Note: This project is no longer actively maintained by Pinterest.
Quasar is a library that provides scalable, lightweight threads for the JVM (see references below for links to more information). Thrift is a code generator for services that works across multiple languages. This package integrates Quasar with Thrift, providing a high-performance Thrift server that takes advantage of Quasar's elegant programming model.
Quasar requires some setup to use correctly, as it uses byte-code instrumentation to implement
lightweight threads. Specifically, Thrift servers using Quasar need to mark some methods as
"suspendable", both in Java code and in separate resource files, and need to use a Java agent. The
best way to get both of these is to copy one of the example projects in the examples
directory
of this repository. To run the examples, first compile and install the quasar-thrift project:
./gradlew clean install
That will install the Maven dependency in your local ~/.m2 directory, where it can be used by the example projects. Then cd into one of the example project directories and run this command in one terminal:
./gradlew clean runServer
Then, in another terminal, in the same directory, run this command:
./gradlew runClient
At this time, the easiest way to use Quasar is with Ant or Gradle, as there is an Ant task that
needs to be run after compilation. The build.gradle
files used by the example servers has setup
that looks like this:
classes {
doFirst {
ant.taskdef(name: 'scanSuspendables',
classname: 'co.paralleluniverse.fibers.instrument.SuspendablesScanner',
classpath: "build/classes/main:build/resources/main:${configurations.runtime.asPath}")
ant.scanSuspendables(
auto: true,
suspendablesFile: "$sourceSets.main.output.resourcesDir/META-INF/suspendables",
supersFile: "$sourceSets.main.output.resourcesDir/META-INF/suspendable-supers",
append: true) {
fileset(dir: sourceSets.main.output.classesDir)
}
}
}
This tells Quasar to find all the methods to be instrumented at compile time, and to put them into
files named suspendables
and suspendable-supers
in the resources output directory. In some
cases you will need to manually mark files for instrumentation as well. Both of the example servers
have done this, and you can see the results in src/resources/META-INF/suspendables
in both of
the project directories.
If you are getting strange exceptions, typically NullPointerException
, you may have missed some
methods for instrumentation. At that point, uncomment this line in the build.grade file to have
Quasar give you a better runtime error:
//systemProperties "co.paralleluniverse.fibers.verifyInstrumentation": "true"
This will seriously degrade the performance of the server, so make sure to comment it back out when you are done.
This package provides a Thrift TServer
that works similarly with all the existing Thrift TServers,
and supports all the available Thrift Protocols and (non-server) Transports. It will work with any
existing Thrift client, in any language (assuming they use the same protocols and transports, of
course, as is true of any Thrift server and client).
This server uses N + 1 Quasar Fibers. A single Fiber is used to accept new connections from clients.
When a connection is accepted, a new Fiber is spawed to manage that connection, reading requests,
processing them, writing responses and then repeating until the connection is closed by the client
or an error. Methods called by the service must either throw SuspendExecution
or have the
@Suspendable
annotation in order for Quasar to find them and instrument them. That may not be
necessary in future versions of Quasar.
We have only run a small, simple load test of this library. We ran the example "echo" server on an EC2 m3.2xlarge server with Ubuntu 12.04 and some basic Linux TCP tuning for throughput. Then, on a separate EC2 instance in the same AZ we ran a load test written using Bender. Quasar/Thrift was able to support 6,000 QPS before the 99.9th percentile latency started to increase. We continued the test up to 10,000 QPS at which point the latency and error rate were relatively high (1% errors, >1s 50th percentile latency). A similar test using Twitter's Finagle libraries produced roughly the same results up to 7,000 QPS, and better results after.
The ParallelUniverse blog is a good source of information about Quasar:
- http://blog.paralleluniverse.co/2014/02/06/fibers-threads-strands/
- http://blog.paralleluniverse.co/2014/05/01/modern-java/
Copyright 2014 Pinterest.com
Licensed 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.
This library includes open source from the following sources:
- Apache Thrift Libraries Copyright 2014 Apache Software Foundation. Licensed under the Apache License v2.0 (http://www.apache.org/licenses/).
- Quasar Libraries Copyright 2014 Parallel Universe. Licensed under the GNU Lesser General Public License (http://www.gnu.org/licenses/lgpl.html).