[feature](tools)profile viewer (#14429)

It is a painful work to read profile, especially there are multi-parallel instances.
This tool helps us to grasp the main information of profile in a graphical view.

The profile is represented by a tree.
Sql operation nodes contains operation type(join, scan...), its node id, its fragment id. The number on the arrow edge means how many rows output by child node. This tool will sum the output rows of the same node in multi-parallel instances, that is if there are 4 parallel instance, and each ScanNode on lineitem table output 10 rows, the label on the arrow beginning with ScanNode(lineitem) is 40.

Here is a demo for tpch Q2
tpch q2 profile viewer

Issue Number: close #xxx
This commit is contained in:
minghong
2022-11-21 10:29:54 +08:00
committed by GitHub
parent 4976021bf7
commit 0613ccda74

329
tools/profile_viewer.py Normal file
View File

@ -0,0 +1,329 @@
#!/usr/bin/env python
# Licensed to the Apache Software Foundation (ASF) under one
# or more contributor license agreements. See the NOTICE file
# distributed with this work for additional information
# regarding copyright ownership. The ASF licenses this file
# to you under the Apache License, Version 2.0 (the
# "License"); you may not use this file except in compliance
# with the License. You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing,
# software distributed under the License is distributed on an
# "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
# KIND, either express or implied. See the License for the
# specific language governing permissions and limitations
# under the License.
import re
import os
import sys
#dot dig -T png -o dig.png
exchanges = {}
class Node:
TERMINAL_NODE_TYPES={'VEXCHANGE_NODE', 'VNewOlapScanNode'}
def __init__(self, id, type, frid):
self.id = id
self.frid = frid
self.type = type
self.parent = None
self.children = []
self.rows = 0
m = re.search("VNewOlapScanNode\((.*)\)", type)
if m != None:
self.table = m.group(1)
self.type = 'VNewOlapScanNode'
else:
self.table = ""
def set_data(self, data):
self.data = data
def parse(self):
for line in self.data:
m = re.search(".*RowsReturned:.+\(([\d]+)\)", line)
if m!= None:
self.rows = int(m.group(1))
break
else:
m = re.search(".*RowsReturned:.+([\d]+)", line)
if m != None:
self.rows = int(m.group(1))
break
def build_tree(self, nodes):
if self.type in Node.TERMINAL_NODE_TYPES:
return 0
self.children.append(nodes[0])
consume = 1 + nodes[0].build_tree(nodes[1:])
if self.type.find("JOIN") > 0:
self.children.append(nodes[consume])
consume = consume + 1 + nodes[consume].build_tree(nodes[consume+1:])
return consume
def __repr__(self):
if self.type == 'VNewOlapScanNode':
return "{}_{}".format(self.table, self.id)
return "{}_{}".format(self.type, self.id)
def to_label(self):
name = self.type
if self.type == 'VNewOlapScanNode':
name = self.table
return '"{}\\n#{}FR{}"'.format(name, self.id, self.frid)
def show(self, indent):
print("{}{}".format(indent, self))
for ch in self.children:
ch.show(" " + indent)
def to_dot(self, dot):
dot = dot + "\n{}[label={}]".format(self, self.to_label())
for ch in reversed(self.children):
dot = dot + "\n{}->{}[label={}]".format(ch, self, ch.rows)
for ch in self.children:
dot = ch.to_dot(dot)
return dot
class Instance:
def __init__(self, id, frid):
self.id = id
self.frid = frid
self.target = ""
self.srcs = []
self.nodes = []
self.has_build_tree = False
def set_data(self, data):
self.data = data
def build_tree(self):
if self.has_build_tree:
return
self.has_build_tree = True
self.root = self.nodes[0]
self.root.build_tree(self.nodes[1:])
def parse(self):
pos = 0
for line in self.data:
pos = pos + 1
m = re.search("\s+VDataStreamSender \(dst_id=(\d+),.*", line)
if m != None:
self.target = m.group(1)
break
for line in self.data:
m = re.search("\s+VEXCHANGE_NODE \(id=(\d+)\):.*", line)
if m != None:
self.srcs.append(m.group(1))
prev = 0
pos = 0
for line in self.data:
m = re.search("\s+([^\s]+N[o|O][d|D][e|E][^\s]*) \(id=(\d+)\.*", line)
if m != None:
type = m.group(1)
id = m.group(2)
node = Node(id, type, self.frid)
if node.type == 'VEXCHANGE_NODE':
exchanges[id] = node
if prev != 0:
#self.nodes
self.nodes[-1].set_data(self.data[prev : pos])
self.nodes.append(node)
prev = pos
pos = pos + 1
self.nodes[-1].set_data(self.data[prev : pos])
for node in self.nodes:
node.parse()
self.build_tree()
def show(self):
self.root.show(" ")
class Fragment:
def __init__(self, id):
self.id = id
self.target = ""
self.srcs = []
self.instances = []
self.children = []
self.parent = None
def set_data(self, data):
self.data = data
def get_instance_id(self, line):
m = re.search("\s+Instance (\w+-\w+)", line)
is_inst = m != None
return (is_inst, m.group(1) if is_inst else "")
def parse(self):
pos = 1
(is_inst, inst_id) = self.get_instance_id(self.data[pos])
inst = Instance(inst_id, self.id)
self.instances.append(inst)
size = len(self.data)
prev = pos
while pos < size - 1:
pos = pos + 1
(is_inst, inst_id) = self.get_instance_id(self.data[pos])
if is_inst:
inst.set_data(self.data[prev : pos])
inst = Instance(inst_id, self.id)
self.instances.append(inst)
prev = pos
inst.set_data(self.data[prev : pos])
for inst in self.instances:
inst.parse()
self.srcs = self.instances[0].srcs
self.target = self.instances[0].target
def show(self, indent):
print("{}{}".format(indent, self.id))
for fr in self.children:
fr.show(indent+" ")
def sum_nodes(self, nodes):
sum = 0
for node in nodes:
sum = sum + node.rows
nodes[0].rows = sum
for (i, ch) in enumerate(nodes[0].children):
chs = [node.children[i] for node in nodes]
self.sum_nodes(chs)
def zip_instances(self, instances):
'''sum rows of same nodes in different instance'''
roots = [inst.root for inst in instances]
self.sum_nodes(roots)
def to_dot(self):
inst = self.instances[0]
dot = ""
dot = inst.root.to_dot(dot)
return dot
class Tree:
def __init__(self, fragments):
self.fragments = fragments
def build(self):
src_fr_map = {}
for fr in self.fragments:
if fr.target == "":
self.root = fr
for src in fr.srcs:
src_fr_map[src] = fr
for fr in self.fragments:
if fr.target != "":
parent = src_fr_map.get(fr.target)
parent.children.append(fr)
fr.parent = parent
def show(self):
self.root.show(" ")
def load_profile(profile):
with open(profile, 'r') as f:
l = f.readlines()
return l
def get_fragment_id(line):
#line in format : Fragment /d+:
#len("Fragment "):10
line = line.strip()
return line[10:-1]
def is_fragment_begin(line):
line = line.strip()
return line.startswith("Fragment")
def cut_fragment(data):
fragments = []
prev = 0
pos = 0
total_len = len(data)
#skip some lines not in fragment
for line in data:
if is_fragment_begin(line):
break
prev = prev + 1
pos = prev
fr_id = get_fragment_id(data[pos])
pos = pos + 1
current_fr = Fragment(fr_id)
fragments = []
fragments.append(current_fr)
while pos < total_len:
line = data[pos]
if is_fragment_begin(line):
current_fr.set_data(data[prev : pos])
fr_id = get_fragment_id(line)
current_fr = Fragment(fr_id)
fragments.append(current_fr)
prev = pos
pos = pos + 1
current_fr.set_data(data[prev: pos - 1])
for fr in fragments:
fr.parse()
return fragments
def print_usage():
print("""
USAGE:
python profile_viewer.py [PATH_TO_PROFILE]
graphviz is required(https://graphviz.org/)
on linux: apt install graphvize
on mac: brew install graphvize
""")
if __name__ == '__main__':
if len(sys.argv) != 2:
print_usage()
exit(0)
prf_file = sys.argv[1]
if not os.path.exists(prf_file):
print("{} is not a file".format(prf_file))
exit(0)
data = load_profile(prf_file)
fragements = []
fragments = cut_fragment(data)
dot_header = """digraph BPMN {
rankdir=BT;
node[shape=rectanble style="rounded,filled" color="lightgoldenrodyellow" ]
"""
dot_tail = "}"
with open(prf_file+".dot", "w") as f:
f.write(dot_header)
for fr in fragments:
# print("fr{}:{} <-{}".format(fr.id, fr.target, fr.srcs))
fr.zip_instances(fr.instances)
f.write(fr.to_dot())
if fr.target != "":
f.write("\n{}->{}[label={}]".format(fr.instances[0].root, exchanges[fr.target], fr.instances[0].root.rows))
f.write(dot_tail)
cmd = "dot {} -T png -o {}".format(prf_file+".dot", prf_file+".png")
status = os.system(cmd)
if status != 0:
print("convert failed")
else:
print("generate " + prf_file + ".png")