Fluent Bit Operator

Fluent Bit Operator facilitates the deployment of Fluent Bit and provides great flexibility in building a logging layer based on Fluent Bit.

Once installed, the Fluent Bit Operator provides the following features:

Fluent Bit Management: Deploy and destroy Fluent Bit DaemonSet automatically.
Custom Configuration: Select input/filter/output plugins via labels.
Dynamic Reloading: Update configuration without rebooting Fluent Bit pods.

Overview
Get Started
Monitoring
API Doc
Best Practice
- Plugin Grouping
- Path Convention
Custom Parser
Roadmap
Development
- Requirements
- Running
Contributing
- Documentation
- Manifests

Overview

Fluent Bit Operator defines five custom resources using CustomResourceDefinition (CRD):

FluentBit: Defines the Fluent Bit DaemonSet and its configs. A custom Fluent Bit image kubesphere/fluent-bit is required to work with FluentBit Operator for dynamic configuration reloading.
FluentBitConfig: Select input/filter/output plugins and generates the final config into a Secret.
Input: Defines input config sections.
Parser: Defines parser config sections.
Filter: Defines filter config sections.
Output: Defines output config sections.

Each Input, Parser, Filter, Output represents a Fluent Bit config section, which are selected by FluentBitConfig via label selectors. The operator watches those objects, constructs the final config, and finally creates a Secret to store the config. This secret will be mounted into the Fluent Bit DaemonSet. The entire workflow looks like below:

To enable fluent-bit to pick up and use the latest config whenever the fluent-bit config changes, a wrapper called fluent-bit watcher is added to restart the fluent-bit process as soon as fluent-bit config changes are detected. This way the fluent-bit pod needn't be restarted to reload the new config. The fluent-bit config is reloaded in this way because there is no reload interface in fluent-bit itself. Please refer to this known issue for more details.

Get Started

Prerequisites

Kubernetes v1.16.13+ is necessary for running Fluent Bit Operator.

Install

Deploy Fluent Bit Operator with YAML

Install the latest stable version

kubectl apply -f https://raw.githubusercontent.com/kubesphere/fluentbit-operator/release-0.12/manifests/setup/setup.yaml

# You can change the namespace in manifests/setup/kustomization.yaml in corresponding release branch 
# and then use command below to install to another namespace
# kubectl kustomize manifests/setup/ | kubectl apply -f -

Install the development version

kubectl apply -f https://raw.githubusercontent.com/kubesphere/fluentbit-operator/master/manifests/setup/setup.yaml

# You can change the namespace in manifests/setup/kustomization.yaml 
# and then use command below to install to another namespace
# kubectl kustomize manifests/setup/ | kubectl apply -f -

Deploy Fluent Bit Operator with Helm

Note: For the Helm-based installation you need Helm v3.2.1 or later.

Fluent Bit Operator supports docker as well as containerd and CRI-O. containerd and CRI-O use the CRI Log format which is slightly different and requires additional parsing to parse JSON application logs. You should set different containerRuntime depending on your container runtime.

If your container runtime is docker

helm install fluentbit-operator  --create-namespace -n kubesphere-logging-system charts/fluentbit-operator/ --set containerRuntime=docker

If your container runtime is containerd

helm install fluentbit-operator --create-namespace -n kubesphere-logging-system charts/fluentbit-operator/  --set containerRuntime=containerd

If your container runtime is cri-o

helm install fluentbit-operator --create-namespace -n kubesphere-logging-system charts/fluentbit-operator/  --set containerRuntime=crio

Quick Start

The quick start instructs you to deploy fluent bit with dummy as input and stdout as output, which is equivalent to execute the binary with fluent-bit -i dummy -o stdout.

kubectl apply -f https://raw.githubusercontent.com/kubesphere/fluentbit-operator/master/manifests/quick-start/quick-start.yaml

Once everything is up, you'll observe messages in fluent bit pod logs like below:

[0] my_dummy: [1587991566.000091658, {"message"=>"dummy"}]
[1] my_dummy: [1587991567.000061572, {"message"=>"dummy"}]
[2] my_dummy: [1587991568.000056842, {"message"=>"dummy"}]
[3] my_dummy: [1587991569.000896217, {"message"=>"dummy"}]
[0] my_dummy: [1587991570.000172328, {"message"=>"dummy"}]

It means the FluentBit Operator works properly if you see the above messages, you can delete the quick start test by

kubectl delete -f https://raw.githubusercontent.com/kubesphere/fluentbit-operator/master/manifests/quick-start/quick-start.yaml

Configure Custom Watch Namespaces

When starting the operator, you can pass an optional set of namespaces to watch for resources in with the --watch-namespaces flag. It takes a comma-separated list of namespaces to watch:

...
      spec:
        containers:
          image: kubesphere/fluentbit-operator
        - args:
          - --watch-namespaces=namespace1,namespace2

Collect Kubernetes logs

This guide provisions a logging pipeline including the Fluent Bit DaemonSet and its log input/filter/output configurations to collect Kubernetes logs including container logs and kubelet logs.

Note that you need a running Elasticsearch v5+ cluster to receive log data before start. Remember to adjust output-elasticsearch.yaml to your own es setup. Kafka and Fluentd outputs are optional and are turned off by default.

Deploy the Kubernetes logging stack with YAML

kubectl apply -f manifests/logging-stack

# You can change the namespace in manifests/logging-stack/kustomization.yaml 
# and then use command below to install to another namespace
# kubectl kustomize manifests/logging-stack/ | kubectl apply -f -

Deploy the Kubernetes logging stack with Helm

If your container runtime is docker

helm upgrade fluentbit-operator --create-namespace -n kubesphere-logging-system charts/fluentbit-operator/  --set Kubernetes=true,containerRuntime=docker

If your container runtime is containerd

helm upgrade fluentbit-operator --create-namespace -n kubesphere-logging-system charts/fluentbit-operator/  --set Kubernetes=true,containerRuntime=containerd

If your container runtime is cri-o

helm upgrade fluentbit-operator --create-namespace -n kubesphere-logging-system charts/fluentbit-operator/  --set Kubernetes=true,containerRuntime=crio

Within a couple of minutes, you should observe an index available:

$ curl localhost:9200/_cat/indices
green open ks-logstash-log-2020.04.26 uwQuoO90TwyigqYRW7MDYQ 1 1  99937 0  31.2mb  31.2mb

Success!

Collect auditd logs

The Linux audit framework provides a CAPP-compliant (Controlled Access Protection Profile) auditing system that reliably collects information about any security-relevant (or non-security-relevant) event on a system. Refer to manifests/logging-stack/auditd, it supports a method for collecting audit logs from the Linux audit framework.

kubectl apply -f manifests/logging-stack/auditd

# You can change the namespace in manifests/logging-stack/auditd/kustomization.yaml 
# and then use command below to install to another namespace
# kubectl kustomize manifests/logging-stack/auditd/ | kubectl apply -f -

Within a couple of minutes, you should observe an index available:

$ curl localhost:9200/_cat/indices
green open ks-logstash-log-2021.04.06 QeI-k_LoQZ2h1z23F3XiHg  5 1 404879 0 298.4mb 149.2mb

Monitoring

Fluent Bit comes with a built-in HTTP Server. According to the official documentation of fluentbit You can enable this by enabling the HTTP server from the fluent bit configuration file:

[SERVICE]
    HTTP_Server  On
    HTTP_Listen  0.0.0.0
    HTTP_PORT    2020

When you use the kubesphere/fluentbit-operator, You can enable this from FluentBitConfig manifest. Example is below:

apiVersion: logging.kubesphere.io/v1alpha2
kind: FluentBitConfig
metadata:
  name: fluent-bit-config
  namespace: logging-system
spec:
  filterSelector:
    matchLabels:
      logging.kubesphere.io/enabled: 'true'
  inputSelector:
    matchLabels:
      logging.kubesphere.io/enabled: 'true'
  outputSelector:
    matchLabels:
      logging.kubesphere.io/enabled: 'true'
  service:
    httpListen: 0.0.0.0
    httpPort: 2020
    httpServer: true
    parsersFile: parsers.conf

Once HTTP server is enabled, you should be able to get the information:

curl <podIP>:2020 | jq .

{
  "fluent-bit": {
    "version": "1.8.3",
    "edition": "Community",
    "flags": [
      "FLB_HAVE_PARSER",
      "FLB_HAVE_RECORD_ACCESSOR",
      "FLB_HAVE_STREAM_PROCESSOR",
      "FLB_HAVE_TLS",
      "FLB_HAVE_OPENSSL",
      "FLB_HAVE_AWS",
      "FLB_HAVE_SIGNV4",
      "FLB_HAVE_SQLDB",
      "FLB_HAVE_METRICS",
      "FLB_HAVE_HTTP_SERVER",
      "FLB_HAVE_SYSTEMD",
      "FLB_HAVE_FORK",
      "FLB_HAVE_TIMESPEC_GET",
      "FLB_HAVE_GMTOFF",
      "FLB_HAVE_UNIX_SOCKET",
      "FLB_HAVE_PROXY_GO",
      "FLB_HAVE_JEMALLOC",
      "FLB_HAVE_LIBBACKTRACE",
      "FLB_HAVE_REGEX",
      "FLB_HAVE_UTF8_ENCODER",
      "FLB_HAVE_LUAJIT",
      "FLB_HAVE_C_TLS",
      "FLB_HAVE_ACCEPT4",
      "FLB_HAVE_INOTIFY"
    ]
  }
}

API Doc

The list below shows supported plugins which are based on Fluent Bit v1.7.x+. For more information, please refer to the API docs of each plugin.

Input
- dummy
- tail
- systemd
Parser
- json
- logfmt
- lstv
- regex
Filter
- kubernetes
- modify
- nest
- parser
- grep
- record modifier
- lua
- throttle
Output
- elasticsearch
- file
- forward
- http
- kafka
- null
- stdout
- tcp
- loki
- syslog
- datadog

Best Practice

Plugin Grouping

Input, filter, and output plugins are connected by label selectors. For input and output plugins, always create Input or Output CRs for every plugin. Don't aggregate multiple inputs or outputs into one Input or Output object, except you have a good reason to do so. Take the demo logging stack for example, we have one yaml file for each output.

However, for filter plugins, if you want a filter chain, the order of filters matters. You need to organize multiple filters into an array as the demo logging stack suggests.

Path Convention

Path to file in Fluent Bit config should be well regulated. Fluent Bit Operator adopts the following convention internally.

Dir Path	Description
/fluent-bit/tail	Stores tail related files, eg. file tracking db. Using fluentbit.spec.positionDB will mount a file `pos.db` under this dir by default.
/fluent-bit/secrets/{secret_name}	Stores secrets, eg. TLS files. Specify secrets to mount in fluentbit.spec.secrets, then you have access.
/fluent-bit/config	Stores the main config file and user-defined parser config file.

Note that ServiceAccount files are mounted at /var/run/secrets/kubernetes.io/serviceaccount.

Custom Parser

To enable parsers, you must set the value of FluentBitConfig.Spec.Service.ParsersFile to parsers.conf. Your custom parsers will be included into the built-in parser config via @INCLUDE /fluent-bit/config/parsers.conf. Note that the parsers.conf contains a few built-in parsers, for example, docker. Read parsers.conf for more information.

Check out the demo in the folder /manifests/regex-parser for how to use a custom regex parser.

Roadmap

Support containerd log format
Add Fluentd CRDs as the log aggregation layer with group name fluentd.fluent.io
Add FluentBit Cluster CRDs with new group name fluentbit.fluent.io
Rename the entire project to Fluent Operator
Support more Fluentd & FluentBit plugins

Development

Requirements

golang v1.16+.requirement
kubectl v1.16.13+.
kubebuilder v2.3+ (the project is build with v2.3.2)
Access to a Kubernetes cluster v1.16.13+

Running

Install CRDs: make install
Run: make run

Contributing

Documentation

API Doc is generated automatically. To modify it, edit the comment above struct fields, then run go run cmd/doc-gen/main.go.

Manifests

Most files under the folder manifests/setup are automatically generated from config. Don't edit them directly, run make manifests instead, then replace these files accordingly.

nevosial/fluentbit-operator