Intercepts all writes to specified directories and send them to apache kafka brokers. Quite suited for log centralization.

Packages for various distros can be installed from these repositories at openSUSE Build Service.

The following should install the new repositories then install fuse_kafka:

# curl -O \
    https://raw.githubusercontent.com/yazgoo/fuse_kafka/master/setup.sh \
    && md5sum -c <(echo "99c26578e926eb807a02d7a22c6c2e82  setup.sh") \
    && chmod +x setup.sh && ./setup.sh

(for more options - e.g. to install on a machine with no access to the repos - see 'setup.sh options' section)

A default configuration file is available in conf/fuse_kafka.properties. An explanation for each parameter is available in this file. The packages should install it in /etc/fuse_kafka.conf.

Here is a capture of a quickstart. (Download it - use ttyplay to view)

If you want to test fuse_kafka, using a clone of this repository.

On Debian and Ubuntu, you should install the following:

• librdkafka-dev
• librdkafka1
• libzookeeper-mt-dev
• libzookeeper-mt2
• libjansson-dev
• libjansson4
• python

First, build it:

$ ./build.py

On another terminal session, start zookeeper (this will download kafka):

$ ./build.py zookeeper_start

On another one, start kafka:

$ ./build.py kafka_start

The default configuration is conf/fuse_kafka.properties. An important piece of configuration is fuse_kafka_directories:

$ grep fuse_kafka_directories conf/fuse_kafka.properties -B2
# directories fuse_kafka will listen to (launch script will try to
# create them if they don't exist)
fuse_kafka_directories=["/tmp/fuse-kafka-test"]

Start fuse_kafka using the init script:

$ src/fuse_kafka.py start

If you're not running as root, you might have to make /etc/fuse.conf readable by your user (here to all users):

$ chmod a+r /etc/fuse.conf

And allow non-root user to specify the allow_other option, by adding a line with user_allow_other in /etc/fuse.conf.

If fuse_kafka is running, you should get the following output when running:

$ src/fuse_kafka.py status
listening on /tmp/fuse-kafka-test
fuse kafka is running

In yet another terminal, start a test consumer:

$ ./build.py kafka_consumer_start

Then start writing to a file under the overlay directory:

$ bash -c 'echo "foo"' > /tmp/fuse-kafka-test/bar

You should have an output from the consumer similar to this:

event:
    group: users
    uid: 1497
    @tags:
        -  test
    @fields:
         hostname: test
    @timestamp: 2014-10-03T09:07:04.000+0000
    pid: 6485
    gid: 604
    command: bash -c echo "foo"
    @message: foo
    path: /tmp/fuse-kafka-test/bar
    @version: 0.1.3
    user: yazgoo

When you're done, you can stop fuse_kafka:

$ src/fuse_kafka.py stop

First, commment fuse_kafka_quota in conf/fuse_kafka.properties. Then, start fuse kafka.

$ src/fuse_kafka.py start

Let's create a segfaulting program:

$ cat first.c
int main(void)
{
    *((int*)0) = 1;
}

$ gcc first.c

Then start a test consumer, displaying only the path and message_size-added fields

Launch the segfaulting program in fuse-kafka-test directory:

$ /path/to/a.out

A new core file should appear in fused directory.

Here is the consumer output:

$ SELECT="message_size-added path" ./build.py kafka_consumer_start
event:
    message_size-added: 4096
    path: /tmp/fuse-kafka-test/core
...
event:
    message_size-added: 4096
    path: /tmp/fuse-kafka-test/core

Here we see many messages.

Then, uncomment fuse_kafka_quota in conf/fuse_kafka.properties and launch the segfaulting program,

$ SELECT="message_size-added path" ./build.py kafka_consumer_start
event:
    message_size-added: 64
    path: /tmp/fuse-kafka-test/core

This time, we only receive the first write.

We use a logstash event, except the message and command are base64 encoded:

{"path": "/var/log/redis_6380.log", "pid": 1262, "uid": 0, "gid": 0,
"@message": "aGVsbG8gd29ybGQ=",
"@timestamp": "2014-09-11T14:19:09.000+0000","user": "root", "group":
"root",
"command": "L3Vzci9sb2NhbC9iaW4vcmVkaXMtc2VydmVyIC",
"@version": "0.1.2",
"@fields": {
    "first_field": "first_value",
    "second_field": "second_value" },
"@tags": ["mytag"]}

# installing prerequisites
$ sudo apt-get install librdkafka-dev libfuse-dev
# building
$ ./build.py 
# testing
$ ./build.py test
# cleaning
$ ./build.py clean
# installing:
$ ./build.py install

You can add c compiling flags via CFLAGS environment variable:

$ ./build.py CFLAGS=-Wall ./build.py

The start script allows to specify a command to append when actually launching the binary, FUSE_KAFKA_PREFIX, which can be used to perform analyses while running (like memcheck):

FUSE_KAFKA_PREFIX="valgrind --leak-check=yes" ./src/fuse_kafka.py start

Fuse-kafka must never make your filesystem accesses hang. Although this should be considered as a major bug, this might happen since the soft is still young. You can run a daemon so that any FS hanging is umounted (the check will occur every minute). To do so on an installed instance:

# service fuse_kafka_umounter start

To do so on a source based instance:

$ ./src/fuse_kafka.py start

Here are available options:

• -r: to do a remote install via ssh: -r private_ssh_key user@host
• -d: download and do not install the packages, generating an archive
• -f: install an archive already built via -d: -f fuse_kafka.tar.bz2

For example, this will download packages on a remote server:

./setup.sh -r mykey.pem root@myserver -d

This will generate an archive that will be copied locally. You can then install that archive via:

./setup.sh -f fuse_kafka.tar.bz2

A more realistic network setup test can be launched (as root) via:

sudo ./build.py mininet

This requires mininet.

This will launch kafka, zookeeper, fuse_kafka, and a consumer on their own mininet virtual hosts with their own network stacks. fuse_kafka is running on h3 (host number three). This will launch a mininet shell. For example, if you want to try and write on fuse_kafka host, issue a:

mininet> h3 echo lol > /tmp/fuse-kafka-test/xd

The consumer log is available via (/tmp/kafka_consumer.log).

quit or ^D will stop mininet and cleanup the virtual network.

A logstash input plugin to read from kafka is available in src/logstash/inputs/kafka.rb

Provided you have kafka installed in . (which ./build.py kafka_start should do), you can try it by downloading logastash and running:

/path/to/bin/logstash -p ./src/ -f ./conf/logstash.conf

To launch unit tests, issue a:

./build.py test

To test against multiple python versions (provided tox is installed), issue a:

tox

(see .travis.yml # prerequisites for tox to install these versions on ubuntu).

To run c unit tests, do a:

$ rm -rf out/ ; mkdir -p out/c ; ./build.py compile_test && ./build.py c_test

Basically, any process that has a file handle opened before fuse_kafka starts won't have it's writes captured. Such a process must open a new file handle after fuse_kafka startup, for example by restarting the process.

For example, If you're using rsyslogd and it is writing to /var/log/syslog, after starting fuse_kafka on /var/log, you should issue a:

service rsyslogd restart

After stopping fuse_kafka, you should also restart rsyslogd so it re-acquires a file descriptor on the actual FS.

Provided you have bonnie++ installed, you can run benchmarks with

./build.py bench

This will generate bench/results.js, which you can see via benchs/benchmarks.html

You might want to have fuse_kafka start ahead of most processes. But when it starts, you might not have all its configuration available yet. Or you might want to add brokers or use new zookeepers.

Dynamic configuration allows to modify the configuration on the fly. You won't be able to add watched directory on the fly, but you will be able to:

• point to new zookeepers/brokers
• update tags, fields

Just update your configuration, then, issue a:

sevice fuse_kafka reload

Or, if you using the developer version:

./src/fuse_kafka.py reload

To use this feature, you must make sure that /var/run/fuse_kafka.args is accessible to fuse_kafka.

licensed under Apache v 2.0, see LICENSE file

• for versions, use github release tags, for example 0.1.4
• for OBS source release number, we use ligthweight tag: for example the package fuse_kafka-0.1.4-20.1.x86_64.rpm will have a tag 0.1.4-20 (20 being the release number)

Please note that this project is released with a Contributor Code of Conduct. By participating in this project you agree to abide by its terms. See CODE OF CONDUCT file.