Collect, aggregate, and move large amounts of data into HDFS.
Flume is a distributed, reliable, and available service for efficiently collecting, aggregating, and moving large amounts of log data. It has a simple and flexible architecture based on streaming data flows. It is robust and fault tolerant with tunable reliability mechanisms and many failover and recovery mechanisms. It uses a simple extensible data model that allows for online analytic application. Learn more at flume.apache.org.
This charm provides a Flume agent designed to ingest events into the shared
filesystem (HDFS) of a connected Hadoop cluster. It is meant to relate to
other Flume agents such as
A working Juju installation is assumed to be present. If Juju is not yet set up, please follow the getting-started instructions prior to deploying this charm.
This charm is intended to be deployed via one of the apache bigtop bundles. For example:
juju deploy hadoop-processing
Note: The above assumes Juju 2.0 or greater. If using an earlier version of Juju, use juju-quickstart with the following syntax:
juju quickstart hadoop-processing.
This will deploy an Apache Bigtop Hadoop cluster. More information about this deployment can be found in the bundle readme.
Now add Flume-HDFS and relate it to the cluster via the hadoop-plugin:
juju deploy apache-flume-hdfs flume-hdfs juju add-relation flume-hdfs plugin
The deployment at this stage isn't very exciting, as the
is waiting for other Flume agents to connect and send data. You'll probably
want to check out apache-flume-syslog or apache-flume-kafka
to provide additional functionality for this deployment.
flume-hdfs receives data, it is stored in a
HDFS subdirectory (configured by the connected Flume charm). You can quickly
verify the data written to HDFS using the command line. SSH to the
unit, locate an event, and cat it:
juju ssh flume-hdfs/0 hdfs dfs -ls /user/flume/<event_dir> # <-- find a date hdfs dfs -ls /user/flume/<event_dir>/<yyyy-mm-dd> # <-- find an event hdfs dfs -cat /user/flume/<event_dir>/<yyyy-mm-dd>/FlumeData.<id>
This process works well for data serialized in
text format (the default).
For data serialized in
avro format, you'll need to copy the file locally
and use the
dfs -text command. For example, replace the
dfs -cat command
from above with the following to view files stored in
hdfs dfs -copyToLocal /user/flume/<event_dir>/<yyyy-mm-dd>/FlumeData.<id> /home/ubuntu/myFile.txt hdfs dfs -text file:///home/ubuntu/myFile.txt
Charms can be deployed in environments with limited network access. To deploy in this environment, configure a Juju model with appropriate proxy and/or mirror options. See Configuring Models for more information.
- (string) The maximum number of events stored in the channel.
- (string) The maximum number of events the channel will take from a source or give to a sink per transaction.
- (int) The DFS replication value. The default (3) is the same default as the Namenode charm, but it may be overriden for this application.
- (string) Ingestion protocol for the agent source. Currently only 'avro' is supported.
- (string) URL from which to fetch resources (e.g., Flume binaries) instead of S3
- (int) Number of events written to file before it is rolled. A value of 0 (the default) means never roll based on number of events.
- (int) Number of seconds to wait before rolling the current file. Default will roll the file after 5 minutes. A value of 0 means never roll based on a time interval.
- (string) File size to trigger roll, in bytes. Default will roll the file once it reaches 10 MB. A value of 0 means never roll based on file size.
- (string) Compression codec for the agent sink. An empty value will write events to HDFS uncompressed. You may specify 'snappy' here to compress written events using the snappy codec.
- (string) Specify the serializer used when the sink writes to HDFS. Either 'avro_event' or 'text' are supported.
- (int) Port on which the agent source is listening.