bachelorThesis/figure_8/figure8.py

275 lines
10 KiB
Python
Executable file

#!/usr/bin/env python3
# coding: utf-8
import os
import json
import pandas as pd
import findspark
findspark.init()
import pyspark
import pyspark.sql
import sys
import gzip
from pyspark import AccumulatorParam
from pyspark.sql.functions import lit
from pyspark.sql import Window
from pyspark.sql.types import *
from decimal import *
TESTDATA = True
if len(sys.argv) is not 4:
print(sys.argv[0] + " {cluster} {tmpdir} {maxram}")
sys.exit()
cluster=sys.argv[1]
spark = pyspark.sql.SparkSession.builder \
.appName("task_slowdown") \
.config("spark.driver.maxResultSize", "128g") \
.config("spark.local.dir", sys.argv[2]) \
.config("spark.driver.memory", sys.argv[3]) \
.getOrCreate()
sc = spark.sparkContext
#
# READING INSTANCE EVENTS DATA
#
dfepath = "/home/claudio/google_2019/instance_events/" + cluster
dfepath += "/" + cluster + ("_instance_events00000000000?.json.gz" if TESTDATA else "_instance_events*.json.gz")
dfe = spark.read.json(dfepath)
try:
dfe["collection_type"] = dfe["collection_type"].cast(ByteType())
except:
dfe = dfe.withColumn("collection_type", lit(None).cast(ByteType()))
#
# READING INSTANCE USAGE DATA
#
dfupath = "/home/claudio/google_2019/instance_usage/" + cluster
dfupath += "/" + cluster + ("_instance_usage00000000000?.csv.gz" if TESTDATA else "_instance_usage*.csv.gz")
usage_schema = StructType() \
.add("start_time", LongType(), True) \
.add("end_time", LongType(), True) \
.add("collection_id", StringType(), True) \
.add("instance_index", StringType(), True) \
.add("machine_id", StringType(), True) \
.add("alloc_collection_id", LongType(), True) \
.add("alloc_instance_index", StringType(), True) \
.add("collection_type", ByteType(), True) \
.add("average_usage_cpus", DoubleType(), True) \
.add("average_usage_memory", DoubleType(), True) \
.add("maximum_usage_cpus", DoubleType(), True) \
.add("maximum_usage_memory", DoubleType(), True) \
.add("random_sample_usage_cpus", DoubleType(), True) \
.add("random_sample_usage_memory", DoubleType(), True) \
.add("assigned_memory", DoubleType(), True) \
.add("page_cache_memory", DoubleType(), True) \
.add("cycles_per_instruction", DoubleType(), True) \
.add("memory_accLesses_per_instruction", DoubleType(), True) \
.add("sample_rate", DoubleType(), True) \
.add("cpu_usage_dist_00", DoubleType(), True) \
.add("cpu_usage_dist_10", DoubleType(), True) \
.add("cpu_usage_dist_20", DoubleType(), True) \
.add("cpu_usage_dist_30", DoubleType(), True) \
.add("cpu_usage_dist_40", DoubleType(), True) \
.add("cpu_usage_dist_50", DoubleType(), True) \
.add("cpu_usage_dist_60", DoubleType(), True) \
.add("cpu_usage_dist_70", DoubleType(), True) \
.add("cpu_usage_dist_80", DoubleType(), True) \
.add("cpu_usage_dist_90", DoubleType(), True) \
.add("cpu_usage_dist_91", DoubleType(), True) \
.add("cpu_usage_dist_92", DoubleType(), True) \
.add("cpu_usage_dist_93", DoubleType(), True) \
.add("cpu_usage_dist_94", DoubleType(), True) \
.add("cpu_usage_dist_95", DoubleType(), True) \
.add("cpu_usage_dist_96", DoubleType(), True) \
.add("cpu_usage_dist_97", DoubleType(), True) \
.add("cpu_usage_dist_98", DoubleType(), True) \
.add("cpu_usage_dist_99", DoubleType(), True)
dfu = spark.read.format("csv") \
.option("header", False) \
.schema(usage_schema) \
.load(dfupath)
# READING MACHINE EVENTS DATA, sort them and save them as broadcast variable
print("Starting to read machine events...")
dfm = pd.read_csv("~/google_2019/machine_events/" + cluster + "_machine_events.csv")
print("Dropping remove events...")
dfm = dfm[(dfm.type==1)|(dfm.type==3)]
print("Dropping missing data events...")
dfm = dfm[dfm.missing_data_reason.notnull()]
print("Projecting on useful columns...")
dfm = dfm[["time", "machine_id", "capacity.cpus", "capacity.memory"]]
print("Sorting by time...")
dfm = dfm.sort_values(by=["machine_id", "time"])
print("Converting to broadcast variable...")
dfm = sc.broadcast([tuple(r) for r in dfm.to_numpy()])
print("Done with machine events.")
def tabid(x):
return Decimal(x.collection_id) + Decimal(x.instance_index) / Decimal(2**64)
# interpolate machine data by extending each last machine report before a gap to cover it
def clean_usage(x):
return [tabid(x), Decimal(x.average_usage_cpus), Decimal(x.average_usage_memory), int(x.start_time),
int(x.end_time), x.machine_id]
def interpolate_usage(ts):
ts = sorted(ts, key=lambda x: x[3])
l = len(ts)
for i in range(1, l-1):
if ts[i+1][3] > ts[i][4]:
ts[i][4] = ts[i+1][3]
return ts
dfu = dfu.rdd \
.filter(lambda x: x.start_time is not None and x.end_time is not None and
x.instance_index is not None and x.collection_id is not None and
x.machine_id is not None) \
.map(clean_usage).groupBy(lambda x: x[0]) \
.flatMap(lambda x: interpolate_usage(x[1])).toDF(["id", "acpu", "aram", "start", "end", "mid"])
dfe = dfe.rdd \
.filter(lambda x: x.time is not None and x.type is not None and x.machine_id is not None and
x.instance_index is not None and x.collection_id is not None and x.resource_request is not None and
x.resource_request.cpus is not None and x.resource_request.memory is not None) \
.map(lambda x: [tabid(x), int(x.time), int(x.type),
Decimal(x.resource_request.cpus), Decimal(x.resource_request.memory), x.machine_id]) \
.toDF(["id", "time", "type", "rcpu", "rram", "mid"])
df = dfe.join(dfu, [dfe.id == dfu.id, dfe.mid == dfu.mid, dfe.time >= dfu.start, dfe.time < dfu.end])
def get_machine_time_resources(machine_id):
def aux(i, j):
mid = (i + j) // 2
print(i, j, mid)
print(dfm.value[mid])
if dfm.value[mid][1] > machine_id:
return aux(i, mid - 1)
elif dfm.value[mid][1] < machine_id:
return aux(mid + 1, j)
else:
start = mid
while dfm.value[start][1] == machine_id and start >= 0: # once found, search for oldest log for machine
start -= 1
start += 1
end = mid
while dfm.value[end][1] == machine_id and end < len(dfm.value):
end += 1
return dfm.value[start:end]
return aux(0, len(dfm.value)-1)
def increment_reserv_bucket(bucket, ceils, taskid, reserv):
idx = 0
while ceils[idx] < reserv and idx < len(ceils):
idx += 1
if taskid not in bucket:
bucket[taskid] = [0] * (len(ceils) + 1)
bucket[taskid][idx] += 1
def bucket_sum_per_termination(bucket, last_term_by_id):
result = {-1: None, 4: None, 5: None, 6: None, 7: None, 8: None}
for tid, vs in bucket.items():
term = last_term_by_id[tid]
if result[term] is None:
result[term] = vs
else:
result[term] = [sum(x) for x in zip(result[term], vs)]
return result
def for_each_joined(x):
machine_id = x[0]
ts = x[1]
ts = sorted(ts, key=lambda x: x.time)
last_req_by_id = {} # map taskid -> last known req [cpu, ram] (data removed when task terminates)
reserv_ceils = [0.2, 0.4, 0.6, 0.8, 1]
cpu_reservs_by_id = {} # map taskid -> [a, b, c, d, e, f] where:
# a: count of event with res. reserv. <0.2
# b: count of event with res. reserv. [0.2, 0.4)
# c: count of event with res. reserv. [0.4, 0.6)
# d: count of event with res. reserv. [0.6, 0.8)
# e: count of event with res. reserv. [0.8, 0.1)
# f: count of event with res. reserv. >=1
ram_reservs_by_id = {}
request_ceils = [0.025, 0.05, 0.075]
cpu_request_by_id = {} # map taskid -> [a, b, c, d] where <0.025, [0.025, 0.05), [0.05,0.075), >=0.075
ram_request_by_id = {}
util_ceils = reserv_ceils
cpu_util_by_id = {}
ram_util_by_id = {}
last_term_by_id = {} # map taskid -> last termination
machine_logs = get_machine_time_resources(machine_id)
for i, t in enumerate(ts):
if len(machine_logs) > 1 and machine_logs[1][0] >= t.time:
machine_logs.pop(0)
if t.id not in last_term_by_id:
last_term_by_id[t.id] = -1
if t.type >= 4 and t.type <= 8:
last_term_by_id[t.id] = t.type
del last_req_by_id[t.id]
else:
if t.rcpu is not None and t.rram is not None:
last_req_by_id[t.id] = (t.rcpu, t.rram)
# 8b
tot_req = [sum(x) for x in zip(*last_req_by_id.values())]
reserv_cpu = tot_req[0] / machine_logs[0][2]
reserv_ram = tot_req[1] / machine_logs[0][3]
increment_reserv_bucket(cpu_reservs_by_id, reserv_ceils, t.id, reserv_cpu)
increment_reserv_bucket(ram_reservs_by_id, reserv_ceils, t.id, reserv_ram)
# 8a
increment_reserv_bucket(cpu_request_by_id, request_ceils, t.id, t.rcpu)
increment_reserv_bucket(ram_request_by_id, request_ceils, t.id, t.rram)
# 8c
increment_reserv_bucket(cpu_util_by_id, util_ceils, t.id, t.acpu)
increment_reserv_bucket(ram_util_by_id, util_ceils, t.id, t.aram)
resobj = {'rcpu': cpu_request_by_id, 'rram': ram_request_by_id, 'rscpu': cpu_reservs_by_id,
'rsram': ram_reservs_by_id, 'ucpu': cpu_util_by_id, 'uram': ram_util_by_id}
for k, v in resobj.items():
resobj[k] = bucket_sum_per_termination(v, last_term_by_id)
return resobj
def fold_resobjs(ro1, ro2):
if ro1 is None:
return ro2
elif ro2 is None:
return ro1
else:
for k in ro1.keys():
for kk in ro1[k].keys():
if ro1[k][kk] is None:
ro1[k][kk] = ro2[k][kk]
elif ro2[k][kk] is None:
continue
else:
ro1[k][kk] = [sum(x) for x in zip(ro1[k][kk], ro2[k][kk])]
return ro1
# TODO: partition by id and in the for-each-row
# function implement lookup to dfm.value to understand its memory capacity
result = df.rdd \
.groupBy(lambda x: x.mid) \
.map(for_each_joined) \
.fold(None, fold_resobjs)
d = os.path.dirname(os.path.realpath(__file__))
with open(d + "/" + cluster + "_figure8.json", "w") as f:
json.dump(result, f)
# vim: set ts=4 sw=4 et tw=120: