2020-12-22 17:51:37 -06:00
|
|
|
#!/usr/bin/python
|
|
|
|
######################################################################
|
|
|
|
#
|
|
|
|
# Example parser for the JSON format history stats output of grpcurl
|
|
|
|
# for the gRPC service provided on a Starlink user terminal.
|
|
|
|
#
|
|
|
|
# Expects input as from the following command:
|
|
|
|
# grpcurl -plaintext -d {\"get_history\":{}} 192.168.100.1:9200 SpaceX.API.Device.Device/Handle
|
|
|
|
#
|
|
|
|
# This script examines the most recent samples from the history data
|
|
|
|
# and computes several different metrics related to packet loss. By
|
|
|
|
# default, it will print the results in CSV format.
|
|
|
|
#
|
|
|
|
######################################################################
|
|
|
|
|
|
|
|
import json
|
|
|
|
import datetime
|
|
|
|
import sys
|
|
|
|
import getopt
|
|
|
|
|
|
|
|
from itertools import chain
|
|
|
|
|
|
|
|
fArgError = False
|
|
|
|
|
|
|
|
try:
|
2020-12-24 16:58:28 -06:00
|
|
|
opts, args = getopt.getopt(sys.argv[1:], "ahrs:vH")
|
2020-12-22 17:51:37 -06:00
|
|
|
except getopt.GetoptError as err:
|
|
|
|
print(str(err))
|
|
|
|
fArgError = True
|
|
|
|
|
|
|
|
# Default to 1 hour worth of data samples.
|
|
|
|
parseSamples = 3600
|
|
|
|
fUsage = False
|
|
|
|
fVerbose = False
|
|
|
|
fParseAll = False
|
|
|
|
fHeader = False
|
2020-12-24 16:58:28 -06:00
|
|
|
fRunLengths = False
|
2020-12-22 17:51:37 -06:00
|
|
|
|
|
|
|
if not fArgError:
|
|
|
|
if len(args) > 1:
|
|
|
|
fArgError = True
|
|
|
|
else:
|
|
|
|
for opt, arg in opts:
|
|
|
|
if opt == "-a":
|
|
|
|
fParseAll = True
|
|
|
|
elif opt == "-h":
|
|
|
|
fUsage = True
|
2020-12-24 16:58:28 -06:00
|
|
|
elif opt == "-r":
|
|
|
|
fRunLengths = True
|
2020-12-22 17:51:37 -06:00
|
|
|
elif opt == "-s":
|
|
|
|
parseSamples = int(arg)
|
|
|
|
elif opt == "-v":
|
|
|
|
fVerbose = True
|
|
|
|
elif opt == "-H":
|
|
|
|
fHeader = True
|
|
|
|
|
|
|
|
if fUsage or fArgError:
|
|
|
|
print("Usage: "+sys.argv[0]+" [options...] [<file>]")
|
|
|
|
print(" where <file> is the file to parse, default: stdin")
|
|
|
|
print("Options:")
|
|
|
|
print(" -a: Parse all valid samples")
|
|
|
|
print(" -h: Be helpful")
|
2020-12-24 16:58:28 -06:00
|
|
|
print(" -r: Include ping drop run length stats")
|
2020-12-22 17:51:37 -06:00
|
|
|
print(" -s <num>: Parse <num> data samples, default: "+str(parseSamples))
|
|
|
|
print(" -v: Be verbose")
|
|
|
|
print(" -H: print CSV header instead of parsing file")
|
|
|
|
sys.exit(1 if fArgError else 0)
|
|
|
|
|
|
|
|
if fHeader:
|
2020-12-24 16:58:28 -06:00
|
|
|
header = "datetimestamp_utc,samples,total_ping_drop,count_full_ping_drop,count_obstructed,total_obstructed_ping_drop,count_full_obstructed_ping_drop,count_unscheduled,total_unscheduled_ping_drop,count_full_unscheduled_ping_drop"
|
|
|
|
if fRunLengths:
|
|
|
|
header+= ",init_run_fragment,final_run_fragment,"
|
|
|
|
header += ",".join("run_seconds_" + str(x) for x in range(1, 61)) + ","
|
|
|
|
header += ",".join("run_minutes_" + str(x) for x in range(1, 60))
|
|
|
|
header += ",run_minutes_60_or_greater"
|
|
|
|
print(header)
|
2020-12-22 17:51:37 -06:00
|
|
|
sys.exit(0)
|
|
|
|
|
|
|
|
# Allow "-" to be specified as file for stdin.
|
|
|
|
if len(args) == 0 or args[0] == "-":
|
|
|
|
jsonData = json.load(sys.stdin)
|
|
|
|
else:
|
|
|
|
jsonFile = open(args[0])
|
|
|
|
jsonData = json.load(jsonFile)
|
|
|
|
jsonFile.close()
|
|
|
|
|
2020-12-24 16:58:28 -06:00
|
|
|
timestamp = datetime.datetime.utcnow()
|
|
|
|
|
2020-12-22 17:51:37 -06:00
|
|
|
historyData = jsonData['dishGetHistory']
|
|
|
|
|
|
|
|
# 'current' is the count of data samples written to the ring buffer,
|
|
|
|
# irrespective of buffer wrap.
|
|
|
|
current = int(historyData['current'])
|
|
|
|
nSamples = len(historyData['popPingDropRate'])
|
|
|
|
|
|
|
|
if fVerbose:
|
|
|
|
print("current: " + str(current))
|
|
|
|
print("All samples: " + str(nSamples))
|
|
|
|
|
|
|
|
nSamples = min(nSamples,current)
|
|
|
|
|
|
|
|
if fVerbose:
|
|
|
|
print("Valid samples: " + str(nSamples))
|
|
|
|
|
|
|
|
# This is ring buffer offset, so both index to oldest data sample and
|
|
|
|
# index to next data sample after the newest one.
|
|
|
|
offset = current % nSamples
|
|
|
|
|
|
|
|
tot = 0
|
|
|
|
totOne = 0
|
|
|
|
totUnsched = 0
|
|
|
|
totUnschedD = 0
|
|
|
|
totUnschedOne = 0
|
|
|
|
totObstruct = 0
|
|
|
|
totObstructD = 0
|
|
|
|
totObstructOne = 0
|
|
|
|
|
2020-12-24 16:58:28 -06:00
|
|
|
secondRuns = [0] * 60
|
|
|
|
minuteRuns = [0] * 60
|
|
|
|
runLength = 0
|
|
|
|
initRun = None
|
|
|
|
|
2020-12-22 17:51:37 -06:00
|
|
|
if fParseAll or nSamples < parseSamples:
|
|
|
|
parseSamples = nSamples
|
|
|
|
|
|
|
|
# Parse the most recent parseSamples-sized set of samples. This will
|
|
|
|
# iterate samples in order from oldest to newest, although that's not
|
|
|
|
# actually required for the current set of stats being computed below.
|
|
|
|
if parseSamples <= offset:
|
|
|
|
sampleRange = range(offset - parseSamples, offset)
|
|
|
|
else:
|
|
|
|
sampleRange = chain(range(nSamples + offset - parseSamples, nSamples), range(0, offset))
|
|
|
|
|
|
|
|
for i in sampleRange:
|
|
|
|
d = historyData["popPingDropRate"][i]
|
|
|
|
tot += d
|
|
|
|
if d >= 1:
|
|
|
|
totOne += d
|
2020-12-24 16:58:28 -06:00
|
|
|
runLength += 1
|
|
|
|
elif runLength > 0:
|
|
|
|
if initRun is None:
|
|
|
|
initRun = runLength
|
|
|
|
else:
|
|
|
|
if runLength <= 60:
|
|
|
|
secondRuns[runLength-1] += runLength
|
|
|
|
else:
|
|
|
|
minuteRuns[min((runLength-1)//60-1, 59)] += runLength
|
|
|
|
runLength = 0
|
|
|
|
elif initRun is None:
|
|
|
|
initRun = 0
|
2020-12-22 17:51:37 -06:00
|
|
|
if not historyData["scheduled"][i]:
|
|
|
|
totUnsched += 1
|
|
|
|
totUnschedD += d
|
|
|
|
if d >= 1:
|
|
|
|
totUnschedOne += d
|
|
|
|
if historyData["obstructed"][i]:
|
|
|
|
totObstruct += 1
|
|
|
|
totObstructD += d
|
|
|
|
if d >= 1:
|
|
|
|
totObstructOne += d
|
|
|
|
|
2020-12-24 16:58:28 -06:00
|
|
|
# If the entire sample set is one big drop run, it will be both initial
|
|
|
|
# fragment (continued from prior sample range) and final one (continued
|
|
|
|
# to next sample range), but to avoid double-reporting, just call it
|
|
|
|
# the initial run.
|
|
|
|
if initRun is None:
|
|
|
|
initRun = runLength
|
|
|
|
runLength = 0
|
|
|
|
|
2020-12-22 17:51:37 -06:00
|
|
|
if fVerbose:
|
|
|
|
print("Parsed samples: " + str(parseSamples))
|
|
|
|
print("Total ping drop: " + str(tot))
|
|
|
|
print("Count of drop == 1: " + str(totOne))
|
|
|
|
print("Obstructed: " + str(totObstruct))
|
|
|
|
print("Obstructed ping drop: " + str(totObstructD))
|
|
|
|
print("Obstructed drop == 1: " + str(totObstructOne))
|
|
|
|
print("Unscheduled: " + str(totUnsched))
|
|
|
|
print("Unscheduled ping drop: " + str(totUnschedD))
|
|
|
|
print("Unscheduled drop == 1: " + str(totUnschedOne))
|
2020-12-24 16:58:28 -06:00
|
|
|
if fRunLengths:
|
|
|
|
print("Initial drop run fragment: " + str(initRun))
|
|
|
|
print("Final drop run fragment: " + str(runLength))
|
|
|
|
print("Per-second drop runs: " + ", ".join(str(x) for x in secondRuns))
|
|
|
|
print("Per-minute drop runs: " + ", ".join(str(x) for x in minuteRuns))
|
2020-12-22 17:51:37 -06:00
|
|
|
else:
|
|
|
|
# NOTE: When changing data output format, also change the -H header printing above.
|
2020-12-24 16:58:28 -06:00
|
|
|
csvData = timestamp.replace(microsecond=0).isoformat() + "," + ",".join(str(x) for x in [parseSamples, tot, totOne, totObstruct, totObstructD, totObstructOne, totUnsched, totUnschedD, totUnschedOne])
|
|
|
|
if fRunLengths:
|
|
|
|
csvData += "," + ",".join(str(x) for x in chain([initRun, runLength], secondRuns, minuteRuns))
|
|
|
|
print(csvData)
|