mirror of
https://github.com/alibaba/arthas.git
synced 2024-04-21 10:21:39 +00:00
519 lines
17 KiB
Python
Executable File
519 lines
17 KiB
Python
Executable File
#!/usr/bin/env python3
|
|
|
|
import argparse
|
|
import requests
|
|
from datetime import datetime
|
|
|
|
def str2bool(v):
|
|
if isinstance(v, bool):
|
|
return v
|
|
if v.lower() in ('yes', 'true', 't', 'y', '1'):
|
|
return True
|
|
elif v.lower() in ('no', 'false', 'f', 'n', '0'):
|
|
return False
|
|
else:
|
|
raise argparse.ArgumentTypeError('Boolean value expected.')
|
|
|
|
parser = argparse.ArgumentParser(description='Trace step by step, find slow method tree.')
|
|
parser.add_argument('--host', help='Arthas server host (default: 127.0.0.1:8563)', default="127.0.0.1:8563")
|
|
parser.add_argument('--times', '-n', help='Max trace times of every round (default: 100)', type=int, default=100)
|
|
parser.add_argument('--method-path', '-m',
|
|
help='Exact method path to trace, eg: "className1::methodName1[,className2::methodName2]"',
|
|
required=True)
|
|
parser.add_argument('--addit-method',
|
|
help='Additional methods to trace, but not in the method path, '
|
|
'eg: "className1::methodName1[,className2::methodName2]"')
|
|
# parser.add_argument('--condition', '-c', help='condition express', default='')
|
|
parser.add_argument('--min-cost', help='min cost in condition: #cost > min_cost', type=int, default=0)
|
|
# TODO filter by min cost of specify method
|
|
parser.add_argument('--timeout', '-t', help='request timeout(ms) (default:30000)', type=int, default=30000)
|
|
parser.add_argument('--skip-jdk-method', help='skip jdk method trace, (default:True)', type=str2bool, default=True)
|
|
parser.add_argument('--reset-on-start', help='reset classes once on start (default:True)', type=str2bool, default=True)
|
|
parser.add_argument('--reset-on-round', help='reset classes on every round (default:False)', type=str2bool, default=False)
|
|
parser.add_argument('--stop-match-times',
|
|
help='If primary call tree matching times was exceeded, assuming no new call tree can be found (default: 10)',
|
|
type=int, default=10)
|
|
args = parser.parse_args()
|
|
# print args
|
|
print(args)
|
|
|
|
url = 'http://' + args.host + "/api"
|
|
min_cost = args.min_cost
|
|
if min_cost > 0:
|
|
condition = '"#cost > %s"' % (min_cost)
|
|
else:
|
|
condition = ''
|
|
trace_times = args.times
|
|
timeout = args.timeout
|
|
stop_match_times = args.stop_match_times
|
|
is_skip_jdk_method = args.skip_jdk_method
|
|
|
|
|
|
# trace method path, 方法的顺序与调用树一致
|
|
# {
|
|
# className: xxx,
|
|
# methodName: yyy
|
|
# }
|
|
trace_method_path = []
|
|
trace_method_path_code = 0
|
|
|
|
# Partial matching method paths
|
|
# [[{},{}], [path]]
|
|
partial_matching_method_paths = []
|
|
|
|
# additional trace methods, not in trace tree
|
|
additional_methods = []
|
|
|
|
call_tree_match_times = 0
|
|
|
|
# call stack tree
|
|
# {
|
|
# className: xxx
|
|
# methodName: xxx
|
|
# children: [{
|
|
# className: class_y
|
|
# methodName: method_y
|
|
# }]
|
|
# }
|
|
# call_stack_tree = None
|
|
|
|
|
|
"""
|
|
Tree rendering
|
|
"""
|
|
STEP_FIRST_CHAR = "`---"
|
|
STEP_NORMAL_CHAR = "+---"
|
|
STEP_HAS_BOARD = "| "
|
|
STEP_EMPTY_BOARD = " "
|
|
|
|
|
|
def nano_to_millis(nanoSeconds):
|
|
return nanoSeconds / 1000000.0
|
|
|
|
def render_node(node):
|
|
str = ''
|
|
if node.get('threadName'):
|
|
# thread
|
|
str += "ts=%s;thread_name=%s;id=%s;is_daemon=%s;priority=%d;TCCL=%s" % (
|
|
datetime.fromtimestamp(node['timestamp']/1000).strftime("%Y-%m-%d %H:%M:%S"),
|
|
node['threadName'],
|
|
hex(node['threadId']),
|
|
node['daemon'],
|
|
node['priority'],
|
|
node['classloader']
|
|
)
|
|
if node.get('traceId'):
|
|
str += ";trace_id="+node['traceId']
|
|
if node.get('rpcId'):
|
|
str += ";rpc_id="+node['rpcId']
|
|
else:
|
|
# cost
|
|
times = node['times']
|
|
if times == 1:
|
|
str += '[%.3fms] ' % nano_to_millis(node['cost'])
|
|
else:
|
|
str += '[min=%.3fms,max=%.3fms,total=%.3fms,count=%d] ' % (nano_to_millis(node['minCost']),
|
|
nano_to_millis(node['maxCost']),
|
|
nano_to_millis(node['totalCost']),
|
|
times)
|
|
# method
|
|
str += "%s:%s()" % (node['className'], node['methodName'])
|
|
if node['lineNumber'] > 0:
|
|
str += " #%d" % node['lineNumber']
|
|
# mark
|
|
if node.get('mark'):
|
|
str += ' ['+node['mark']+']'
|
|
return str
|
|
|
|
def print_node(prefix, node, is_last):
|
|
current_prefix = (prefix+STEP_FIRST_CHAR) if is_last else (prefix+STEP_NORMAL_CHAR)
|
|
print("%s%s" % (current_prefix, render_node(node)))
|
|
# children
|
|
if node.get('children'):
|
|
children = node['children']
|
|
size = len(children)
|
|
for index in range(size):
|
|
current_prefix = (prefix + STEP_EMPTY_BOARD) if is_last else (prefix + STEP_HAS_BOARD)
|
|
is_last_child = (index == size-1)
|
|
print_node(current_prefix, children[index], is_last_child)
|
|
|
|
|
|
def print_trace_tree(root):
|
|
print_node('', root, True)
|
|
|
|
#------------------------ Tree Rendering End ---------------------------#
|
|
|
|
|
|
def init_session():
|
|
resp = requests.post(url, json={
|
|
"action": "init_session"
|
|
})
|
|
# print(resp.text)
|
|
result = resp.json()
|
|
if resp.status_code == 200 and result['state'] == 'SUCCEEDED':
|
|
session_id = result['sessionId']
|
|
consumer_id = result['consumerId']
|
|
return (session_id, consumer_id)
|
|
|
|
raise Exception('init http session failed: ' + resp.text)
|
|
|
|
|
|
def interrupt_job(session_id):
|
|
resp = requests.post(url, json={
|
|
"action": "interrupt_job",
|
|
"sessionId": session_id
|
|
})
|
|
# print(resp.text)
|
|
result = resp.json()
|
|
if resp.status_code == 200: # and result['state'] == 'SUCCEEDED'
|
|
return result
|
|
else:
|
|
raise Exception('init http session failed: ' + resp.text)
|
|
|
|
|
|
# Execute command sync
|
|
def exec_command(session_id, command):
|
|
resp = requests.post(url, json={
|
|
"action": "exec",
|
|
"command": command,
|
|
"sessionId": session_id
|
|
})
|
|
# print(resp.text)
|
|
result = resp.json()
|
|
state = result['state']
|
|
if resp.status_code == 200 and state == 'SUCCEEDED':
|
|
return result['body']['results']
|
|
else:
|
|
raise Exception('exec command failed: ' + resp.text)
|
|
|
|
|
|
# Execute command async
|
|
def async_exec(session_id, command):
|
|
resp = requests.post(url, json={
|
|
"action": "async_exec",
|
|
"command": command,
|
|
"sessionId": session_id
|
|
})
|
|
# print(resp.text)
|
|
result = resp.json()
|
|
state = result['state']
|
|
if resp.status_code == 200 and state == 'SCHEDULED':
|
|
return result['body']['jobId']
|
|
else:
|
|
raise Exception('async exec command failed: ' + resp.text)
|
|
|
|
|
|
# pull results of job
|
|
def pull_results(session_id, consumer_id, job_id, handler):
|
|
while True:
|
|
resp = requests.post(url, json={
|
|
"action": "pull_results",
|
|
"sessionId": session_id,
|
|
"consumerId": consumer_id
|
|
})
|
|
# print(resp.text)
|
|
json_resp = resp.json()
|
|
state = json_resp['state']
|
|
if resp.status_code == 200 and state == 'SUCCEEDED':
|
|
results = json_resp['body']['results']
|
|
for result in results:
|
|
if result.get('jobId'):
|
|
res_job_id = result['jobId'];
|
|
if res_job_id == job_id:
|
|
handler(result)
|
|
# check call tree match times
|
|
if call_tree_match_times >= stop_match_times:
|
|
print("The primary call tree matching times is exceeded, assuming no new call tree can be found.")
|
|
return
|
|
# receive status code of job, the job is terminated.
|
|
if result['type'] == 'status':
|
|
if result.get('message'): print(result['message'])
|
|
return
|
|
elif res_job_id > job_id:
|
|
# new job is executing
|
|
return
|
|
# TODO handle no response, timeout, cancel job
|
|
else:
|
|
raise Exception('pull results failed: ' + resp.text)
|
|
|
|
|
|
def handle_trace_result(result):
|
|
type = result['type']
|
|
if type == 'trace':
|
|
root = result['root']
|
|
trace_tree = root['children'][0]
|
|
method_path = match_call_tree(trace_tree)
|
|
if method_path:
|
|
# cancel job for executing other command
|
|
interrupt_job(session_id)
|
|
# add new method to path
|
|
index = len(trace_method_path)
|
|
size = len(method_path)
|
|
while index < size:
|
|
tm = method_path[index]
|
|
add_trace_method(tm['className'], tm['methodName'])
|
|
index += 1
|
|
|
|
# print
|
|
method_path_code = get_method_path_hash(method_path)
|
|
print("New primary call tree [%x]" % method_path_code)
|
|
print_trace_tree(root)
|
|
print_method_path(method_path, method_path_code)
|
|
print("")
|
|
|
|
|
|
def get_class_detail(class_name):
|
|
command = "sc -d " + class_name
|
|
results = exec_command(session_id, command)
|
|
for result in results:
|
|
type = result['type']
|
|
if type == 'class' and result['classInfo']['name'] == class_name:
|
|
return result['classInfo']
|
|
pass
|
|
|
|
|
|
def is_derived_from(class_detail, super_class):
|
|
super_classes = class_detail['superClass']
|
|
for sc in super_classes:
|
|
if sc == super_class:
|
|
return True
|
|
return False
|
|
|
|
|
|
def add_additional_method(class_name, method_name):
|
|
# class_name = replace_regex_chars(class_name)
|
|
# method_name = replace_regex_chars(method_name)
|
|
additional_methods.append({'className': class_name, 'methodName': method_name})
|
|
|
|
|
|
def add_trace_method(class_name, method_name):
|
|
# class_name = replace_regex_chars(class_name)
|
|
# method_name = replace_regex_chars(method_name)
|
|
tm = {
|
|
'className': class_name,
|
|
'methodName': method_name,
|
|
}
|
|
global trace_method_path_code
|
|
trace_method_path.append(tm)
|
|
trace_method_path_code = get_method_path_hash(trace_method_path)
|
|
|
|
# add java.lang.reflect.InvocationHandler for java.lang.reflect.Proxy instance
|
|
if 'java.lang.reflect.InvocationHandler' not in additional_methods:
|
|
class_detail = get_class_detail(class_name)
|
|
if is_derived_from(class_detail, 'java.lang.reflect.Proxy'):
|
|
add_additional_method('java.lang.reflect.InvocationHandler', 'invoke')
|
|
return tm
|
|
|
|
|
|
def print_trace_method_path():
|
|
print("trace method path: ")
|
|
for tm in trace_method_path:
|
|
print(" %s:%s()" % (tm['className'], tm['methodName']))
|
|
|
|
print("additional methods: ")
|
|
for am in additional_methods:
|
|
print(" %s:%s()" % (am['className'], am['methodName']))
|
|
pass
|
|
|
|
# replace regex chars
|
|
def replace_regex_chars(str):
|
|
return str.replace("$", "\\\\$")
|
|
#.replace(".", "\\.")
|
|
|
|
def start_trace():
|
|
# concat trace regex match pattern
|
|
# filter duplicated item by set
|
|
global is_skip_jdk_method
|
|
class_names = []
|
|
method_names = []
|
|
for tm in trace_method_path:
|
|
class_name = tm['className']
|
|
method_name = tm['methodName']
|
|
if class_name not in class_names:
|
|
class_names.append(class_name)
|
|
if method_name not in method_names:
|
|
method_names.append(method_name)
|
|
|
|
# append additional methods
|
|
for am in additional_methods:
|
|
class_name = am['className']
|
|
method_name = am['methodName']
|
|
if class_name not in class_names:
|
|
class_names.append(class_name)
|
|
if method_name not in method_names:
|
|
method_names.append(method_name)
|
|
|
|
class_pattern = "|".join(class_names)
|
|
method_pattern = "|".join(method_names)
|
|
class_pattern = replace_regex_chars(class_pattern)
|
|
method_pattern = replace_regex_chars(method_pattern)
|
|
|
|
command = "trace -E {0} {1} {2} -n {3}".format(class_pattern, method_pattern, condition, trace_times)
|
|
if not is_skip_jdk_method:
|
|
command += " --skipJDKMethod false"
|
|
|
|
print("")
|
|
print_trace_method_path()
|
|
print("command: %s" % command)
|
|
|
|
# async exec trace
|
|
job_id = async_exec(session_id, command)
|
|
print("job_id: %d" % job_id)
|
|
|
|
return job_id
|
|
|
|
|
|
def match_node(node, class_name, method_name):
|
|
return node['className'] == class_name and node['methodName'] == method_name
|
|
|
|
|
|
# 遍历调用树，生成关键方法路径
|
|
def create_method_path_from_tree(root):
|
|
method_path = []
|
|
node = root
|
|
while node:
|
|
class_name = node['className']
|
|
method_name = node['methodName']
|
|
method_path.append({'className': class_name, 'methodName': method_name})
|
|
node = replace_duplicated_node(node, class_name, method_name)
|
|
node = get_max_cost_node(node)
|
|
|
|
return method_path
|
|
|
|
|
|
def get_method_path_hash(method_path):
|
|
return hash(str(method_path))
|
|
|
|
|
|
def print_method_path(method_path, method_path_code):
|
|
print("slow method path [%x]: " % method_path_code)
|
|
for m in method_path:
|
|
print(" %s:%s()" % (m["className"], m["methodName"]))
|
|
|
|
|
|
def get_match_size(method_path1, method_path2):
|
|
for index in range(len(method_path1)):
|
|
if method_path1[index] != method_path2[index]:
|
|
return index
|
|
return len(method_path1)
|
|
|
|
# compare trace method path
|
|
# return:
|
|
# method_path: new call tree found
|
|
# None: match none / exact match / partial matching
|
|
def match_call_tree(trace_tree):
|
|
# compare trace method path
|
|
global call_tree_match_times
|
|
method_path = create_method_path_from_tree(trace_tree)
|
|
match_size = get_match_size(trace_method_path, method_path)
|
|
if match_size == len(trace_method_path):
|
|
if match_size == len(method_path):
|
|
# exact match
|
|
call_tree_match_times+=1
|
|
print("Exact matching primary call tree [%x] times: %d" % (trace_method_path_code, call_tree_match_times))
|
|
return None
|
|
elif len(method_path) > match_size:
|
|
# new call tree
|
|
return method_path
|
|
else:
|
|
# error, len(method_path) < match_size
|
|
raise Exception("Matching call tree error")
|
|
elif match_size > 0:
|
|
# TODO 本次结果与之前的不完全匹配，如果方法时间比之前的大，应该进行修正
|
|
# print partial match on first meet
|
|
method_path_code = get_method_path_hash(method_path)
|
|
print("Partial matching call tree [%x]" % method_path_code)
|
|
if method_path not in partial_matching_method_paths:
|
|
partial_matching_method_paths.append(method_path)
|
|
print_trace_tree(trace_tree)
|
|
print_method_path(method_path, method_path_code)
|
|
print("")
|
|
return None
|
|
else:
|
|
# match none
|
|
# TODO match interface and it's impl class
|
|
return None
|
|
|
|
|
|
def get_max_cost_node(node):
|
|
children = node.get('children')
|
|
if not children:
|
|
return None
|
|
|
|
next_node = None
|
|
for child in children:
|
|
if next_node:
|
|
if next_node['maxCost'] < child['maxCost']:
|
|
next_node = child
|
|
else:
|
|
next_node = child
|
|
return next_node
|
|
|
|
|
|
def replace_duplicated_node(node, class_name, method_name):
|
|
# ignore non-invoking node (fix Arthas duplicate enhance problem: https://github.com/alibaba/arthas/issues/599 )
|
|
children = node.get('children')
|
|
if children and len(children) == 1:
|
|
sub_node = children[0]
|
|
if match_node(sub_node, class_name, method_name) and not sub_node.get('invoking'):
|
|
node = sub_node
|
|
return node
|
|
|
|
def reset_classes():
|
|
exec_command(session_id, 'reset')
|
|
|
|
|
|
"""
|
|
Main
|
|
"""
|
|
# init session
|
|
(session_id, consumer_id) = init_session()
|
|
print("session_id: {0}, consumer_id: {1}".format(session_id, consumer_id))
|
|
|
|
try:
|
|
# parse method path
|
|
methods = args.method_path.split(',')
|
|
for m in methods:
|
|
strs = m.split(':')
|
|
class_name = strs[0].strip()
|
|
method_name = strs[1].replace('()','').strip()
|
|
# add trace method
|
|
add_trace_method(class_name, method_name)
|
|
|
|
# parse addit-method
|
|
if args.addit_method:
|
|
methods = args.addit_method.split(',')
|
|
for m in methods:
|
|
strs = m.split(':')
|
|
class_name = strs[0].strip()
|
|
method_name = strs[1].replace('()','').strip()
|
|
# additional method
|
|
add_additional_method(class_name, method_name)
|
|
|
|
if args.reset_on_start:
|
|
reset_classes()
|
|
|
|
trace_method_size = 0
|
|
while trace_method_size < len(trace_method_path):
|
|
# async trace
|
|
job_id = start_trace()
|
|
trace_method_size = len(trace_method_path)
|
|
|
|
# pull results
|
|
pull_results(session_id, consumer_id, job_id, handle_trace_result)
|
|
# reset on round
|
|
if args.reset_on_round:
|
|
reset_classes()
|
|
|
|
|
|
# print("")
|
|
# print("")
|
|
# print_trace_method_path()
|
|
print("Job is finished.")
|
|
except KeyboardInterrupt:
|
|
print("")
|
|
print("Job is canceled.")
|
|
finally:
|
|
interrupt_job(session_id)
|
|
reset_classes()
|