalize and clean up """
if self.client_conn:
try:
self.client_conn.close()
except Exception:
pass
self.client_conn = None
if __name__ == '__main__':
try:
P = PCP2Spark()
P.connect()
P.validate_config()
P.execute()
P.finalize()
except pmapi.pmErr as error:
sys.stderr.write("%s: %s" % (error.progname(), error.message()))
if error.message() == "Connection refused":
sys.stderr.write("; is pmcd running?")
sys.stderr.write("\n")
sys.exit(1)
except pmapi.pmUsageErr as usage:
usage.message()
sys.exit(1)
except IOError as error:
if error.errno != errno.EPIPE:
sys.stderr.write("%s\n" % str(error))
sys.exit(1)
except KeyboardInterrupt:
sys.stdout.write("\n")
P.finalize()