Home

Awesome

Fluentd Daemonset for Kubernetes

[!CAUTION] README.md is generated from templates/README.md.erb

Docker Stars Docker Pulls ImageLayers Size ImageLayers Layers

Supported tags and respective Dockerfile links

See also dockerhub tags page: https://hub.docker.com/r/fluent/fluentd-kubernetes-daemonset/tags

Debian

Current stable

[!TIP] Since v1.17.0, the container image build process has been migrated from automated builds on hub.docker.com to GitHub Actions. This is because there were limitation about the number of automated builds on hub.docker.com. Now, there is no limitation about the number of build pipelines.

Note that there were some restrictions to ship daemonset images for v1.16.5 or older ones before:

Multi-Arch images
x86_64 images
arm64 images

You can also use v1-debian-PLUGIN tag to refer latest v1 image, e.g. v1-debian-elasticsearch. On production, strict tag is better to avoid unexpected update.

See dockerhub's tags page for older tags.

Old stable

v0.12 development has been ended. These images are never updated.

Alpine Linux (This is deprecated. Use Debian images instead)

What is Fluentd?

Fluentd Logo

Fluentd is an open source data collector, which lets you unify the data collection and consumption for a better use and understanding of data.

www.fluentd.org

Image versions

Fluentd versioning is as follows:

SeriesDescription
v1.xcurrent stable
v0.12Old stable, no longer updated

Settings

Default image version

Default YAML uses latest v1 images like fluent/fluentd-kubernetes-daemonset:v1-debian-kafka. If you want to avoid unexpected image update, specify exact version for image like fluent/fluentd-kubernetes-daemonset:v1.8.0-debian-kafka-1.0.

Run as root

This is for v0.12 images.

In Kubernetes and default setting, fluentd needs root permission to read logs in /var/log and write pos_file to /var/log. To avoid permission error, you need to set FLUENT_UID environment variable to 0 in your Kubernetes configuration.

Use your configuration

These images have default configuration and support some environment variables for parameters but it sometimes doesn't fit your case. If you want to use your configuration, use ConfigMap feature.

Each image has following configurations:

Overwrite conf file via ConfigMap. See also several examples:

Use CRI parser for containerd/cri-o logs

This feature is available since v1.12.0-xxx-1.1.

By default, these images use json parser for /var/log/containers/ files because docker generates json formatted logs. On the other hand, containerd/cri-o use different log format. To parse such logs, you need to use cri parser instead.

You can use cri parser by overwriting tail_container_parse.conf via ConfigMap.

# configuration example
<parse>
  @type cri
</parse>

See also CRI parser README

Use FLUENT_CONTAINER_TAIL_PATH to change container logs folder

You can update the default path for the container logs i.e /var/log/container/*.log and also one can add multiple path as defined in this fluentd document https://docs.fluentd.org/input/tail#path

Use FLUENT_CONTAINER_TAIL_EXCLUDE_PATH to exclude specific container logs

Since v1.9.3 or later images.

You can exclude container logs from /var/log/containers/ with FLUENT_CONTAINER_TAIL_EXCLUDE_PATH. If you have a trouble with specific log, use this envvar, e.g. ["/var/log/containers/logname-*"].

Use LD_PRELOAD to enable jemalloc memory allocator

Since v1.17.0-1.3/v1.16.5-1.3, jemalloc memory allocator is disabled by default. This is because that combination of systemd plugin and jemalloc memory allocator causes a crash bug. (free(): invalid pointer in typical)

If you don't use systemd plugin at all, you can enable jemalloc memory allocator explicitly via env: parameter.

env:
  ...
  LD_PRELOAD="/usr/lib/libjemalloc.so.2"

Disable systemd input

If you don't setup systemd in the container, fluentd shows following messages by default configuration.

[warn]: #0 [in_systemd_bootkube] Systemd::JournalError: No such file or directory retrying in 1s
[warn]: #0 [in_systemd_kubelet] Systemd::JournalError: No such file or directory retrying in 1s
[warn]: #0 [in_systemd_docker] Systemd::JournalError: No such file or directory retrying in 1s

You can suppress these messages by setting disable to FLUENTD_SYSTEMD_CONF environment variable in your kubernetes configuration.

Disable prometheus input plugins

By default, latest images launch prometheus plugins to monitor fluentd. You can disable prometheus input plugin by setting disable to FLUENTD_PROMETHEUS_CONF environment variable in your kubernetes configuration.

Disable sed execution on elasticsearch image

This is for older images. Latest elasticsearch images don't use sed.

By historical reason, elasaticsearch image executes sed command during startup phase when FLUENT_ELASTICSEARCH_USER or FLUENT_ELASTICSEARCH_PASSWORD is specified. This sometimes causes a problem with read only mount. To avoid this problem, set "true" to FLUENT_ELASTICSEARCH_SED_DISABLE environment variable in your kubernetes configuration.

Running on OpenShift

This daemonset setting mounts /var/log as service account fluentd so you need to run containers as privileged container. Here is command example:

oc project kube-system
oc create -f https://raw.githubusercontent.com/fluent/fluentd-kubernetes-daemonset/master/fluentd-daemonset-elasticsearch-rbac.yaml
oc adm policy add-scc-to-user privileged -z fluentd
oc patch ds fluentd -p "spec:
  template:
    spec:
      containers:
      - name: fluentd
        securityContext:
          privileged: true"
oc delete pod -l k8s-app=fluentd-logging

This is from nekop's japanese article.

Multiple fluentd

When you want to have multiple fluentd. For example push to multiple destination like: elsticsearch + S3. You need to use FLUENT_POS_EXTRA_DIR add additional directory for pos file. Otherwise they share same pos file. You may found some log only on one destination.

Note

kafka image doesn't support zookeeper parameters

zookeeper gem doesn't work on Debian 10, so kafka image doesn't include zookeeper gem.

Windows k8s daemonset not supported in this repository

Maintainers don't have k8s experience on Windows. Some users create k8s daemonset on Windows:

Please check them out.

kafka image suggestion

Using debian-kafka2/debian-kafka2-arm64 images are better than using debian-kafka/debian-kafka-arm64 images. Because debian-kafka2/debian-kafka2-arm64 images use out_kafka2 plugin but debian-kafka/debian-kafka-arm64 images use deprecated out_kafka_buffered plugin.

Maintainers

Some images are contributed by users. If you have a problem/question for following images, ask it to contributors.

Currently, we don't accept new destination request without contribution. See https://github.com/fluent/fluentd-kubernetes-daemonset/issues/293

References

Kubernetes Logging with Fluentd

Issues

We can't notice comments in the DockerHub so don't use them for reporting issues or asking question.

If you have any problems with or questions about this image, please contact us through a GitHub issue.

Pull Request

Update templates files instead of docker-image files. docker-image files are automatically generated from templates.

Note: This file is generated from templates/README.md.erb