MapReduceIndexerTool
MapReduceIndexerTool is a MapReduce batch job driver that takes a morphline and creates a set of Solr index shards from a set of input files and writes the indexes into HDFS in a flexible, scalable, and fault-tolerant manner.
For more information on Morphlines, see:
- Extracting, Transforming, and Loading Data With Cloudera Morphlines for an introduction to Morphlines.
- Example Morphline Usage for morphline examples, discussion of those examples, and links to additional information.
MapReduceIndexerTool also supports merging the output shards into a set of live customer-facing Solr servers, typically a SolrCloud.
The indexer creates an offline index on HDFS in the output directory specified by the --output-dir parameter. If the --go-live parameter is specified, Solr merges the resulting offline index into the live running service. Thus, the Solr service must have read access to the contents of the output directory in order to complete the --go-live step. In an environment with restrictive permissions, such as one with an HDFS umask of 077, the Solr user may not be able to read the contents of the newly created directory. To address this issue, the indexer automatically applies the HDFS ACLs to enable Solr to read the output directory contents. These ACLs are only applied if HDFS ACLs are enabled on the HDFS NameNode. For more information, see HDFS Extended ACLs.
The indexer only makes ACL updates to the output directory and its contents. If the output directory's parent directories do not include the execute permission, the Solr service is not be able to access the output directory. Solr must have execute permissions from standard permissions or ACLs on the parent directories of the output directory.
MapReduceIndexerTool Input Splits
Different from some other indexing tools, the MapReduceIndexerTool does not operate on HDFS blocks as input splits. This means that when indexing a smaller number of large files, fewer nodes may be involved. For example, indexing two files that are each one GB results in two nodes acting as mappers. If these files were stored on a system with a 128 MB block size, other mappers might divide the work on the two files among 16 mappers, corresponding to the 16 HDFS blocks that store the two files.
This intentional design choice aligns with MapReduceIndexerTool supporting indexing non-splittable file formats such as JSON, XML, jpg, or log4j.
In theory, this could result in inefficient use of resources when a single node indexes a large file while many other nodes sit idle. In reality, this indexing strategy typically results in satisfactory performance in production environments because in most cases the number of files is large enough that work is spread throughout the cluster.
While dividing tasks by input splits does not present problems in most cases, users may still want to divide indexing tasks along HDFS splits. In that case, use the CrunchIndexerTool, which can work with Hadoop input splits using the input-file-format option.
Invoking Command-Line Help
- To invoke the command-line help in a default parcels installation, use:
$ hadoop jar /opt/cloudera/parcels/CDH-*/jars/search-mr-*-job.jar \ org.apache.solr.hadoop.MapReduceIndexerTool --help
- To invoke the command-line help in a default packages installation, use:
$ hadoop jar /usr/lib/solr/contrib/mr/search-mr-*-job.jar \ org.apache.solr.hadoop.MapReduceIndexerTool --help
usage: hadoop [GenericOptions]... jar search-mr-*-job.jar org.apache.solr.hadoop.MapReduceIndexerTool
[--help] --output-dir HDFS_URI [--input-list URI]
--morphline-file FILE [--morphline-id STRING] [--solr-home-dir DIR]
[--update-conflict-resolver FQCN] [--mappers INTEGER]
[--reducers INTEGER] [--max-segments INTEGER]
[--fair-scheduler-pool STRING] [--dry-run] [--log4j FILE]
[--verbose] [--show-non-solr-cloud] [--zk-host STRING] [--go-live]
[--collection STRING] [--go-live-threads INTEGER]
[HDFS_URI [HDFS_URI ...]]
MapReduce batch job driver that takes a morphline and creates a set of
Solr index shards from a set of input files and writes the indexes into
HDFS, in a flexible, scalable and fault-tolerant manner. It also supports
merging the output shards into a set of live customer facing Solr
servers, typically a SolrCloud. The program proceeds in several
consecutive MapReduce based phases, as follows:
1) Randomization phase: This (parallel) phase randomizes the list of
input files in order to spread indexing load more evenly among the
mappers of the subsequent phase.
2) Mapper phase: This (parallel) phase takes the input files, extracts
the relevant content, transforms it and hands SolrInputDocuments to a set
of reducers. The ETL functionality is flexible and customizable using
chains of arbitrary morphline commands that pipe records from one
transformation command to another. Commands to parse and transform a set
of standard data formats such as Avro, CSV, Text, HTML, XML, PDF, Word,
Excel, etc. are provided out of the box, and additional custom commands
and parsers for additional file or data formats can be added as morphline
plugins. This is done by implementing a simple Java interface that
consumes a record (e.g. a file in the form of an InputStream plus some
headers plus contextual metadata) and generates as output zero or more
records. Any kind of data format can be indexed and any Solr documents
for any kind of Solr schema can be generated, and any custom ETL logic
can be registered and executed.
Record fields, including MIME types, can also explicitly be passed by
force from the CLI to the morphline, for example: hadoop ... -D
morphlineField._attachment_mimetype=text/csv
3) Reducer phase: This (parallel) phase loads the mapper's
SolrInputDocuments into one EmbeddedSolrServer per reducer. Each such
reducer and Solr server can be seen as a (micro) shard. The Solr servers
store their data in HDFS.
4) Mapper-only merge phase: This (parallel) phase merges the set of
reducer shards into the number of solr shards expected by the user, using
a mapper-only job. This phase is omitted if the number of shards is
already equal to the number of shards expected by the user.
5) Go-live phase: This optional (parallel) phase merges the output shards
of the previous phase into a set of live customer facing Solr servers,
typically a SolrCloud. If this phase is omitted you can explicitly point
each Solr server to one of the HDFS output shard directories.
Fault Tolerance: Mapper and reducer task attempts are retried on failure
per the standard MapReduce semantics. On program startup all data in the
--output-dir is deleted if that output directory already exists. If the
whole job fails you can retry simply by rerunning the program again using
the same arguments.
positional arguments:
HDFS_URI HDFS URI of file or directory tree to index.
(default: [])
optional arguments:
--help, -help, -h Show this help message and exit
--input-list URI Local URI or HDFS URI of a UTF-8 encoded file
containing a list of HDFS URIs to index, one URI
per line in the file. If '-' is specified, URIs
are read from the standard input. Multiple --
input-list arguments can be specified.
--morphline-id STRING The identifier of the morphline that shall be
executed within the morphline config file
specified by --morphline-file. If the --
morphline-id option is ommitted the first (i.e.
top-most) morphline within the config file is
used. Example: morphline1
--solr-home-dir DIR Optional relative or absolute path to a local
dir containing Solr conf/ dir and in particular
conf/solrconfig.xml and optionally also lib/
dir. This directory will be uploaded to each MR
task. Example: src/test/resources/solr/minimr
--update-conflict-resolver FQCN
Fully qualified class name of a Java class that
implements the UpdateConflictResolver interface.
This enables deduplication and ordering of a
series of document updates for the same unique
document key. For example, a MapReduce batch job
might index multiple files in the same job where
some of the files contain old and new versions
of the very same document, using the same unique
document key.
Typically, implementations of this interface
forbid collisions by throwing an exception, or
ignore all but the most recent document version,
or, in the general case, order colliding updates
ascending from least recent to most recent
(partial) update. The caller of this interface
(i.e. the Hadoop Reducer) will then apply the
updates to Solr in the order returned by the
orderUpdates() method.
The default
RetainMostRecentUpdateConflictResolver
implementation ignores all but the most recent
document version, based on a configurable
numeric Solr field, which defaults to the
file_last_modified timestamp (default: org.
apache.solr.hadoop.dedup.
RetainMostRecentUpdateConflictResolver)
--mappers INTEGER Tuning knob that indicates the maximum number of
MR mapper tasks to use. -1 indicates use all map
slots available on the cluster. (default: -1)
--reducers INTEGER Tuning knob that indicates the number of
reducers to index into. -1 indicates use all
reduce slots available on the cluster. 0
indicates use one reducer per output shard,
which disables the mtree merge MR algorithm. The
mtree merge MR algorithm improves scalability by
spreading load (in particular CPU load) among a
number of parallel reducers that can be much
larger than the number of solr shards expected
by the user. It can be seen as an extension of
concurrent lucene merges and tiered lucene
merges to the clustered case. The subsequent
mapper-only phase merges the output of said
large number of reducers to the number of shards
expected by the user, again by utilizing more
available parallelism on the cluster. (default:
-1)
--max-segments INTEGER
Tuning knob that indicates the maximum number of
segments to be contained on output in the index
of each reducer shard. After a reducer has built
its output index it applies a merge policy to
merge segments until there are <= maxSegments
lucene segments left in this index. Merging
segments involves reading and rewriting all data
in all these segment files, potentially multiple
times, which is very I/O intensive and time
consuming. However, an index with fewer segments
can later be merged faster, and it can later be
queried faster once deployed to a live Solr
serving shard. Set maxSegments to 1 to optimize
the index for low query latency. In a nutshell,
a small maxSegments value trades indexing
latency for subsequently improved query latency.
This can be a reasonable trade-off for batch
indexing systems. (default: 1)
--fair-scheduler-pool STRING
Optional tuning knob that indicates the name of
the fair scheduler pool to submit jobs to. The
Fair Scheduler is a pluggable MapReduce
scheduler that provides a way to share large
clusters. Fair scheduling is a method of
assigning resources to jobs such that all jobs
get, on average, an equal share of resources
over time. When there is a single job running,
that job uses the entire cluster. When other
jobs are submitted, tasks slots that free up are
assigned to the new jobs, so that each job gets
roughly the same amount of CPU time. Unlike the
default Hadoop scheduler, which forms a queue of
jobs, this lets short jobs finish in reasonable
time while not starving long jobs. It is also an
easy way to share a cluster between multiple of
users. Fair sharing can also work with job
priorities - the priorities are used as weights
to determine the fraction of total compute time
that each job gets.
--dry-run Run in local mode and print documents to stdout
instead of loading them into Solr. This executes
the morphline in the client process (without
submitting a job to MR) for quicker turnaround
during early trial & debug sessions. (default:
false)
--log4j FILE Relative or absolute path to a log4j.properties
config file on the local file system. This file
will be uploaded to each MR task. Example:
/path/to/log4j.properties
--verbose, -v Turn on verbose output. (default: false)
--show-non-solr-cloud Also show options for Non-SolrCloud mode as part
of --help. (default: false)
Required arguments:
--output-dir HDFS_URI HDFS directory to write Solr indexes to. Inside
there one output directory per shard will be
generated. Example: hdfs://c2202.mycompany.
com/user/$USER/test
--morphline-file FILE Relative or absolute path to a local config file
that contains one or more morphlines. The file
must be UTF-8 encoded. Example:
/path/to/morphline.conf
Cluster arguments:
Arguments that provide information about your Solr cluster.
--zk-host STRING The address of a ZooKeeper ensemble being used
by a SolrCloud cluster. This ZooKeeper ensemble
will be examined to determine the number of
output shards to create as well as the Solr URLs
to merge the output shards into when using the --
go-live option. Requires that you also pass the
--collection to merge the shards into.
The --zk-host option implements the same
partitioning semantics as the standard SolrCloud
Near-Real-Time (NRT) API. This enables to mix
batch updates from MapReduce ingestion with
updates from standard Solr NRT ingestion on the
same SolrCloud cluster, using identical unique
document keys.
Format is: a list of comma separated host:port
pairs, each corresponding to a zk server.
Example: '127.0.0.1:2181,127.0.0.1:
2182,127.0.0.1:2183' If the optional chroot
suffix is used the example would look like:
'127.0.0.1:2181/solr,127.0.0.1:2182/solr,
127.0.0.1:2183/solr' where the client would be
rooted at '/solr' and all paths would be
relative to this root - i.e.
getting/setting/etc... '/foo/bar' would result
in operations being run on '/solr/foo/bar' (from
the server perspective).
If --solr-home-dir is not specified, the Solr
home directory for the collection will be
downloaded from this ZooKeeper ensemble.
Go live arguments:
Arguments for merging the shards that are built into a live Solr
cluster. Also see the Cluster arguments.
--go-live Allows you to optionally merge the final index
shards into a live Solr cluster after they are
built. You can pass the ZooKeeper address with --
zk-host and the relevant cluster information
will be auto detected. (default: false)
--collection STRING The SolrCloud collection to merge shards into
when using --go-live and --zk-host. Example:
collection1
--go-live-threads INTEGER
Tuning knob that indicates the maximum number of
live merges to run in parallel at one time.
(default: 1000)
Generic options supported are
--conf <configuration file>
specify an application configuration file
-D <property=value> use value for given property
--fs <local|namenode:port>
specify a namenode
--jt <local|jobtracker:port>
specify a job tracker
--files <comma separated list of files>
specify comma separated files to be copied to
the map reduce cluster
--libjars <comma separated list of jars>
specify comma separated jar files to include in
the classpath.
--archives <comma separated list of archives>
specify comma separated archives to be
unarchived on the compute machines.
The general command line syntax is
bin/hadoop command [genericOptions] [commandOptions]
Examples:
# (Re)index an Avro based Twitter tweet file:
sudo -u hdfs hadoop \
--config /etc/hadoop/conf.cloudera.mapreduce1 \
jar target/search-mr-*-job.jar org.apache.solr.hadoop.MapReduceIndexerTool \
-D 'mapred.child.java.opts=-Xmx500m' \
--log4j src/test/resources/log4j.properties \
--morphline-file ../search-core/src/test/resources/test-morphlines/tutorialReadAvroContainer.conf \
--solr-home-dir src/test/resources/solr/minimr \
--output-dir hdfs://c2202.mycompany.com/user/$USER/test \
--shards 1 \
hdfs:///user/$USER/test-documents/sample-statuses-20120906-141433.avro
# (Re)index all files that match all of the following conditions:
# 1) File is contained in dir tree hdfs:///user/$USER/solrloadtest/twitter/tweets
# 2) file name matches the glob pattern 'sample-statuses*.gz'
# 3) file was last modified less than 100000 minutes ago
# 4) file size is between 1 MB and 1 GB
# Also include extra library jar file containing JSON tweet Java parser:
hadoop jar target/search-mr-*-job.jar org.apache.solr.hadoop.HdfsFindTool \
-find hdfs:///user/$USER/solrloadtest/twitter/tweets \
-type f \
-name 'sample-statuses*.gz' \
-mmin -1000000 \
-size -100000000c \
-size +1000000c \
| sudo -u hdfs hadoop \
--config /etc/hadoop/conf.cloudera.mapreduce1 \
jar target/search-mr-*-job.jar org.apache.solr.hadoop.MapReduceIndexerTool \
--libjars /path/to/kite-morphlines-twitter-0.10.0.jar \
-D 'mapred.child.java.opts=-Xmx500m' \
--log4j src/test/resources/log4j.properties \
--morphline-file ../search-core/src/test/resources/test-morphlines/tutorialReadJsonTestTweets.conf \
--solr-home-dir src/test/resources/solr/minimr \
--output-dir hdfs://c2202.mycompany.com/user/$USER/test \
--shards 100 \
--input-list -
# Go live by merging resulting index shards into a live Solr cluster
# (explicitly specify Solr URLs - for a SolrCloud cluster see next example):
sudo -u hdfs hadoop \
--config /etc/hadoop/conf.cloudera.mapreduce1 \
jar target/search-mr-*-job.jar org.apache.solr.hadoop.MapReduceIndexerTool \
-D 'mapred.child.java.opts=-Xmx500m' \
--log4j src/test/resources/log4j.properties \
--morphline-file ../search-core/src/test/resources/test-morphlines/tutorialReadAvroContainer.conf \
--solr-home-dir src/test/resources/solr/minimr \
--output-dir hdfs://c2202.mycompany.com/user/$USER/test \
--shard-url http://solr001.mycompany.com:8983/solr/collection1 \
--shard-url http://solr002.mycompany.com:8983/solr/collection1 \
--go-live \
hdfs:///user/foo/indir
# Go live by merging resulting index shards into a live SolrCloud cluster
# (discover shards and Solr URLs through ZooKeeper):
sudo -u hdfs hadoop \
--config /etc/hadoop/conf.cloudera.mapreduce1 \
jar target/search-mr-*-job.jar org.apache.solr.hadoop.MapReduceIndexerTool \
-D 'mapred.child.java.opts=-Xmx500m' \
--log4j src/test/resources/log4j.properties \
--morphline-file ../search-core/src/test/resources/test-morphlines/tutorialReadAvroContainer.conf \
--output-dir hdfs://c2202.mycompany.com/user/$USER/test \
--zk-host zk01.mycompany.com:2181/solr \
--collection collection1 \
--go-live \
hdfs:///user/foo/indir
# MapReduce on Yarn - Pass custom JVM arguments (including a custom tmp directory)
HADOOP_CLIENT_OPTS='-DmaxConnectionsPerHost=10000 -DmaxConnections=10000 -Djava.io.tmpdir=/my/tmp/dir/'; \
sudo -u hdfs hadoop \
--config /etc/hadoop/conf.cloudera.mapreduce1 \
jar target/search-mr-*-job.jar org.apache.solr.hadoop.MapReduceIndexerTool \
-D 'mapreduce.map.java.opts=-DmaxConnectionsPerHost=10000 -DmaxConnections=10000' \
-D 'mapreduce.reduce.java.opts=-DmaxConnectionsPerHost=10000 -DmaxConnections=10000' \
--log4j src/test/resources/log4j.properties \
--morphline-file ../search-core/src/test/resources/test-morphlines/tutorialReadAvroContainer.conf \
--solr-home-dir src/test/resources/solr/minimr \
--output-dir hdfs://c2202.mycompany.com/user/$USER/test \
--shards 1 \
hdfs:///user/$USER/test-documents/sample-statuses-20120906-141433.avro
# MapReduce on MR1 - Pass custom JVM arguments (including a custom tmp directory)
HADOOP_CLIENT_OPTS='-DmaxConnectionsPerHost=10000 -DmaxConnections=10000 -Djava.io.tmpdir=/my/tmp/dir/'; \
sudo -u hdfs hadoop \
--config /etc/hadoop/conf.cloudera.mapreduce1 \
jar target/search-mr-*-job.jar org.apache.solr.hadoop.MapReduceIndexerTool \
-D 'mapred.child.java.opts=-DmaxConnectionsPerHost=10000 -DmaxConnections=10000' \
--log4j src/test/resources/log4j.properties \
--morphline-file ../search-core/src/test/resources/test-morphlines/tutorialReadAvroContainer.conf \
--solr-home-dir src/test/resources/solr/minimr \
--output-dir hdfs://c2202.mycompany.com/user/$USER/test \
--shards 1 \
hdfs:///user/$USER/test-documents/sample-statuses-20120906-141433.avro