Recommandé
Recommandé
Contenu connexe
Tendances
Tendances (20)
En vedette
En vedette (10)
Similaire à Large Scale ETL for Hadoop and Cloudera Search using Morphlines
Similaire à Large Scale ETL for Hadoop and Cloudera Search using Morphlines (20)
Dernier
Dernier (20)
Large Scale ETL for Hadoop and Cloudera Search using Morphlines
- 1. 1 Large Scale ETL for Hadoop and Cloudera Search using Morphlines Wolfgang Hoschek (@whoschek) Silicon Valley Java User Group Meetup Sept 2013
- 2. Agenda • Hadoop, ETL and Search – seLng the stage • Cloudera Morphlines Architecture • Component Deep Dive • Cloudera Search Use Cases • What’s next? Feel free to ask quesTons as we go!
- 3. Example ETL Use Case: Distributed Search on Hadoop Flume Hue UI Custom UI Custom App Solr Solr Solr SolrCloud query query query Index (ETL) Hadoop Cluster MR HDFS Index (ETL) HBase Index (ETL)
- 4. Cloudera Morphlines Architecture Solr Solr Solr SolrCloud Logs, tweets, social media, html, images, pdf, text…. Anything you want to index Flume, MR Indexer, HBase indexer, etc... Or your applicaTon! Morphline Library Morphlines can be embedded in any applicaTon… Your App!
- 5. Cloudera Morphlines • Open Source framework for simple ETL • Consume any kind of data from any kind of data source, process and load into any app or storage system • Designed for Near Real Time apps & Batch apps • Ships as part Cloudera Developer Kit (CDK) and Cloudera Search • It’s a Java library • ASL licensed on github hbps://github.com/cloudera/cdk • Similar to Unix pipelines, but more convenient & efficient • ConfiguraTon over coding (reduce Tme & skills) • Supports common file formats • Log Files & Text • Avro, Sequence file • JSON, HTML & XML • Etc… (pluggable) • Extensible set of transformaTon commands
- 6. ExtracTon, TransformaTon and Loading • Chain of pipelined commands • Simple and flexible data mapping & transformaTon • Reusable across mulTple index workloads • Over Tme, extend and re-‐ use across plaiorm workloads syslog Flume Agent Solr sink Command: readLine Command: grok Command: loadSolr Solr Event Record Record Record Document Morphline Library
- 7. Like a Unix Pipeline • Like Unix pipelines where the data model is generalized to work with streams of generic records, including arbitrary binary payloads • Designed to be embedded into Hadoop components such as Search, Flume, MapReduce, Pig, Hive, Sqoop
- 8. Stdlib + plugins • Framework ships with a set of frequently used high level transformaTon and I/O commands that can be combined in applicaTon specific ways • The plugin system allows the adding of new transformaTons and I/O commands and integrates exisTng funcTonality and third party systems in a straighiorward manne
- 9. Flexible Data Model • A record is a set of named fields where each field has an ordered list of one or more Java Objects (i.e. Guava’s ArrayListMulTmap) • Field can have mulTple values and any two records need not use common field names • Corresponds exactly to Solr/Lucene data model • Pass not only structured data, but also arbitrary binary data
- 10. Passing Binary Data • _abachment_body field (opTonal) • java.io.InputStream or Java byte[] • opTonal fields assist w/ detecTng & parsing data type • _abachment_mimetype field • e.g. "applicaTon/pdf" • _abachment_charset field • e.g. "UTF-‐8" • _abachment_name field • e.g. "cars.pdf” • Conceptually similar to email and HTTP headers/body
- 11. Processing Model • Morphline commands manipulate conTnuous or arbitrarily large streams of records • A command transforms a record into zero or more records • The output records of a command are passed to the next command in the chain • A command can contain nested commands • A morphline is a tree of commands, essenTally a push-‐based data flow engine
- 12. Processing Model Non-‐Goals • Designed to embedded into mulTple host systems, thus… • No noTon of persistence or durability or distributed compuTng or node failover • Basically just a chain of in-‐memory transformaTons in the current thread • No need to manage mulTple nodes or threads -‐ already covered by host systems such as MapReduce, Flume, Storm, Samza, etc. • However, a morphline does support passing noTficaTons • E.g. BEGIN_TRANSACTION, COMMIT_TRANSACTION, ROLLBACK_TRANSACTION, SHUTDOWN
- 13. Performance and Scaling • The runTme compiles morphline on the fly • The runTme processes all commands of a given morphline in the same thread • Piping a record from one command to another is fast • just a cheap Java method call • no queues, no handoffs among threads, no context switches, and no serializaTon between commands • For scalability, deploy many morphline instances on a cluster in many Flume agents and MapReduce tasks
- 14. Syntax • HOCON format (Human-‐OpTmized Config Object NotaTon) • Basically JSON slightly adjusted for the configuraTon file use case • Came out of typesafe.com • Also used by Akka and Play frameworks
- 15. Example: Indexing log4j w/ stacktraces juil. 25, 2012 10:49:40 AM hudson.triggers.SafeTimerTask run ok juil. 25, 2012 10:49:46 AM hudson.triggers.SafeTimerTask run failed com.amazonaws.AmazonClientException: Unable to calculate a request signature at com.amazonaws.auth.AbstractAWSSigner.signAndBase64Encode(AbstractAWSSigner.java:71) at java.util.TimerThread.run(Timer.java:505) Caused by: com.amazonaws.AmazonClientException: Unable to calculate a request signature at com.amazonaws.auth.AbstractAWSSigner.sign(AbstractAWSSigner.java:90) at com.amazonaws.auth.AbstractAWSSigner.signAndBase64Encode(AbstractAWSSigner.java:68) ... 14 more Caused by: java.lang.IllegalArgumentException: Empty key at javax.crypto.spec.SecretKeySpec.<init>(SecretKeySpec.java:96) at com.amazonaws.auth.AbstractAWSSigner.sign(AbstractAWSSigner.java:87) ... 15 more juil. 25, 2012 10:49:54 AM hudson.slaves.SlaveComputer tryReconnect Record 1 Record 2 Record 3
- 16. Example: Indexing log4j w/ stacktraces morphlines : [ { id : morphline1 importCommands : ["com.cloudera.**", "org.apache.solr.**"] commands : [ { readMultiLine { regex : "(^.+Exception: .+)|(^s+at .+)|(^s+... d+ more)|(^s*Caused by:.+)" what : previous charset : UTF-8 } } { logDebug { format : "output record: {}", args : ["@{}"] } } { loadSolr {} ] } ]
- 17. Example: Escape to Java Code morphlines : [ { id : morphline1 importCommands : ["com.cloudera.**", "org.apache.solr.**"] commands : [ { java { code: """ List tags = record.get("tags"); if (!tags.contains("hello")) { return false; } tags.add("world"); return child.process(record); """ } } ] } ]
- 18. Current Command Library • Integrate with and load into Apache Solr • Flexible log file analysis • Single-‐line record, mulT-‐line records, CSV files • Regex based pabern matching and extracTon • IntegraTon with Avro, JSON, XML, HTML • IntegraTon with Apache Hadoop Sequence Files • IntegraTon with SolrCell and all Apache Tika parsers • Auto-‐detecTon of MIME types from binary data using Apache Tika
- 19. Current Command Library (cont’d) • ScripTng support for dynamic java code • OperaTons on fields for assignment and comparison • OperaTons on fields with list and set semanTcs • if-‐then-‐else condiTonals • A small rules engine (tryRules) • String and Tmestamp conversions • slf4j logging • Yammer metrics and counters • Decompression and unpacking of arbitrarily nested container file formats • etc
- 20. Plugin Commands • Easy to add new I/O & transformaTon cmds • Integrate exisTng funcTonality and third party systems • Implement Java interface Command or subclass AbstractCommand • Add it to Java classpath • No registraTon or other administraTve acTon required
- 21. Morphline Example – syslog with grok morphlines : [ { id : morphline1 importCommands : ["com.cloudera.**", "org.apache.solr.**"] commands : [ { readLine {} } { grok { dicTonaryFiles : [/tmp/grok-‐dicTonaries] expressions : { message : """<%{POSINT:syslog_pri}>%{SYSLOGTIMESTAMP:syslog_Tmestamp} % {SYSLOGHOST:syslog_hostname} %{DATA:syslog_program}(?:[%{POSINT:syslog_pid}])?: % {GREEDYDATA:syslog_message}""" } } } { loadSolr {} } ] } ] Example Input <164>Feb 4 10:46:14 syslog sshd[607]: listening on 0.0.0.0 port 22 Output Record syslog_pri:164 syslog_Tmestamp:Feb 4 10:46:14 syslog_hostname:syslog syslog_program:sshd syslog_pid:607 syslog_message:listening on 0.0.0.0 port 22.
- 22. Example Java Driver Program /** Usage: java ... <morphline.conf> <dataFile1> ... <dataFileN> */ public static void main(String[] args) { // compile morphline.conf file on the fly File conf= new File(args[0]); MorphlineContext ctx= new MorphlineContext.Builder().build(); Command morphline = new Compiler().compile(conf, null, ctx, null); // process each input data file Notifications.notifyBeginTransaction(morphline); for (int i = 1; i < args.length; i++) { InputStream in = new FileInputStream(new File(args[i])); Record record = new Record(); record.put(Fields.ATTACHMENT_BODY, in); morphline.process(record); in.close(); } Notifications.notifyCommitTransaction(morphline); }
- 23. PotenTal New Plugin Commands • Extract, clean, transform, join, integrate, enrich and decorate records • Examples • join records with external data sources such as relaTonal databases, key-‐value stores, local files or IP Geo lookup tables. • Perform DNS resoluTon, expand shortened URLs • fetch linked metadata from social networks • do senTment analysis & annotate record accordingly • conTnuously maintain stats over sliding windows • compute exact or approx. disTnct values & quanTles
- 24. Example Command ImplementaTon (1/2) public final class ToStringBuilder implements CommandBuilder { @Override public Collection<String> getNames() { return Collections.singletonList("toString"); } @Override public Command build(Config config, Command parent, Command child, MorphlineContext context) { return new ToString(config, parent, child, context); } private static final class ToString extends AbstractCommand { @Override protected boolean doProcess(Record record) { // some custom processing goes here return super.doProcess(record); // pass to next command in chain } } }
- 25. Example Command ImplementaTon (2/2) private static final class ToString extends AbstractCommand { private final String fieldName; private final boolean trim; public ToString(Config config, Command parent, Command child, MorphlineContext context) { super(config, parent, child, context); this.fieldName = getConfigs().getString(config, "field"); this.trim = getConfigs().getBoolean(config, "trim", false); validateArguments(); } @Override protected boolean doProcess(Record record) { ListIterator iter = record.get(fieldName).listIterator(); while (iter.hasNext()) { String str = iter.next().toString(); iter.set(trim ? str.trim() : str); } return super.doProcess(record); // pass to next command in chain } } }
- 26. Use Case: Cloudera Search An Integrated Part of the Hadoop System One pool of data One security framework One set of system resources One management interface
- 27. What is Cloudera Search? • Full-‐text, interacTve search and faceted navigaTon • Batch, near real-‐Tme, and on-‐demand indexing • Apache Solr integrated with CDH • Established, mature search with vibrant community • Separate runTme like MapReduce, Impala • Incorporated as part of the Hadoop ecosystem • Open Source • 100% Apache, 100% Solr • Standard Solr APIs
- 28. ETL for Distributed Search on Apache Hadoop Flume Hue UI Custom UI Custom App Solr Solr Solr SolrCloud query query query Index (ETL) Hadoop Cluster MR HDFS Index (ETL) HBase Index (ETL)
- 29. Near Real Time ETL & Indexing with Flume Log File Apache Solr and Apache Flume • Data ingest at scale • Flexible extracTon and mapping • Indexing at data ingest • Packaged as Flume Morphline Solr Sink HDFS Flume Agent Indexer w/ Morphline Other Log File Flume Agent Indexer w/ Morphline 29 agent.sinks.solrSink.type = org.apache.flume.sink.solr.morphline.MorphlineSolrSink agent.sinks.solrSink.morphlineFile = /etc/flume-ng/conf/morphline.conf Flume.conf
- 30. Cloudera Manager Flume Morphline GUI 30
- 31. Scalable Batch ETL & Indexing Index shard Files Index shard Indexer w/ Morphline Files Solr server Indexer w/ Morphline Solr server 31 HDFS Solr and MapReduce • Flexible, scalable batch indexing • Start serving new indices with no downTme • On-‐demand indexing, cost-‐ efficient re-‐indexing • Packaged as MapReduceIndexerTool hadoop ... MapReduceIndexerTool --morphline-file morphline.conf ...
- 32. MapReduceIndexerTool 32 hadoop ... MapReduceIndexerTool --morphline-file morphline.conf ... S0_0_0 Extractors (Mappers) Leaf Shards (Reducers) Root Shards (Mappers) S0_0_1 S0S0_1_0 S0_1_1 S1_0_0 S1_0_1 S1S1_1_0 S1_1_1 Input Files ... ... ... ... • Morphline runs inside Mapper
- 33. Near Real Time indexing of Apache HBase HDFS HBase interacTve load Lily HBase Indexer(s) with Morphline Triggers on updates Solr server Solr server Solr server Solr server Solr server Search + = Large scale tabular data immediate access & updates fast & flexible informaDon discovery BIG DATA DATAMANAGEMENT
- 34. Batch & Near Real Time ETL Tweets Flume Solr Hue UI HDFS MapReduceIndexerTool, Impala, HBase, Mahout, EDW, MR, etc Lily HBase Indexer HdfsSink Query MapReduce IndexerTool Log Formats Social Media HTML Images PDF Custom UI Query Custom App ... Morphline Morphline MorphlineSink Morphline HBase OLTP
- 35. What’s next • More work on Apache Hbase IntegraTon • IntegraTon into Apache Hive & Sqoop • Stream AnalyTcs
- 36. Conclusion • Cloudera Development Kit w/ Morphlines • Open Source -‐ ASL License • Version 0.7.0 shipping • Extensive documentaTon • Send your quesTons and feedback to cdk-‐dev mailing list • Also ships integrated with Cloudera Search • Free QuickStart VM also available!