|
|
|
# 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 json
|
|
|
|
import os.path
|
|
|
|
import requests
|
|
|
|
from requests.adapters import HTTPAdapter
|
|
|
|
from requests.packages.urllib3 import Retry
|
|
|
|
|
|
|
|
from ducktape.services.service import Service
|
KAFKA-6060; Add workload generation capabilities to Trogdor
Previously, Trogdor only handled "Faults." Now, Trogdor can handle
"Tasks" which may be either faults, or workloads to execute in the
background.
The Agent and Coordinator have been refactored from a
mutexes-and-condition-variables paradigm into a message passing
paradigm. No locks are necessary, because only one thread can access
the task state or worker state. This makes them a lot easier to reason
about.
The MockTime class can now handle mocking deferred message passing
(adding a message to an ExecutorService with a delay). I added a
MockTimeTest.
MiniTrogdorCluster now starts up Agent and Coordinator classes in
paralle in order to minimize junit test time.
RPC messages now inherit from a common Message.java class. This class
handles implementing serialization, equals, hashCode, etc.
Remove FaultSet, since it is no longer necessary.
Previously, if CoordinatorClient or AgentClient hit a networking
problem, they would throw an exception. They now retry several times
before giving up. Additionally, the REST RPCs to the Coordinator and
Agent have been changed to be idempotent. If a response is lost, and
the request is resent, no harm will be done.
Author: Colin P. Mccabe <cmccabe@confluent.io>
Reviewers: Rajini Sivaram <rajinisivaram@googlemail.com>, Ismael Juma <ismael@juma.me.uk>
Closes #4073 from cmccabe/KAFKA-6060
7 years ago
|
|
|
from ducktape.utils.util import wait_until
|
|
|
|
from kafkatest.directory_layout.kafka_path import KafkaPathResolverMixin
|
|
|
|
|
|
|
|
|
|
|
|
class TrogdorService(KafkaPathResolverMixin, Service):
|
|
|
|
"""
|
|
|
|
A ducktape service for running the trogdor fault injection daemons.
|
|
|
|
|
|
|
|
Attributes:
|
|
|
|
PERSISTENT_ROOT The root filesystem path to store service files under.
|
|
|
|
COORDINATOR_STDOUT_STDERR The path where we store the coordinator's stdout/stderr output.
|
|
|
|
AGENT_STDOUT_STDERR The path where we store the agents's stdout/stderr output.
|
|
|
|
COORDINATOR_LOG The path where we store the coordinator's log4j output.
|
|
|
|
AGENT_LOG The path where we store the agent's log4j output.
|
|
|
|
AGENT_LOG4J_PROPERTIES The path to the agent log4j.properties file for log config.
|
|
|
|
COORDINATOR_LOG4J_PROPERTIES The path to the coordinator log4j.properties file for log config.
|
|
|
|
CONFIG_PATH The path to the trogdor configuration file.
|
|
|
|
DEFAULT_AGENT_PORT The default port to use for trogdor_agent daemons.
|
|
|
|
DEFAULT_COORDINATOR_PORT The default port to use for trogdor_coordinator daemons.
|
|
|
|
REQUEST_TIMEOUT The request timeout in seconds to use for REST requests.
|
|
|
|
REQUEST_HEADERS The request headers to use when communicating with trogdor.
|
|
|
|
"""
|
|
|
|
|
|
|
|
PERSISTENT_ROOT="/mnt/trogdor"
|
|
|
|
COORDINATOR_STDOUT_STDERR = os.path.join(PERSISTENT_ROOT, "trogdor-coordinator-stdout-stderr.log")
|
|
|
|
AGENT_STDOUT_STDERR = os.path.join(PERSISTENT_ROOT, "trogdor-agent-stdout-stderr.log")
|
|
|
|
COORDINATOR_LOG = os.path.join(PERSISTENT_ROOT, "trogdor-coordinator.log")
|
|
|
|
AGENT_LOG = os.path.join(PERSISTENT_ROOT, "trogdor-agent.log")
|
|
|
|
COORDINATOR_LOG4J_PROPERTIES = os.path.join(PERSISTENT_ROOT, "trogdor-coordinator-log4j.properties")
|
|
|
|
AGENT_LOG4J_PROPERTIES = os.path.join(PERSISTENT_ROOT, "trogdor-agent-log4j.properties")
|
|
|
|
CONFIG_PATH = os.path.join(PERSISTENT_ROOT, "trogdor.conf")
|
|
|
|
DEFAULT_AGENT_PORT=8888
|
|
|
|
DEFAULT_COORDINATOR_PORT=8889
|
|
|
|
REQUEST_TIMEOUT=5
|
|
|
|
REQUEST_HEADERS = {"Content-type": "application/json"}
|
|
|
|
|
|
|
|
logs = {
|
|
|
|
"trogdor_coordinator_stdout_stderr": {
|
|
|
|
"path": COORDINATOR_STDOUT_STDERR,
|
|
|
|
"collect_default": True},
|
|
|
|
"trogdor_agent_stdout_stderr": {
|
|
|
|
"path": AGENT_STDOUT_STDERR,
|
|
|
|
"collect_default": True},
|
|
|
|
"trogdor_coordinator_log": {
|
|
|
|
"path": COORDINATOR_LOG,
|
|
|
|
"collect_default": True},
|
|
|
|
"trogdor_agent_log": {
|
|
|
|
"path": AGENT_LOG,
|
|
|
|
"collect_default": True},
|
|
|
|
}
|
|
|
|
|
KAFKA-6060; Add workload generation capabilities to Trogdor
Previously, Trogdor only handled "Faults." Now, Trogdor can handle
"Tasks" which may be either faults, or workloads to execute in the
background.
The Agent and Coordinator have been refactored from a
mutexes-and-condition-variables paradigm into a message passing
paradigm. No locks are necessary, because only one thread can access
the task state or worker state. This makes them a lot easier to reason
about.
The MockTime class can now handle mocking deferred message passing
(adding a message to an ExecutorService with a delay). I added a
MockTimeTest.
MiniTrogdorCluster now starts up Agent and Coordinator classes in
paralle in order to minimize junit test time.
RPC messages now inherit from a common Message.java class. This class
handles implementing serialization, equals, hashCode, etc.
Remove FaultSet, since it is no longer necessary.
Previously, if CoordinatorClient or AgentClient hit a networking
problem, they would throw an exception. They now retry several times
before giving up. Additionally, the REST RPCs to the Coordinator and
Agent have been changed to be idempotent. If a response is lost, and
the request is resent, no harm will be done.
Author: Colin P. Mccabe <cmccabe@confluent.io>
Reviewers: Rajini Sivaram <rajinisivaram@googlemail.com>, Ismael Juma <ismael@juma.me.uk>
Closes #4073 from cmccabe/KAFKA-6060
7 years ago
|
|
|
|
|
|
|
def __init__(self, context, agent_nodes=None, client_services=None,
|
|
|
|
agent_port=DEFAULT_AGENT_PORT, coordinator_port=DEFAULT_COORDINATOR_PORT):
|
|
|
|
"""
|
|
|
|
Create a Trogdor service.
|
|
|
|
|
|
|
|
:param context: The test context.
|
|
|
|
:param agent_nodes: The nodes to run the agents on.
|
KAFKA-6060; Add workload generation capabilities to Trogdor
Previously, Trogdor only handled "Faults." Now, Trogdor can handle
"Tasks" which may be either faults, or workloads to execute in the
background.
The Agent and Coordinator have been refactored from a
mutexes-and-condition-variables paradigm into a message passing
paradigm. No locks are necessary, because only one thread can access
the task state or worker state. This makes them a lot easier to reason
about.
The MockTime class can now handle mocking deferred message passing
(adding a message to an ExecutorService with a delay). I added a
MockTimeTest.
MiniTrogdorCluster now starts up Agent and Coordinator classes in
paralle in order to minimize junit test time.
RPC messages now inherit from a common Message.java class. This class
handles implementing serialization, equals, hashCode, etc.
Remove FaultSet, since it is no longer necessary.
Previously, if CoordinatorClient or AgentClient hit a networking
problem, they would throw an exception. They now retry several times
before giving up. Additionally, the REST RPCs to the Coordinator and
Agent have been changed to be idempotent. If a response is lost, and
the request is resent, no harm will be done.
Author: Colin P. Mccabe <cmccabe@confluent.io>
Reviewers: Rajini Sivaram <rajinisivaram@googlemail.com>, Ismael Juma <ismael@juma.me.uk>
Closes #4073 from cmccabe/KAFKA-6060
7 years ago
|
|
|
:param client_services: Services whose nodes we should run agents on.
|
|
|
|
:param agent_port: The port to use for the trogdor_agent daemons.
|
|
|
|
:param coordinator_port: The port to use for the trogdor_coordinator daemons.
|
|
|
|
"""
|
|
|
|
Service.__init__(self, context, num_nodes=1)
|
|
|
|
self.coordinator_node = self.nodes[0]
|
KAFKA-6060; Add workload generation capabilities to Trogdor
Previously, Trogdor only handled "Faults." Now, Trogdor can handle
"Tasks" which may be either faults, or workloads to execute in the
background.
The Agent and Coordinator have been refactored from a
mutexes-and-condition-variables paradigm into a message passing
paradigm. No locks are necessary, because only one thread can access
the task state or worker state. This makes them a lot easier to reason
about.
The MockTime class can now handle mocking deferred message passing
(adding a message to an ExecutorService with a delay). I added a
MockTimeTest.
MiniTrogdorCluster now starts up Agent and Coordinator classes in
paralle in order to minimize junit test time.
RPC messages now inherit from a common Message.java class. This class
handles implementing serialization, equals, hashCode, etc.
Remove FaultSet, since it is no longer necessary.
Previously, if CoordinatorClient or AgentClient hit a networking
problem, they would throw an exception. They now retry several times
before giving up. Additionally, the REST RPCs to the Coordinator and
Agent have been changed to be idempotent. If a response is lost, and
the request is resent, no harm will be done.
Author: Colin P. Mccabe <cmccabe@confluent.io>
Reviewers: Rajini Sivaram <rajinisivaram@googlemail.com>, Ismael Juma <ismael@juma.me.uk>
Closes #4073 from cmccabe/KAFKA-6060
7 years ago
|
|
|
if client_services is not None:
|
|
|
|
for client_service in client_services:
|
|
|
|
for node in client_service.nodes:
|
|
|
|
self.nodes.append(node)
|
|
|
|
if agent_nodes is not None:
|
|
|
|
for agent_node in agent_nodes:
|
|
|
|
self.nodes.append(agent_node)
|
|
|
|
if (len(self.nodes) == 1):
|
|
|
|
raise RuntimeError("You must supply at least one agent node to run the service on.")
|
|
|
|
self.agent_port = agent_port
|
|
|
|
self.coordinator_port = coordinator_port
|
|
|
|
|
|
|
|
def free(self):
|
|
|
|
# We only want to deallocate the coordinator node, not the agent nodes. So we
|
|
|
|
# change self.nodes to include only the coordinator node, and then invoke
|
|
|
|
# the base class' free method.
|
|
|
|
if self.coordinator_node is not None:
|
|
|
|
self.nodes = [self.coordinator_node]
|
|
|
|
self.coordinator_node = None
|
|
|
|
Service.free(self)
|
|
|
|
|
|
|
|
def _create_config_dict(self):
|
|
|
|
"""
|
|
|
|
Create a dictionary with the Trogdor configuration.
|
|
|
|
|
|
|
|
:return: The configuration dictionary.
|
|
|
|
"""
|
|
|
|
dict_nodes = {}
|
|
|
|
for node in self.nodes:
|
|
|
|
dict_nodes[node.name] = {
|
|
|
|
"hostname": node.account.ssh_hostname,
|
|
|
|
}
|
KAFKA-6060; Add workload generation capabilities to Trogdor
Previously, Trogdor only handled "Faults." Now, Trogdor can handle
"Tasks" which may be either faults, or workloads to execute in the
background.
The Agent and Coordinator have been refactored from a
mutexes-and-condition-variables paradigm into a message passing
paradigm. No locks are necessary, because only one thread can access
the task state or worker state. This makes them a lot easier to reason
about.
The MockTime class can now handle mocking deferred message passing
(adding a message to an ExecutorService with a delay). I added a
MockTimeTest.
MiniTrogdorCluster now starts up Agent and Coordinator classes in
paralle in order to minimize junit test time.
RPC messages now inherit from a common Message.java class. This class
handles implementing serialization, equals, hashCode, etc.
Remove FaultSet, since it is no longer necessary.
Previously, if CoordinatorClient or AgentClient hit a networking
problem, they would throw an exception. They now retry several times
before giving up. Additionally, the REST RPCs to the Coordinator and
Agent have been changed to be idempotent. If a response is lost, and
the request is resent, no harm will be done.
Author: Colin P. Mccabe <cmccabe@confluent.io>
Reviewers: Rajini Sivaram <rajinisivaram@googlemail.com>, Ismael Juma <ismael@juma.me.uk>
Closes #4073 from cmccabe/KAFKA-6060
7 years ago
|
|
|
if node.name == self.coordinator_node.name:
|
|
|
|
dict_nodes[node.name]["trogdor.coordinator.port"] = self.coordinator_port
|
|
|
|
else:
|
|
|
|
dict_nodes[node.name]["trogdor.agent.port"] = self.agent_port
|
|
|
|
|
|
|
|
return {
|
|
|
|
"platform": "org.apache.kafka.trogdor.basic.BasicPlatform",
|
|
|
|
"nodes": dict_nodes,
|
|
|
|
}
|
|
|
|
|
|
|
|
def start_node(self, node):
|
|
|
|
node.account.mkdirs(TrogdorService.PERSISTENT_ROOT)
|
|
|
|
|
|
|
|
# Create the configuration file on the node.
|
|
|
|
str = json.dumps(self._create_config_dict(), indent=2)
|
|
|
|
self.logger.info("Creating configuration file %s with %s" % (TrogdorService.CONFIG_PATH, str))
|
|
|
|
node.account.create_file(TrogdorService.CONFIG_PATH, str)
|
|
|
|
|
|
|
|
if self.is_coordinator(node):
|
|
|
|
self._start_coordinator_node(node)
|
|
|
|
else:
|
|
|
|
self._start_agent_node(node)
|
|
|
|
|
|
|
|
def _start_coordinator_node(self, node):
|
|
|
|
node.account.create_file(TrogdorService.COORDINATOR_LOG4J_PROPERTIES,
|
|
|
|
self.render('log4j.properties',
|
|
|
|
log_path=TrogdorService.COORDINATOR_LOG))
|
|
|
|
self._start_trogdor_daemon("coordinator", TrogdorService.COORDINATOR_STDOUT_STDERR,
|
|
|
|
TrogdorService.COORDINATOR_LOG4J_PROPERTIES,
|
|
|
|
TrogdorService.COORDINATOR_LOG, node)
|
|
|
|
self.logger.info("Started trogdor coordinator on %s." % node.name)
|
|
|
|
|
|
|
|
def _start_agent_node(self, node):
|
|
|
|
node.account.create_file(TrogdorService.AGENT_LOG4J_PROPERTIES,
|
|
|
|
self.render('log4j.properties',
|
|
|
|
log_path=TrogdorService.AGENT_LOG))
|
|
|
|
self._start_trogdor_daemon("agent", TrogdorService.AGENT_STDOUT_STDERR,
|
|
|
|
TrogdorService.AGENT_LOG4J_PROPERTIES,
|
|
|
|
TrogdorService.AGENT_LOG, node)
|
|
|
|
self.logger.info("Started trogdor agent on %s." % node.name)
|
|
|
|
|
|
|
|
def _start_trogdor_daemon(self, daemon_name, stdout_stderr_capture_path,
|
|
|
|
log4j_properties_path, log_path, node):
|
|
|
|
cmd = "export KAFKA_LOG4J_OPTS='-Dlog4j.configuration=file:%s'; " % log4j_properties_path
|
|
|
|
cmd += "%s %s --%s.config %s --node-name %s 1>> %s 2>> %s &" % \
|
|
|
|
(self.path.script("trogdor.sh", node),
|
|
|
|
daemon_name,
|
|
|
|
daemon_name,
|
|
|
|
TrogdorService.CONFIG_PATH,
|
|
|
|
node.name,
|
|
|
|
stdout_stderr_capture_path,
|
|
|
|
stdout_stderr_capture_path)
|
|
|
|
node.account.ssh(cmd)
|
|
|
|
with node.account.monitor_log(log_path) as monitor:
|
KAFKA-6060; Add workload generation capabilities to Trogdor
Previously, Trogdor only handled "Faults." Now, Trogdor can handle
"Tasks" which may be either faults, or workloads to execute in the
background.
The Agent and Coordinator have been refactored from a
mutexes-and-condition-variables paradigm into a message passing
paradigm. No locks are necessary, because only one thread can access
the task state or worker state. This makes them a lot easier to reason
about.
The MockTime class can now handle mocking deferred message passing
(adding a message to an ExecutorService with a delay). I added a
MockTimeTest.
MiniTrogdorCluster now starts up Agent and Coordinator classes in
paralle in order to minimize junit test time.
RPC messages now inherit from a common Message.java class. This class
handles implementing serialization, equals, hashCode, etc.
Remove FaultSet, since it is no longer necessary.
Previously, if CoordinatorClient or AgentClient hit a networking
problem, they would throw an exception. They now retry several times
before giving up. Additionally, the REST RPCs to the Coordinator and
Agent have been changed to be idempotent. If a response is lost, and
the request is resent, no harm will be done.
Author: Colin P. Mccabe <cmccabe@confluent.io>
Reviewers: Rajini Sivaram <rajinisivaram@googlemail.com>, Ismael Juma <ismael@juma.me.uk>
Closes #4073 from cmccabe/KAFKA-6060
7 years ago
|
|
|
monitor.wait_until("Starting %s process." % daemon_name, timeout_sec=60, backoff_sec=.25,
|
|
|
|
err_msg=("%s on %s didn't finish startup" % (daemon_name, node.name)))
|
|
|
|
|
|
|
|
def wait_node(self, node, timeout_sec=None):
|
|
|
|
if self.is_coordinator(node):
|
|
|
|
return len(node.account.java_pids(self.coordinator_class_name())) == 0
|
|
|
|
else:
|
|
|
|
return len(node.account.java_pids(self.agent_class_name())) == 0
|
|
|
|
|
|
|
|
def stop_node(self, node):
|
|
|
|
"""Halt trogdor processes on this node."""
|
|
|
|
if self.is_coordinator(node):
|
|
|
|
node.account.kill_java_processes(self.coordinator_class_name())
|
|
|
|
else:
|
|
|
|
node.account.kill_java_processes(self.agent_class_name())
|
|
|
|
|
|
|
|
def clean_node(self, node):
|
|
|
|
"""Clean up persistent state on this node - e.g. service logs, configuration files etc."""
|
|
|
|
self.stop_node(node)
|
|
|
|
node.account.ssh("rm -rf -- %s" % TrogdorService.PERSISTENT_ROOT)
|
|
|
|
|
|
|
|
def _coordinator_url(self, path):
|
|
|
|
return "http://%s:%d/coordinator/%s" % \
|
|
|
|
(self.coordinator_node.account.ssh_hostname, self.coordinator_port, path)
|
|
|
|
|
|
|
|
def request_session(self):
|
|
|
|
"""
|
|
|
|
Creates a new request session which will retry for a while.
|
|
|
|
"""
|
|
|
|
session = requests.Session()
|
|
|
|
session.mount('http://',
|
|
|
|
HTTPAdapter(max_retries=Retry(total=4, backoff_factor=0.3)))
|
|
|
|
return session
|
|
|
|
|
KAFKA-6060; Add workload generation capabilities to Trogdor
Previously, Trogdor only handled "Faults." Now, Trogdor can handle
"Tasks" which may be either faults, or workloads to execute in the
background.
The Agent and Coordinator have been refactored from a
mutexes-and-condition-variables paradigm into a message passing
paradigm. No locks are necessary, because only one thread can access
the task state or worker state. This makes them a lot easier to reason
about.
The MockTime class can now handle mocking deferred message passing
(adding a message to an ExecutorService with a delay). I added a
MockTimeTest.
MiniTrogdorCluster now starts up Agent and Coordinator classes in
paralle in order to minimize junit test time.
RPC messages now inherit from a common Message.java class. This class
handles implementing serialization, equals, hashCode, etc.
Remove FaultSet, since it is no longer necessary.
Previously, if CoordinatorClient or AgentClient hit a networking
problem, they would throw an exception. They now retry several times
before giving up. Additionally, the REST RPCs to the Coordinator and
Agent have been changed to be idempotent. If a response is lost, and
the request is resent, no harm will be done.
Author: Colin P. Mccabe <cmccabe@confluent.io>
Reviewers: Rajini Sivaram <rajinisivaram@googlemail.com>, Ismael Juma <ismael@juma.me.uk>
Closes #4073 from cmccabe/KAFKA-6060
7 years ago
|
|
|
def _coordinator_post(self, path, message):
|
|
|
|
"""
|
|
|
|
Make a POST request to the Trogdor coordinator.
|
|
|
|
|
|
|
|
:param path: The URL path to use.
|
|
|
|
:param message: The message object to send.
|
|
|
|
:return: The response as an object.
|
|
|
|
"""
|
|
|
|
url = self._coordinator_url(path)
|
|
|
|
self.logger.info("POST %s %s" % (url, message))
|
|
|
|
response = self.request_session().post(url, json=message,
|
|
|
|
timeout=TrogdorService.REQUEST_TIMEOUT,
|
|
|
|
headers=TrogdorService.REQUEST_HEADERS)
|
|
|
|
response.raise_for_status()
|
|
|
|
return response.json()
|
|
|
|
|
|
|
|
def _coordinator_put(self, path, message):
|
|
|
|
"""
|
|
|
|
Make a PUT request to the Trogdor coordinator.
|
|
|
|
|
|
|
|
:param path: The URL path to use.
|
|
|
|
:param message: The message object to send.
|
|
|
|
:return: The response as an object.
|
|
|
|
"""
|
|
|
|
url = self._coordinator_url(path)
|
|
|
|
self.logger.info("PUT %s %s" % (url, message))
|
|
|
|
response = self.request_session().put(url, json=message,
|
|
|
|
timeout=TrogdorService.REQUEST_TIMEOUT,
|
|
|
|
headers=TrogdorService.REQUEST_HEADERS)
|
|
|
|
response.raise_for_status()
|
|
|
|
return response.json()
|
|
|
|
|
|
|
|
def _coordinator_get(self, path, message):
|
|
|
|
"""
|
|
|
|
Make a GET request to the Trogdor coordinator.
|
|
|
|
|
|
|
|
:param path: The URL path to use.
|
|
|
|
:param message: The message object to send.
|
|
|
|
:return: The response as an object.
|
|
|
|
"""
|
|
|
|
url = self._coordinator_url(path)
|
|
|
|
self.logger.info("GET %s %s" % (url, message))
|
|
|
|
response = self.request_session().get(url, json=message,
|
|
|
|
timeout=TrogdorService.REQUEST_TIMEOUT,
|
|
|
|
headers=TrogdorService.REQUEST_HEADERS)
|
|
|
|
response.raise_for_status()
|
|
|
|
return response.json()
|
|
|
|
|
KAFKA-6060; Add workload generation capabilities to Trogdor
Previously, Trogdor only handled "Faults." Now, Trogdor can handle
"Tasks" which may be either faults, or workloads to execute in the
background.
The Agent and Coordinator have been refactored from a
mutexes-and-condition-variables paradigm into a message passing
paradigm. No locks are necessary, because only one thread can access
the task state or worker state. This makes them a lot easier to reason
about.
The MockTime class can now handle mocking deferred message passing
(adding a message to an ExecutorService with a delay). I added a
MockTimeTest.
MiniTrogdorCluster now starts up Agent and Coordinator classes in
paralle in order to minimize junit test time.
RPC messages now inherit from a common Message.java class. This class
handles implementing serialization, equals, hashCode, etc.
Remove FaultSet, since it is no longer necessary.
Previously, if CoordinatorClient or AgentClient hit a networking
problem, they would throw an exception. They now retry several times
before giving up. Additionally, the REST RPCs to the Coordinator and
Agent have been changed to be idempotent. If a response is lost, and
the request is resent, no harm will be done.
Author: Colin P. Mccabe <cmccabe@confluent.io>
Reviewers: Rajini Sivaram <rajinisivaram@googlemail.com>, Ismael Juma <ismael@juma.me.uk>
Closes #4073 from cmccabe/KAFKA-6060
7 years ago
|
|
|
def create_task(self, id, spec):
|
|
|
|
"""
|
KAFKA-6060; Add workload generation capabilities to Trogdor
Previously, Trogdor only handled "Faults." Now, Trogdor can handle
"Tasks" which may be either faults, or workloads to execute in the
background.
The Agent and Coordinator have been refactored from a
mutexes-and-condition-variables paradigm into a message passing
paradigm. No locks are necessary, because only one thread can access
the task state or worker state. This makes them a lot easier to reason
about.
The MockTime class can now handle mocking deferred message passing
(adding a message to an ExecutorService with a delay). I added a
MockTimeTest.
MiniTrogdorCluster now starts up Agent and Coordinator classes in
paralle in order to minimize junit test time.
RPC messages now inherit from a common Message.java class. This class
handles implementing serialization, equals, hashCode, etc.
Remove FaultSet, since it is no longer necessary.
Previously, if CoordinatorClient or AgentClient hit a networking
problem, they would throw an exception. They now retry several times
before giving up. Additionally, the REST RPCs to the Coordinator and
Agent have been changed to be idempotent. If a response is lost, and
the request is resent, no harm will be done.
Author: Colin P. Mccabe <cmccabe@confluent.io>
Reviewers: Rajini Sivaram <rajinisivaram@googlemail.com>, Ismael Juma <ismael@juma.me.uk>
Closes #4073 from cmccabe/KAFKA-6060
7 years ago
|
|
|
Create a new task.
|
|
|
|
|
KAFKA-6060; Add workload generation capabilities to Trogdor
Previously, Trogdor only handled "Faults." Now, Trogdor can handle
"Tasks" which may be either faults, or workloads to execute in the
background.
The Agent and Coordinator have been refactored from a
mutexes-and-condition-variables paradigm into a message passing
paradigm. No locks are necessary, because only one thread can access
the task state or worker state. This makes them a lot easier to reason
about.
The MockTime class can now handle mocking deferred message passing
(adding a message to an ExecutorService with a delay). I added a
MockTimeTest.
MiniTrogdorCluster now starts up Agent and Coordinator classes in
paralle in order to minimize junit test time.
RPC messages now inherit from a common Message.java class. This class
handles implementing serialization, equals, hashCode, etc.
Remove FaultSet, since it is no longer necessary.
Previously, if CoordinatorClient or AgentClient hit a networking
problem, they would throw an exception. They now retry several times
before giving up. Additionally, the REST RPCs to the Coordinator and
Agent have been changed to be idempotent. If a response is lost, and
the request is resent, no harm will be done.
Author: Colin P. Mccabe <cmccabe@confluent.io>
Reviewers: Rajini Sivaram <rajinisivaram@googlemail.com>, Ismael Juma <ismael@juma.me.uk>
Closes #4073 from cmccabe/KAFKA-6060
7 years ago
|
|
|
:param id: The task id.
|
|
|
|
:param spec: The task spec.
|
|
|
|
"""
|
KAFKA-6060; Add workload generation capabilities to Trogdor
Previously, Trogdor only handled "Faults." Now, Trogdor can handle
"Tasks" which may be either faults, or workloads to execute in the
background.
The Agent and Coordinator have been refactored from a
mutexes-and-condition-variables paradigm into a message passing
paradigm. No locks are necessary, because only one thread can access
the task state or worker state. This makes them a lot easier to reason
about.
The MockTime class can now handle mocking deferred message passing
(adding a message to an ExecutorService with a delay). I added a
MockTimeTest.
MiniTrogdorCluster now starts up Agent and Coordinator classes in
paralle in order to minimize junit test time.
RPC messages now inherit from a common Message.java class. This class
handles implementing serialization, equals, hashCode, etc.
Remove FaultSet, since it is no longer necessary.
Previously, if CoordinatorClient or AgentClient hit a networking
problem, they would throw an exception. They now retry several times
before giving up. Additionally, the REST RPCs to the Coordinator and
Agent have been changed to be idempotent. If a response is lost, and
the request is resent, no harm will be done.
Author: Colin P. Mccabe <cmccabe@confluent.io>
Reviewers: Rajini Sivaram <rajinisivaram@googlemail.com>, Ismael Juma <ismael@juma.me.uk>
Closes #4073 from cmccabe/KAFKA-6060
7 years ago
|
|
|
self._coordinator_post("task/create", { "id": id, "spec": spec.message()})
|
|
|
|
return TrogdorTask(id, self)
|
|
|
|
|
KAFKA-6060; Add workload generation capabilities to Trogdor
Previously, Trogdor only handled "Faults." Now, Trogdor can handle
"Tasks" which may be either faults, or workloads to execute in the
background.
The Agent and Coordinator have been refactored from a
mutexes-and-condition-variables paradigm into a message passing
paradigm. No locks are necessary, because only one thread can access
the task state or worker state. This makes them a lot easier to reason
about.
The MockTime class can now handle mocking deferred message passing
(adding a message to an ExecutorService with a delay). I added a
MockTimeTest.
MiniTrogdorCluster now starts up Agent and Coordinator classes in
paralle in order to minimize junit test time.
RPC messages now inherit from a common Message.java class. This class
handles implementing serialization, equals, hashCode, etc.
Remove FaultSet, since it is no longer necessary.
Previously, if CoordinatorClient or AgentClient hit a networking
problem, they would throw an exception. They now retry several times
before giving up. Additionally, the REST RPCs to the Coordinator and
Agent have been changed to be idempotent. If a response is lost, and
the request is resent, no harm will be done.
Author: Colin P. Mccabe <cmccabe@confluent.io>
Reviewers: Rajini Sivaram <rajinisivaram@googlemail.com>, Ismael Juma <ismael@juma.me.uk>
Closes #4073 from cmccabe/KAFKA-6060
7 years ago
|
|
|
def stop_task(self, id):
|
|
|
|
"""
|
KAFKA-6060; Add workload generation capabilities to Trogdor
Previously, Trogdor only handled "Faults." Now, Trogdor can handle
"Tasks" which may be either faults, or workloads to execute in the
background.
The Agent and Coordinator have been refactored from a
mutexes-and-condition-variables paradigm into a message passing
paradigm. No locks are necessary, because only one thread can access
the task state or worker state. This makes them a lot easier to reason
about.
The MockTime class can now handle mocking deferred message passing
(adding a message to an ExecutorService with a delay). I added a
MockTimeTest.
MiniTrogdorCluster now starts up Agent and Coordinator classes in
paralle in order to minimize junit test time.
RPC messages now inherit from a common Message.java class. This class
handles implementing serialization, equals, hashCode, etc.
Remove FaultSet, since it is no longer necessary.
Previously, if CoordinatorClient or AgentClient hit a networking
problem, they would throw an exception. They now retry several times
before giving up. Additionally, the REST RPCs to the Coordinator and
Agent have been changed to be idempotent. If a response is lost, and
the request is resent, no harm will be done.
Author: Colin P. Mccabe <cmccabe@confluent.io>
Reviewers: Rajini Sivaram <rajinisivaram@googlemail.com>, Ismael Juma <ismael@juma.me.uk>
Closes #4073 from cmccabe/KAFKA-6060
7 years ago
|
|
|
Stop a task.
|
|
|
|
|
KAFKA-6060; Add workload generation capabilities to Trogdor
Previously, Trogdor only handled "Faults." Now, Trogdor can handle
"Tasks" which may be either faults, or workloads to execute in the
background.
The Agent and Coordinator have been refactored from a
mutexes-and-condition-variables paradigm into a message passing
paradigm. No locks are necessary, because only one thread can access
the task state or worker state. This makes them a lot easier to reason
about.
The MockTime class can now handle mocking deferred message passing
(adding a message to an ExecutorService with a delay). I added a
MockTimeTest.
MiniTrogdorCluster now starts up Agent and Coordinator classes in
paralle in order to minimize junit test time.
RPC messages now inherit from a common Message.java class. This class
handles implementing serialization, equals, hashCode, etc.
Remove FaultSet, since it is no longer necessary.
Previously, if CoordinatorClient or AgentClient hit a networking
problem, they would throw an exception. They now retry several times
before giving up. Additionally, the REST RPCs to the Coordinator and
Agent have been changed to be idempotent. If a response is lost, and
the request is resent, no harm will be done.
Author: Colin P. Mccabe <cmccabe@confluent.io>
Reviewers: Rajini Sivaram <rajinisivaram@googlemail.com>, Ismael Juma <ismael@juma.me.uk>
Closes #4073 from cmccabe/KAFKA-6060
7 years ago
|
|
|
:param id: The task id.
|
|
|
|
"""
|
|
|
|
self._coordinator_put("task/stop", { "id": id })
|
|
|
|
|
|
|
|
def tasks(self):
|
|
|
|
"""
|
|
|
|
Get the tasks which are on the coordinator.
|
|
|
|
|
|
|
|
:returns: A map of task id strings to task state objects.
|
|
|
|
Task state objects contain a 'spec' field with the spec
|
|
|
|
and a 'state' field with the state.
|
|
|
|
"""
|
KAFKA-6060; Add workload generation capabilities to Trogdor
Previously, Trogdor only handled "Faults." Now, Trogdor can handle
"Tasks" which may be either faults, or workloads to execute in the
background.
The Agent and Coordinator have been refactored from a
mutexes-and-condition-variables paradigm into a message passing
paradigm. No locks are necessary, because only one thread can access
the task state or worker state. This makes them a lot easier to reason
about.
The MockTime class can now handle mocking deferred message passing
(adding a message to an ExecutorService with a delay). I added a
MockTimeTest.
MiniTrogdorCluster now starts up Agent and Coordinator classes in
paralle in order to minimize junit test time.
RPC messages now inherit from a common Message.java class. This class
handles implementing serialization, equals, hashCode, etc.
Remove FaultSet, since it is no longer necessary.
Previously, if CoordinatorClient or AgentClient hit a networking
problem, they would throw an exception. They now retry several times
before giving up. Additionally, the REST RPCs to the Coordinator and
Agent have been changed to be idempotent. If a response is lost, and
the request is resent, no harm will be done.
Author: Colin P. Mccabe <cmccabe@confluent.io>
Reviewers: Rajini Sivaram <rajinisivaram@googlemail.com>, Ismael Juma <ismael@juma.me.uk>
Closes #4073 from cmccabe/KAFKA-6060
7 years ago
|
|
|
return self._coordinator_get("tasks", {})
|
|
|
|
|
|
|
|
def is_coordinator(self, node):
|
|
|
|
return node == self.coordinator_node
|
|
|
|
|
|
|
|
def agent_class_name(self):
|
|
|
|
return "org.apache.kafka.trogdor.agent.Agent"
|
|
|
|
|
|
|
|
def coordinator_class_name(self):
|
|
|
|
return "org.apache.kafka.trogdor.coordinator.Coordinator"
|
KAFKA-6060; Add workload generation capabilities to Trogdor
Previously, Trogdor only handled "Faults." Now, Trogdor can handle
"Tasks" which may be either faults, or workloads to execute in the
background.
The Agent and Coordinator have been refactored from a
mutexes-and-condition-variables paradigm into a message passing
paradigm. No locks are necessary, because only one thread can access
the task state or worker state. This makes them a lot easier to reason
about.
The MockTime class can now handle mocking deferred message passing
(adding a message to an ExecutorService with a delay). I added a
MockTimeTest.
MiniTrogdorCluster now starts up Agent and Coordinator classes in
paralle in order to minimize junit test time.
RPC messages now inherit from a common Message.java class. This class
handles implementing serialization, equals, hashCode, etc.
Remove FaultSet, since it is no longer necessary.
Previously, if CoordinatorClient or AgentClient hit a networking
problem, they would throw an exception. They now retry several times
before giving up. Additionally, the REST RPCs to the Coordinator and
Agent have been changed to be idempotent. If a response is lost, and
the request is resent, no harm will be done.
Author: Colin P. Mccabe <cmccabe@confluent.io>
Reviewers: Rajini Sivaram <rajinisivaram@googlemail.com>, Ismael Juma <ismael@juma.me.uk>
Closes #4073 from cmccabe/KAFKA-6060
7 years ago
|
|
|
|
|
|
|
class TrogdorTask(object):
|
|
|
|
PENDING_STATE = "PENDING"
|
|
|
|
RUNNING_STATE = "RUNNING"
|
|
|
|
STOPPING_STATE = "STOPPING"
|
|
|
|
DONE_STATE = "DONE"
|
|
|
|
|
|
|
|
def __init__(self, id, trogdor):
|
|
|
|
self.id = id
|
|
|
|
self.trogdor = trogdor
|
|
|
|
|
|
|
|
def done(self):
|
|
|
|
"""
|
|
|
|
Check if this task is done.
|
|
|
|
|
|
|
|
:raises RuntimeError: If the task encountered an error.
|
|
|
|
:returns: True if the task is in DONE_STATE;
|
|
|
|
False if it is in a different state.
|
|
|
|
"""
|
|
|
|
task_state = self.trogdor.tasks()["tasks"][self.id]
|
|
|
|
if task_state is None:
|
|
|
|
raise RuntimeError("Coordinator did not know about %s." % self.id)
|
|
|
|
error = task_state.get("error")
|
|
|
|
if error is None or error == "":
|
|
|
|
return task_state["state"] == TrogdorTask.DONE_STATE
|
|
|
|
raise RuntimeError("Failed to gracefully stop %s: got task error: %s" % (self.id, error))
|
|
|
|
|
|
|
|
def stop(self):
|
|
|
|
"""
|
|
|
|
Stop this task.
|
|
|
|
|
|
|
|
:raises RuntimeError: If the task encountered an error.
|
|
|
|
"""
|
|
|
|
if self.done():
|
|
|
|
return
|
|
|
|
self.trogdor.stop_task(self.id)
|
|
|
|
|
|
|
|
def wait_for_done(self, timeout_sec=360):
|
|
|
|
wait_until(lambda: self.done(),
|
|
|
|
timeout_sec=timeout_sec,
|
|
|
|
err_msg="%s failed to finish in the expected amount of time." % self.id)
|