[:en] Operational problems with Zookeeper

This post is a summary of what has been presented by Kathleen Ting on StrangeLoop conference. You can watch the original here: http://www.infoq.com/presentations/Misconfiguration-ZooKeeper I’ve decided to put this selection here for quick reference. …This post is a summary of what has been presented by Kathleen Ting on StrangeLoop conference. You can watch the original here: http://www.infoq.com/presentations/Misconfiguration-ZooKeeper I’ve decided to put this selection here for quick reference. …

This post is a summary of what has been presented by Kathleen Ting on
StrangeLoop conference. You can watch the original here:
http://www.infoq.com/presentations/Misconfiguration-ZooKeeper

I’ve decided to put this selection here for quick reference.

Connection mismanagement

  • too many connections
    WARN [NIOServerCxn.Factory: 0.0.0.0/0.0.0.0:2181:NIOServerCnxn$Factory@247] - Too many connections from /xx.x.xx.xxx - max is 60
    
  • running out of ZK connections?
    • set maxClientCnxns=200 in zoo.cfg
  • HBase client leaking connections?
    • fixed in HBASE-3777, HBASE-4773, HBASE-5466
    • manually close connections
  • connection closes prematurely
    ERROR: org.apache.hadoop.hbase.ZooKeeperConnectionException: HBase is able to connect to ZooKeeper but the connection closes immediately.
  • in hbase-site.xml set hbase.zookeeper.recoverable.waittime=30000ms
  • pig hangs connecting to HBase

    CAUSE: location of ZK quorum is not known to Pig

    • use Pig 10, which includes PIG-2115
    • if there is an overlap between TaskTrackers and ZK quorum nodes
      • set hbase.zookeeper.quorum to final in hbase-site.xml
      • otherwise add hbaze.zoopeeker.quorum=hadoophbasemaster.lan:2181 in pig.properties
WARN org.apache.zookeeper.ClientCnxn: Session 0x0 for server null, unexpected error, closing socket connection and attempting reconnect java.net.ConnectionException: Connection refused!

Time mismanagement

  • client session timed out
INFO org.apache.zookeeper.server.ZooKeeperServer: Expiring session <id>, timeout of 40000ms exceeded
    • ZK and HBase need the same session timeout values
      • zoo.cfg: maxSession=Timeout=180000
      • hbase-site.xml: zookeeper.session.timeout=180000
    • don’t co-locate ZK with IO-intense DataNode or RegionServer
    • specify right amount of heap and tune GC flags
      • turn on parallel/CMS/incremental GC
  •  
  • clients lose connections
    WARN org.apache.zookeeper.ClientCnxn - Session <id> for server <name>, unexpected error, closing socket connection and attempting reconnect java.io.IOException: Broken pipe
    
    • don’t use SSD drive for ZK transaction log

Disk management

  • unable to load database – unable to run quorum server
    FATAL Unable to load database on disk !  java.io.IOException: Failed to process transaction type: 2 error: KeeperErrorCode = NoNode for <file> at org.apache.zookeeper.server.persistence.FileTxnSnapLog.restore(FileTxnSnapLog.java:152)!
    
    • archive and wipe /var/zookeeper/version-2 if other two ZK servers
      are running
  • unable to load database – unreasonable length exception
    FATAL Unable to load database on disk java.io.IOException: Unreasonable length = 1048583 at org.apache.jute.BinaryInputArchive.readBuffer(BinaryInputArchive.java:100)
    • server allows a client to set data larger than the server can read from disk
    • if a znode is not readable, increase jute.maxbuffer
      • look for "Packet len <xx> is out of range" in the client log
      • increase it by 20%
      • set in JVMFLAGS="-Djute.maxbuffer=yy" bin/zkCli.sh
      • fixed in ZOOKEEPER-151
  • failure to follow leader
    WARN org.apache.zookeeper.server.quorum.Learner: Exception when following the leader java.net.SocketTimeoutException: Read timed out

    CAUSE:

    • disk IO contention, network issues
    • ZK snapshot is too large (lots of ZK nodes)

    SOLVE:

    • reduce IO contention by putting dataDir on dedicated spindle
    • increase initLimit on all ZK servers and restart, see
      ZOOKEEPER-1521
    • monitor network

Best Practices

DOs

  • separate spindles for dataDir & dataLogDir
  • allocate 3 or 5 ZK servers
  • tune garbage collection
  • run zkCleanup.sh script via cron

DON’Ts

  • dont’ co-locate ZK with I/O intense DataNode or RegionServer
  • don’t use SSD drive for ZK transaction log

You may use Zookeeper as an observer – a non-voting member:

  • in zoo.cfg
    peerType=observer
You May Also Like

Zookeeper + Curator = Distributed sync

An application developed for one of my recent projects at TouK involved multiple servers. There was a requirement to ensure failover for the system’s components. Since I had already a few separate components I didn’t want to add more of that, and since there already was a Zookeeper ensemble running - required by one of the services, I’ve decided to go that way with my solution.

What is Zookeeper?

Just a crude distributed synchronization framework. However, it implements Paxos-style algorithms (http://en.wikipedia.org/wiki/Paxos_(computer_science)) to ensure no split-brain scenarios would occur. This is quite an important feature, since I don’t have to care about that kind of problems while using this app. You just need to create an ensemble of a couple of its instances - to ensure high availability. It is basically a virtual filesystem, with files, directories and stuff. One could ask why another filesystem? Well this one is a rather special one, especially for distributed systems. The reason why creating all the locking algorithms on top of Zookeeper is easy is its Ephemeral Nodes - which are just files that exist as long as connection for them exists. After it disconnects - such file disappears.

With such paradigms in place it’s fairly easy to create some high level algorithms for synchronization.

Having that in place, it can safely integrate multiple services ensuring loose coupling in a distributed way.

Zookeeper from developer’s POV

With all the base services for Zookeeper started, it seems there is nothing else, than just connect to it and start implementing necessary algorithms. Unfortunately, the API is quite basic and offers files and directories abstractions with the addition of different node type (file types) - ephemeral and sequence. It is also possible to watch a node for changes.

Using bare Zookeeper is hard!

Creating connections is tedious - and there is lots of things to take care of. Handling an established connection is hard - when establishing connection to ensemble, it’s necessary to negotiate a session also. During the whole process a number of exceptions can occur - these are “recoverable” exceptions, that can be gracefully handled and not break the connection.

    class="c8"><span>So, Zookeeper API is hard.</span></p><p class="c1"><span></span></p><p class="c8"><span>Even if one is proficient with that API, then there come recipes. The reason for using Zookeeper is to be able to implement some more sophisticated algorithms on top of it. Unfortunately those aren&rsquo;t trivial and it is again quite hard to implement them without bugs.</span>

And since distributed systems are hard, why would anyone want another difficult to handle tool?

Enter Curator

<p
    class="c8"><span>Happily, guys from Netflix implemented a nice abstraction for dealing with Zookeeper internals. They called it Curator and use it extensively in the company&rsquo;s environment. Curator offers consistent API for Zookeeper&rsquo;s functionality. It even implements a couple of recipes for distributed systems.</span>

File read/write

<p
    class="c8"><span>The basic use of Zookeeper is as a distributed configuration repository. For this scenario I only need read/write capabilities, to be able to write and read files from the Zookeeper filesystem. This code snippet writes a sample json to a file on ZK filesystem.</span>

<a href="#"
                                                                                                  name="0"></a>

EnsurePath ensurePath = new EnsurePath(markerPath);
ensurePath.ensure(client.getZookeeperClient());
String json = “...”;
if (client.checkExists().forPath(statusFile(core)) != null)
     client.setData().forPath(statusFile(core), json.getBytes());
else
     client.create().forPath(statusFile(core), json.getBytes());


Distributed locking

Having multiple systems there may be a need of using an exclusive lock for some resource, or perhaps some big system requires it’s components to synchronize based on locks. This “recipe” is an ideal match for those situations.

ref="#"
                                                                                    name="b0329bbbf14b79ffaba1139881914aea887ef6a3"></a>



lock = new InterProcessSemaphoreMutex(client, lockPath);
lock.acquire(5, TimeUnit.MINUTES);
… do sth …
lock.release();


 (from https://github.com/zygm0nt/curator-playground/blob/master/src/main/java/pl/touk/curator/LockingRemotely.java)

Sevice Advertisement

<p

    class="c8"><span>This is quite an interesting use case. With many small services on different servers it is not wise to exchange ip addresses and ports between them. When some of those services may go down, while other will try to replace them - the task gets even harder. </span>

That’s why, with Zookeeper in place, it can be utilised as a registry of existing services.

If a service starts, it registers into the ServiceRegistry, offering basic information, like it’s purpose, role, address, and port.

Services that want to use a specific kind of service request an access to some instance. This way of configuring easily decouples services from their configuration.

Basically this scenario needs ? steps:

<span>1. Service starts and registers its presence (</span><span class="c5"><a class="c0"
                                                                               href="https://github.com/zygm0nt/curator-playground/blob/master/src/main/java/pl/touk/curator/WorkerAdvertiser.java#L44">https://github.com/zygm0nt/curator-playground/blob/master/src/main/java/pl/touk/curator/WorkerAdvertiser.java#L44</a></span><span>)</span><span>:</span>



ServiceDiscovery discovery = getDiscovery();
            discovery.start();
            ServiceInstance si = getInstance();
            log.info(si);
            discovery.registerService(si);



2. Another service - on another host or in another JVM on the same machine tries to discover who is implementing the service (https://github.com/zygm0nt/curator-playground/blob/master/src/main/java/pl/touk/curator/WorkerFinder.java#L50):

<a href="#"

                                                                                                  name="3"></a>

instances = discovery.queryForInstances(serviceName);

The whole concept here is ridiculously simple - the service advertising its presence just stores a file with its whereabouts. The service that is looking for service providers just look into specific directory and read stored definitions.

In my example, the structure advertised by services looks like this (+ some getters and constructor - the rest is here: https://github.com/zygm0nt/curator-playground/blob/master/src/main/java/pl/touk/model/WorkerMetadata.java):



public final class WorkerMetadata {
    private final UUID workerId;
    private final String listenAddress;
    private final int listenPort;
}


Source code

<p

    class="c8"><span>The above recipes are available in Curator library (</span><span class="c5"><a class="c0"
                                                                                                    href="http://curator.incubator.apache.org/">http://curator.incubator.apache.org/</a></span><span>). Recipes&rsquo;
usage examples are in my github repo at </span><span class="c5"><a class="c0"
                                                                   href="https://github.com/zygm0nt/curator-playground">https://github.com/zygm0nt/curator-playground</a></span>

Conclusion

<p
    class="c8"><span>If you&rsquo;re in need of a reliable platform for exchanging data and managing synchronization, and you need to do it in a distributed fashion - just choose Zookeeper. Then add Curator for the ease of using it. Enjoy!</span>


  1. image comes from: http://www.flickr.com/photos/jfgallery/2993361148
  2. all source code fragments taken from this repo: https://github.com/zygm0nt/curator-playground

An application developed for one of my recent projects at TouK involved multiple servers. There was a requirement to ensure failover for the system’s components. Since I had already a few separate components I didn’t want to add more of that, and since there already was a Zookeeper ensemble running - required by one of the services, I’ve decided to go that way with my solution.

What is Zookeeper?

Just a crude distributed synchronization framework. However, it implements Paxos-style algorithms (http://en.wikipedia.org/wiki/Paxos_(computer_science)) to ensure no split-brain scenarios would occur. This is quite an important feature, since I don’t have to care about that kind of problems while using this app. You just need to create an ensemble of a couple of its instances - to ensure high availability. It is basically a virtual filesystem, with files, directories and stuff. One could ask why another filesystem? Well this one is a rather special one, especially for distributed systems. The reason why creating all the locking algorithms on top of Zookeeper is easy is its Ephemeral Nodes - which are just files that exist as long as connection for them exists. After it disconnects - such file disappears.

With such paradigms in place it’s fairly easy to create some high level algorithms for synchronization.

Having that in place, it can safely integrate multiple services ensuring loose coupling in a distributed way.

Zookeeper from developer’s POV

With all the base services for Zookeeper started, it seems there is nothing else, than just connect to it and start implementing necessary algorithms. Unfortunately, the API is quite basic and offers files and directories abstractions with the addition of different node type (file types) - ephemeral and sequence. It is also possible to watch a node for changes.

Using bare Zookeeper is hard!

Creating connections is tedious - and there is lots of things to take care of. Handling an established connection is hard - when establishing connection to ensemble, it’s necessary to negotiate a session also. During the whole process a number of exceptions can occur - these are “recoverable” exceptions, that can be gracefully handled and not break the connection.

    class="c8"><span>So, Zookeeper API is hard.</span></p><p class="c1"><span></span></p><p class="c8"><span>Even if one is proficient with that API, then there come recipes. The reason for using Zookeeper is to be able to implement some more sophisticated algorithms on top of it. Unfortunately those aren&rsquo;t trivial and it is again quite hard to implement them without bugs.</span>

And since distributed systems are hard, why would anyone want another difficult to handle tool?

Enter Curator

<p
    class="c8"><span>Happily, guys from Netflix implemented a nice abstraction for dealing with Zookeeper internals. They called it Curator and use it extensively in the company&rsquo;s environment. Curator offers consistent API for Zookeeper&rsquo;s functionality. It even implements a couple of recipes for distributed systems.</span>

File read/write

<p
    class="c8"><span>The basic use of Zookeeper is as a distributed configuration repository. For this scenario I only need read/write capabilities, to be able to write and read files from the Zookeeper filesystem. This code snippet writes a sample json to a file on ZK filesystem.</span>

<a href="#"
                                                                                                  name="0"></a>

EnsurePath ensurePath = new EnsurePath(markerPath);
ensurePath.ensure(client.getZookeeperClient());
String json = “...”;
if (client.checkExists().forPath(statusFile(core)) != null)
     client.setData().forPath(statusFile(core), json.getBytes());
else
     client.create().forPath(statusFile(core), json.getBytes());


Distributed locking

Having multiple systems there may be a need of using an exclusive lock for some resource, or perhaps some big system requires it’s components to synchronize based on locks. This “recipe” is an ideal match for those situations.

ref="#"
                                                                                    name="b0329bbbf14b79ffaba1139881914aea887ef6a3"></a>



lock = new InterProcessSemaphoreMutex(client, lockPath);
lock.acquire(5, TimeUnit.MINUTES);
… do sth …
lock.release();


 (from https://github.com/zygm0nt/curator-playground/blob/master/src/main/java/pl/touk/curator/LockingRemotely.java)

Sevice Advertisement

<p

    class="c8"><span>This is quite an interesting use case. With many small services on different servers it is not wise to exchange ip addresses and ports between them. When some of those services may go down, while other will try to replace them - the task gets even harder. </span>

That’s why, with Zookeeper in place, it can be utilised as a registry of existing services.

If a service starts, it registers into the ServiceRegistry, offering basic information, like it’s purpose, role, address, and port.

Services that want to use a specific kind of service request an access to some instance. This way of configuring easily decouples services from their configuration.

Basically this scenario needs ? steps:

<span>1. Service starts and registers its presence (</span><span class="c5"><a class="c0"
                                                                               href="https://github.com/zygm0nt/curator-playground/blob/master/src/main/java/pl/touk/curator/WorkerAdvertiser.java#L44">https://github.com/zygm0nt/curator-playground/blob/master/src/main/java/pl/touk/curator/WorkerAdvertiser.java#L44</a></span><span>)</span><span>:</span>



ServiceDiscovery discovery = getDiscovery();
            discovery.start();
            ServiceInstance si = getInstance();
            log.info(si);
            discovery.registerService(si);



2. Another service - on another host or in another JVM on the same machine tries to discover who is implementing the service (https://github.com/zygm0nt/curator-playground/blob/master/src/main/java/pl/touk/curator/WorkerFinder.java#L50):

<a href="#"

                                                                                                  name="3"></a>

instances = discovery.queryForInstances(serviceName);

The whole concept here is ridiculously simple - the service advertising its presence just stores a file with its whereabouts. The service that is looking for service providers just look into specific directory and read stored definitions.

In my example, the structure advertised by services looks like this (+ some getters and constructor - the rest is here: https://github.com/zygm0nt/curator-playground/blob/master/src/main/java/pl/touk/model/WorkerMetadata.java):



public final class WorkerMetadata {
    private final UUID workerId;
    private final String listenAddress;
    private final int listenPort;
}


Source code

<p

    class="c8"><span>The above recipes are available in Curator library (</span><span class="c5"><a class="c0"
                                                                                                    href="http://curator.incubator.apache.org/">http://curator.incubator.apache.org/</a></span><span>). Recipes&rsquo;
usage examples are in my github repo at </span><span class="c5"><a class="c0"
                                                                   href="https://github.com/zygm0nt/curator-playground">https://github.com/zygm0nt/curator-playground</a></span>

Conclusion

<p
    class="c8"><span>If you&rsquo;re in need of a reliable platform for exchanging data and managing synchronization, and you need to do it in a distributed fashion - just choose Zookeeper. Then add Curator for the ease of using it. Enjoy!</span>


  1. image comes from: http://www.flickr.com/photos/jfgallery/2993361148
  2. all source code fragments taken from this repo: https://github.com/zygm0nt/curator-playground

Using WsLite in practice

TL;DR

There is a example working GitHub project which covers unit testing and request/response logging when using WsLite.

Why Groovy WsLite ?

I’m a huge fan of Groovy WsLite project for calling SOAP web services. Yes, in a real world you have to deal with those - big companies have huge amount of “legacy” code and are crazy about homogeneous architecture - only SOAP, Java, Oracle, AIX…

But I also never been comfortable with XFire/CXF approach of web service client code generation. I wrote a bit about other posibilites in this post. With JAXB you can also experience some freaky classloading errors - as Tomek described on his blog. In a large commercial project the “the less code the better” principle is significant. And the code generated from XSD could look kinda ugly - especially more complicated structures like sequences, choices, anys etc.

Using WsLite with native Groovy concepts like XmlSlurper could be a great choice. But since it’s a dynamic approach you have to be really careful - write good unit tests and log requests. Below are my few hints for using WsLite in practice.

Unit testing

Suppose you have some invocation of WsLite SOAPClient (original WsLite example):

def getMothersDay(long _year) {
    def response = client.send(SOAPAction: action) {
       body {
           GetMothersDay('xmlns':'http://www.27seconds.com/Holidays/US/Dates/') {
              year(_year)
           }
       }
    }
    response.GetMothersDayResponse.GetMothersDayResult.text()
}

How can the unit test like? My suggestion is to mock SOAPClient and write a simple helper to test that builded XML is correct. Example using great SpockFramework:

void setup() {
   client = Mock(SOAPClient)
   service.client = client
}

def "should pass year to GetMothersDay and return date"() {
  given:
      def year = 2013
  when:
      def date = service.getMothersDay(year)
  then:
      1 * client.send(_, _) >> { Map params, Closure requestBuilder ->
            Document doc = buildAndParseXml(requestBuilder)
            assertXpathEvaluatesTo("$year", '//ns:GetMothersDay/ns:year', doc)
            return mockResponse(Responses.mothersDay)
      }
      date == "2013-05-12T00:00:00"
}

This uses a real cool feature of Spock - even when you mock the invocation with “any mark” (_), you are able to get actual arguments. So we can build XML that would be passed to SOAPClient's send method and check that specific XPaths are correct:

void setup() {
    engine = XMLUnit.newXpathEngine()
    engine.setNamespaceContext(new SimpleNamespaceContext(namespaces()))
}

protected Document buildAndParseXml(Closure xmlBuilder) {
    def writer = new StringWriter()
    def builder = new MarkupBuilder(writer)
    builder.xml(xmlBuilder)
    return XMLUnit.buildControlDocument(writer.toString())
}

protected void assertXpathEvaluatesTo(String expectedValue,
                                      String xpathExpression, Document doc) throws XpathException {
    Assert.assertEquals(expectedValue,
            engine.evaluate(xpathExpression, doc))
}

protected Map namespaces() {
    return [ns: 'http://www.27seconds.com/Holidays/US/Dates/']
}

The XMLUnit library is used just for XpathEngine, but it is much more powerful for comparing XML documents. The NamespaceContext is needed to use correct prefixes (e.g. ns:GetMothersDay) in your Xpath expressions.

Finally - the mock returns SOAPResponse instance filled with envelope parsed from some constant XML:

protected SOAPResponse mockResponse(String resp) {
    def envelope = new XmlSlurper().parseText(resp)
    new SOAPResponse(envelope: envelope)
}

Request and response logging

The WsLite itself doesn’t use any logging framework. We usually handle it by adding own sendWithLogging method:

private SOAPResponse sendWithLogging(String action, Closure cl) {
    SOAPResponse response = client.send(SOAPAction: action, cl)
    log(response?.httpRequest, response?.httpResponse)
    return response
}

private void log(HTTPRequest request, HTTPResponse response) {
    log.debug("HTTPRequest $request with content:\n${request?.contentAsString}")
    log.debug("HTTPResponse $response with content:\n${response?.contentAsString}")
}

This logs the actual request and response send through SOAPClient. But it logs only when invocation is successful and errors are much more interesting… So here goes withExceptionHandler method:

private SOAPResponse withExceptionHandler(Closure cl) {
    try {
        cl.call()
    } catch (SOAPFaultException soapEx) {
        log(soapEx.httpRequest, soapEx.httpResponse)
        def message = soapEx.hasFault() ? soapEx.fault.text() : soapEx.message
        throw new InfrastructureException(message)
    } catch (HTTPClientException httpEx) {
        log(httpEx.request, httpEx.response)
        throw new InfrastructureException(httpEx.message)
    }
}
def send(String action, Closure cl) {
    withExceptionHandler {
        sendWithLogging(action, cl)
    }
}

XmlSlurper gotchas

Working with XML document with XmlSlurper is generally great fun, but is some cases could introduce some problems. A trivial example is parsing an id with a number to Long value:

def id = Long.valueOf(edit.'@id' as String)

The Attribute class (which edit.'@id' evaluates to) can be converted to String using as operator, but converting to Long requires using valueOf.

The second example is a bit more complicated. Consider following XML fragment:

<edit id="3">
   <params>
      <param value="label1" name="label"/>
      <param value="2" name="param2"/>
   </params>
   <value>123</value>
</edit>
<edit id="6">
   <params>
      <param value="label2" name="label"/>
      <param value="2" name="param2"/>
   </params>
   <value>456</value>
</edit>

We want to find id of edit whose label is label1. The simplest solution seems to be:

def param = doc.edit.params.param.find { it['@value'] == 'label1' }
def edit = params.parent().parent()

But it doesn’t work! The parent method returns multiple edits, not only the one that is parent of given param

Here’s the correct solution:

doc.edit.find { edit ->
    edit.params.param.find { it['@value'] == 'label1' }
}

Example

The example working project covering those hints could be found on GitHub.