Flume without agents on web server

[#1]

I am new to Flume and I want to use Flume to push the web server and application server logs into HDFS (these servers not part of Hadoop cluster). Is it mandatory to install Flume agent (client) on these servers? Can I pull logs from these servers to HDFS without installing Flume agents on them?

[#2 · Accepted solution]

You need some method to forward the logs to the flume agents. You could do something like rsyslog with the imfile input module (http://www.rsyslog.com/doc/v8-stable/configuration/modules/imfile.html ) to forward to a syslog source on the flume agents, or you could just install the standalone flume agent ( without the rest of the CDH) via rpms or tarball: http://www.cloudera.com/documentation/enterprise/release-notes/topics/cdh_vd_cdh_download.html

If you run the standalone flume agents, they could use spooldir or the new taildir source (in flume for CDH5.7) to monitor the files and forward via avro to flume agents within your cluster.

-pd

[#3 · Accepted solution]

You need some method to forward the logs to the flume agents. You could do something like rsyslog with the imfile input module (http://www.rsyslog.com/doc/v8-stable/configuration/modules/imfile.html ) to forward to a syslog source on the flume agents, or you could just install the standalone flume agent ( without the rest of the CDH) via rpms or tarball: http://www.cloudera.com/documentation/enterprise/release-notes/topics/cdh_vd_cdh_download.html

If you run the standalone flume agents, they could use spooldir or the new taildir source (in flume for CDH5.7) to monitor the files and forward via avro to flume agents within your cluster.

-pd

[#4]

Taildir source is documented here: http://archive.cloudera.com/cdh5/cdh/5/flume-ng/FlumeUserGuide.html#taildir-source

-pd

[#5]

Thank you for detail solution.

I am also wondering , how I can trigger Flume agent on cluster from Edge node (edge node not passwordless ssh to hadoop cluster) as I dont have permission to log in hadoop cluster.

[#6]

Please define what you mean by "Trigger Flume agent". Are you referring to starting the flume agent, or being able to deliver events to hdfs?

[#7]

I am executing Flumge-ng agnet command on Edge node.

As you already explained in another post, I need to run Flume as service on Edge node to start / stop flume agent.

[#8 · Accepted solution]

This documentation goes over stopping and starting flume when not using Cloudera Manager. This assumes you are running packages and not parcels on this edge node: http://www.cloudera.com/documentation/enterprise/latest/topics/cdh_ig_flume_run.html