The Kubernetes-native platform (v2).
The Package manager for Kubernetes.
The Kubernetes-native Service Broker.
Fluentd runs on every worker node of the cluster and is deployed as a Daemon Set. The Fluentd pods capture all of the stderr and stdout streams of every container running on the host (even those not hosted directly by kubernetes). Once the log message arrives in our custom fluentd plugin we determine where the message originated.
If the message is from the Workflow Router we build an Influxdb compatible message and send it to the same NSQD instance but instead place the message on the metrics topic.
Logger then acts as a consumer reading messages off of the NSQ logs topic storing those messages in a local Redis instance. When a user wants to retrieve log entries using the
deis logs command we make an HTTP request from Controller to Logger which then fetches the appropriate data from Redis.
Even though we provide a redis instance with the default Workflow install, it is recommended that operators use a third-party source like Elasticache or similar offering. This way your data is durable across upgrades or outages. If you have a third-party Redis installation you would like to use all you need to do is set the following values in your helm chart:
These can be changed by running
helm inspect values hephy/workflow > values.yaml before using
helm install to complete the installation. To customize the redis credentials, edit
and modify the
redis section of the file to tune these settings.
deis logs command encounters an error it will return the following message:
Error: There are currently no log messages. Please check the following things: 1) Logger and fluentd pods are running. 2) The application is writing logs to the logger component by checking that an entry in the ring buffer was created: kubectl --namespace=deis logs <logger pod> 3) Making sure that the container logs were mounted properly into the fluentd pod: kubectl --namespace=deis exec <fluentd pod> ls /var/log/containers
┌────────┐ │ Router │ ┌────────┐ ┌─────┐ └────────┘ │ Logger │◀───▶│Redis│ │ └────────┘ └─────┘ Log file ▲ │ │ ▼ │ ┌────────┐ ┌─────────┐ logs/metrics ┌─────┐ │App Logs│──Log File──▶│ fluentd │───────topics─────▶│ NSQ │ └────────┘ └─────────┘ └─────┘ │ │ ┌─────────────┐ │ │ HOST │ ▼ │ Telegraf │───┐ ┌────────┐ └─────────────┘ │ │Telegraf│ │ └────────┘ ┌─────────────┐ │ │ │ HOST │ │ ┌───────────┐ │ │ Telegraf │───┼───▶│ InfluxDB │◀────Wire ───────────┘ └─────────────┘ │ └───────────┘ Protocol │ ▲ ┌─────────────┐ │ │ │ HOST │ │ ▼ │ Telegraf │───┘ ┌──────────┐ └─────────────┘ │ Grafana │ └──────────┘
By default the Fluentd pod can be configured to talk to numerous syslog endpoints. So for example it is possible to have Fluentd send log messages to both the Logger component and Papertrail. This allows production deployments of Deis to satisfy stringent logging requirements such as offsite backups of log data.
Configuring Fluentd to talk to multiple syslog endpoints means modifying the Fluentd daemonset
manifest. This means you will need to fetch the chart with
helm fetch hephy/workflow --untar, then
workflow/charts/fluentd/templates/logger-fluentd-daemon.yaml with the following:
env: - name: "SYSLOG_HOST_1" value: "my.syslog.host" - name: "SYSLOG_PORT_1" value: "5144" .... - name: "SYSLOG_HOST_N" value: "my.syslog.host.n" - name: "SYSLOG_PORT_N" value: "51333"
If you only need to talk to 1 Syslog endpoint you can use the following configuration within your chart:
env: - name: "SYSLOG_HOST" value: "my.syslog.host" - name: "SYSLOG_PORT" value: "5144"
helm install ./workflow --namespace deis to install the modified chart.
We currently support logging information to Syslog, Elastic Search, and Sumo Logic. However, we will gladly accept pull requests that add support to other locations. For more information please visit the fluentd repository.
That are many output plugins available for Fluentd. But, we purposefully do not ship our Fluentd image with these installed. Instead, we provide a mechanism that allows users to install a plugin at startup time of the container and configure it.
If you would like to install a plugin you can set an environment variable such as the following:
FLUENTD_PLUGIN_N=some-fluentd-plugin where N is a positive integer that is incremented for every plugin you wish to install. After you set this value you must then set the configuration text for the
STORE plugin you are installing. You can do that by setting
CUSTOM_STORE_N=configuration text where N is the corresponding index value of the plugin you just installed.
Here is an example of setting the values directly in the manifest of the daemonset.
env: - name: "FLUENTD_PLUGIN_1" value: "fluent-plugin-kafka" - name: "CUSTOM_STORE_1" value: | <store> @type kafka \ default_topic some_topic </store>
Or you could configure it using the
daemon_environment key in the
fluentd: daemon_environment: FLUENTD_PLUGIN_1: "fluent-plugin-kafka" CUSTOM_STORE_1: "|\n <store>\n @type kafka\n default_topic some_topic\n </store>" INSTALL_BUILD_TOOLS: "|\n true"
For more information please see the Custom Plugins section of the README.