"/healthcheck" is too common and conflicts with potential Annotation scan of URI template "/healthcheck(/.*)

Let's change "/healthcheck" to "/surohealthcheck" and "/sinkstat" to "/surosinkstat"

Message is the simple container of String routing key and byte[] message payload. For filtering the message, message payload should be deserialized and it can be cached for another filters. Also, some sink needs specific deserialized format, such as Map<String, Object> or Json string like ElasticSearch.

Right now, Suro supports logging from Java only. A Python client should be developed.

Treat all Sink and source types as independent modules that are enabled by adding them to the withModules of the injector creation step. Various implementations will be made available using guice mapbinder.

We didn't find the root cause of this unit test failure but instead of using ZkClient's test zookeeper server, using curator-test looks better.

populationLatch.await(a few seconds, TimeUnit.SECONDS) should be...

We have a requirement to route messages with the different topic name. 'alias' will be added to routing map configuration. If that property is valid, messages will be routed with that 'alias', not the original routingKey.

• What went wrong:
  Execution failed for task ':suro-core:javadoc'.

  Javadoc generation failed.

Whenever QueuedSink fails to write successfully data, it should retry messages not to lose messages. When the queue is full, messages will be dropped and this should be reported.

'''
2014. 5. 23 오후 4:11:40 org.apache.catalina.loader.WebappClassLoader clearReferencesThreads
심각: The web application [] appears to have started a thread named [NFLoadBalancer-PingTimer-suroClient] but has failed to stop it. This is very likely to create a memory leak.
...
...
2014. 5. 23 오후 4:11:40 org.apache.catalina.loader.WebappClassLoader clearReferencesThreads
심각: The web application [] appears to have started a thread named [AsyncSuroClientPoller-0] but has failed to stop it. This is very likely to create a memory leak.
2014. 5. 23 오후 4:11:40 org.apache.catalina.loader.WebappClassLoader clearReferencesThreads
...
...
정보: Stopping Coyote HTTP/1.1 on http-10004
Exception in thread "NFLoadBalancer-PingTimer-suroClient" java.lang.NullPointerException
at com.netflix.loadbalancer.BaseLoadBalancer$PingTask.run(BaseLoadBalancer.java:583)
at java.util.TimerThread.mainLoop(Timer.java:512)
at java.util.TimerThread.run(Timer.java:462)
'''

LifecycleManager.close() should call SinkManager.shutdown() for automatic resource management.

EventBus filter is the part of Netflix Common now. Remove EventBus filter source code and make the dependency on Netflix Common.

The current version of Suro uses an auto-incremented long variable as partition key, so messages are partitioned in the way of round-robin fashion by default. To support wider usecases Suro must have extension point to add custom partitioning logic.

Refer for more details https://groups.google.com/forum/#!topic/suro-users/6wz7ZDpbjFU

Move common code such as DefaultObjectMapper into a common core module

API for creating a consumer to SQS, CassQueue, Kafka, etc...

Implement end-to-end message consumer - filter - route - produce using Rx.

After building suro-client and starting the server following instructions on the wiki, we're getting the following error:

ERROR org.apache.thrift.server.TNonblockingServer - Read an invalid frame size of 0. Are you using TFramedTransport on the client side?

Break out Kafka into a separate sub-project

Both SuroService and ThriftServer use System.exit(). They shouldn't. Only Suro driver should decide whether to use System.exit() or not.

Implementat a Disruptor based MessageQueue. Part of this will also be to determine whether distruptor actually makes sense.

Define reserved keywords first

• date
• static
• property

Users can combine them together for generating remote prefix.

'date' needs date format string
'static' is static string value
'property' is replaced by the property value from ConfigurationManager.getConfigInstance().getProperty()

In the master branch, there is DataConverter interface which is injectable. Instead of injectable DataConverter interface, we'd better use groovy script.

[root@localhost suro]# ./gradlew clean build

FAILURE: Build failed with an exception.

• What went wrong:
  Could not resolve all dependencies for configuration ':classpath'.

  Artifact 'com.google.code.findbugs:jsr305:1.3.9@jar' not found.

• Try:
  Run with --stacktrace option to get the stack trace. Run with --info or --debug option to get more log output.

BUILD FAILED

Total time: 5.438 secs
[root@localhost suro]#

Generic API for sending messages to SQS, Kafka, Suro, CassQueue, etc..

Create abstraction for MessageQueue so we can try various implementations instead of just LinkedBlockingQueue.

Steps:

Bring up a new Amazon Linux instance

sudo yum -y update
sudo sum -y install git

git clone https://github.com/Netflix/suro.git
cd suro
./gradlew build

Errors:

:suro-server

com.netflix.suro.server.TestSuroControl > testOnlyExitCommandWorks FAILED
    java.net.ConnectException at TestSuroControl.java:67

This error is intermittent. Retrying the builds sometimes succeeds. This may be because of the findAvailablePort method is returning a used port.

The source line is:

Socket client = new Socket("127.0.0.1", port);

:suro-client

com.netflix.suro.connection.TestConnectionOutPool > testOutPool FAILED
    java.lang.AssertionError at TestConnectionOutPool.java:67

Source:

final ConnectionPool pool = injector.getInstance(ConnectionPool.class);
assertEquals(pool.getPoolSize(), 1);

Suro message is using hadoop for Writable implementation but this looks an extra dependency because Writable implementation is used for only writing it to hadoop sequence file. Let's remove it but it can break backward compatibility because value class of hadoop sequence file is changed.

Suro-server side sink should be reusable but its dependency is a little heavy including jetty, jersey, aws sdk, jets3t. Let's break down these dependencies into

suro-file and suro-s3.

Adding row id would be helpful for de-dup or finding lost messages.

Suro client can set up row id explicitly, otherwise, MD5Hash value based on byte[] payload would be added automatically.

FileQueue4Sink is used in QueuedSink. Its run() function calls drainTo and check message list is full or expired. FileQueue4Sink doesn't do autocommit by default and this behavior is draining same messages multiple times.

The following fields should not be part of TagKey.java. Maybe we should put all the counters into a single utility class along with all the convenience methods.

public static final String SENT_COUNT = "sentMessageCount";
public static final String RECV_COUNT = "receivedMessageCount";
public static final String LOST_COUNT = "lostMessageCount";
public static final String RESTORED_COUNT = "restoredMessageCount";
public static final String RETRIED_COUNT = "retriedCount";

ConcurrentModificationException can be thrown on a rare error situations, where ConnectionPool is refreshed very frequently due to lots of malfunctioning suro servers.

Currently, suro's input is only thrift server. We need to make them various including at least kafka consumer.

There will be one more dynamic configurable property called SuroServer.inputConfig json format.

If one of inputConfig element is 'thrift', all suroserver's configuration should be matched to the previous configuration which starts with SuroServer.

Fork join message queue as alternative to simple LinkedBlockingQueue. This should improve performance.