DDevelopment & Design

[: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

— Previous article

TouK na warsztatach z Gita

Next article —

<!--:pl-->Testy - nie takie straszne, jak je malują<!--:--><!--:en-->Tests - not so scary as they used to<!--:-->

You May Also Like
DDevelopment & Design

33rd Degree day 1 review

33rd Degree is over. After the one last year, my expectations were very high, but Grzegorz Duda once again proved he's more than able to deliver. With up to five tracks (most of the time: four presentations + one workshop), and ~650 attendees,  there was a lot to see and a lot to do, thus everyone will probably have a little bit different story to tell. Here is mine.

Twitter: From Ruby on Rails to the JVM

Raffi Krikorian talking about Twitter and JVM
The conference started with  Raffi Krikorian from Twitter, talking about their use for JVM. Twitter was build with Ruby but with their performance management a lot of the backend was moved to Scala, Java and Closure. Raffi noted, that for Ruby programmers Scala was easier to grasp than Java, more natural, which is quite interesting considering how many PHP guys move to Ruby these days because of the same reasons. Perhaps the path of learning Jacek Laskowski once described (Java -> Groovy -> Scala/Closure) may be on par with PHP -> Ruby -> Scala. It definitely feels like Scala is the holy grail of languages these days.

Raffi also noted, that while JVM delivered speed and a concurrency model to Twitter stack, it wasn't enough, and they've build/customized their own Garbage Collector. My guess is that Scala/Closure could also be used because of a nice concurrency solutions (STM, immutables and so on).

Raffi pointed out, that with the scale of Twitter, you easily get 3 million hits per second, and that means you probably have 3 edge cases every second. I'd love to learn listen to lessons they've learned from this.

 

Complexity of Complexity


The second keynote of the first day, was Ken Sipe talking about complexity. He made a good point that there is a difference between complex and complicated, and that we often recognize things as complex only because we are less familiar with them. This goes more interesting the moment you realize that the shift in last 20 years of computer languages, from the "Less is more" paradigm (think Java, ASM) to "More is better" (Groovy/Scala/Closure), where you have more complex language, with more powerful and less verbose syntax, that is actually not more complicated, it just looks less familiar.

So while 10 years ago, I really liked Java as a general purpose language for it's small set of rules that could get you everywhere, it turned out that to do most of the real world stuff, a lot of code had to be written. The situation got better thanks to libraries/frameworks and so on, but it's just patching. New languages have a lot of stuff build into, which makes their set of rules and syntax much more complex, but once you get familiar, the real world usage is simple, faster, better, with less traps laying around, waiting for you to fall.

Ken also pointed out, that while Entity Service Bus looks really simple on diagrams, it's usually very difficult and complicated to use from the perspective of the programmer. And that's probably why it gets chosen so often - the guys selling/buying it, look no deeper than on the diagram.

 

Pointy haired bosses and pragmatic programmers: Facts and Fallacies of Software Development

Venkat Subramaniam with Dima
Dima got lucky. Or maybe not.

Venkat Subramaniam is the kind of a speaker that talk about very simple things in a way, which makes everyone either laugh or reflect. Yes, he is a showman, but hey, that's actually good, because even if you know the subject quite well, his talks are still very entertaining.
This talk was very generic (here's my thesis: the longer the title, the more generic the talk will be), interesting and fun, but at the end I'm unable to see anything new I'd have learned, apart from the distinction between Dynamic vs Static and Strong vs Weak typing, which I've seen the last year, but managed to forgot. This may be a very interesting argument for all those who are afraid of Groovy/Ruby, after bad experience with PHP or Perl.

Build Trust in Your Build to Deployment Flow!


Frederic Simon talked about DevOps and deployment, and that was a miss in my  schedule, because of two reasons. First, the talk was aimed at DevOps specifically, and while the subject is trendy lately, without big-scale problems, deployment is a process I usually set up and forget about. It just works, mostly because I only have to deal with one (current) project at a time. 
Not much love for Dart.
Second, while Frederic has a fabulous accent and a nice, loud voice, he tends to start each sentence loud and fade the sound at the end. This, together with mics failing him badly, made half of the presentation hard to grasp unless you were sitting in the first row.
I'm not saying the presentation was bad, far from it, it just clearly wasn't for me.
I've left a few minutes before the end, to see how many people came to Dart presentation by Mike West. I was kind of interested, since I'm following Warsaw Google Technology User Group and heard a few voices about why I should pay attentions to that new Google language. As you can see from the picture on the right, the majority tends to disagree with that opinion.

 

Non blocking, composable reactive web programming with Iteratees

Sadek Drobi's talk about Iteratees in Play 2.0 was very refreshing. Perhaps because I've never used Play before, but the presentation was flawless, with well explained problems, concepts and solutions.
Sadek started with a reflection on how much CPU we waste waiting for IO in web development, then moved to Play's Iteratees, to explain the concept and implementation, which while very different from the that overused Request/Servlet model, looked really nice and simple. I'm not sure though, how much the problem is present when you have a simple service, serving static content before your app server. Think apache (and faster) before tomcat. That won't fix the upload/download issue though, which is beautifully solved in Play 2.0

The Future of the Java Platform: Java SE 8 & Beyond


Simon Ritter is an intriguing fellow. If you take a glance at his work history (AT&T UNIX System Labs -> Novell -> Sun -> Oracle), you can easily see, he's a heavy weight player.
His presentation was rich in content, no corpo-bullshit. He started with a bit of history of JCP and how it looks like right now, then moved to the most interesting stuff, changes. Now I could give you a summary here, but there is really no point: you'd be much better taking look at the slides. There are only 48 of them, but everything is self-explanatory.
While I'm very disappointed with the speed of changes, especially when compared to the C# world, I'm glad with the direction and the fact that they finally want to BREAK the compatibility with the broken stuff (generics, etc.).  Moving to other languages I guess I won't be the one to scream "My god, finally!" somewhere in 2017, though. All the changes together look very promising, it's just that I'd like to have them like... now? Next year max, not near the heat death of the universe.

Simon also revealed one of the great mysteries of Java, to me:
The original idea behind JNI was to make it hard to write, to discourage people form using it.
On a side note, did you know Tegra3 has actually 5 cores? You use 4 of them, and then switch to the other one, when you battery gets low.

BOF: Spring and CloudFoundry


Having most of my folks moved to see "Typesafe stack 2.0" fabulously organized by Rafał Wasilewski and  Wojtek Erbetowski (with both of whom I had a pleasure to travel to the conference) and knowing it will be recorded, I've decided to see what Josh Long has to say about CloudFoundry, a subject I find very intriguing after the de facto fiasco of Google App Engine.

The audience was small but vibrant, mostly users of Amazon EC2, and while it turned out that Josh didn't have much, with pricing and details not yet public, the fact that Spring Source has already created their own competition (Could Foundry is both an Open Source app and a service), takes a lot from my anxiety.

For the review of the second day of the conference, go here.