Send logs and metrics to Amazon CloudWatch
The Amazon CloudWatch output plugin allows to ingest your records into the CloudWatch Logs service. Support for CloudWatch Metrics is also provided via EMF.
This is the documentation for the core Fluent Bit CloudWatch plugin written in C. It can replace the aws/amazon-cloudwatch-logs-for-fluent-bit Golang Fluent Bit plugin released last year. The Golang plugin was named cloudwatch; this new high performance CloudWatch plugin is called cloudwatch_logs to prevent conflicts/confusion. Check the amazon repo for the Golang plugin for details on the deprecation/migration plan for the original plugin.
See here for details on how AWS credentials are fetched.
In order to send records into Amazon Cloudwatch, you can run the plugin from the command line or through the configuration file:
The cloudwatch plugin, can read the parameters from the command line through the -p argument (property), e.g:
In your main configuration file append the following Output section:
The following AWS IAM permissions are required to use this plugin:
Fluent Bit 1.7 adds a new feature called workers which enables outputs to have dedicated threads. This cloudwatch_logs plugin has partial support for workers. The plugin can support a single worker; enabling multiple workers will lead to errors/indeterminate behavior.
Example:
If you enable a single worker, you are enabling a dedicated thread for your CloudWatch output. We recommend starting without workers, evaluating the performance, and then enabling a worker if needed. For most users, the plugin can provide sufficient throughput without workers.
Sometimes, you may want the log group or stream name to be based on the contents of the log record itself. This plugin supports templating log group and stream names using Fluent Bit record_accessor syntax.
Here is an example usage, for a common use case- templating log group and stream names based on Kubernetes metadata.
Recall that the kubernetes filter can add metadata which will look like the following:
Using record_accessor, we can build a template based on this object.
Here is our output configuration:
With the above kubernetes metadata, the log group name will be application-logs-ip-10-1-128-166.us-east-2.compute.internal.my-namespace. And the log stream name will be myapp-5468c5d4d7-n2swr.myapp.
If the kubernetes structure is not found in the log record, then the log_group_name and log_stream_prefix will be used instead, and Fluent Bit will log an error like:
Notice in the example above, that the template values are separated by dot characters. This is important; the Fluent Bit record_accessor library has a limitation in the characters that can separate template variables- only dots and commas (. and ,) can come after a template variable. This is because the templating library must parse the template and determine the end of a variable.
Assume that your log records contain the metadata keys container_name and task. The following would be invalid templates because the two template variables are not separated by commas or dots:
$task-$container_name
$task/$container_name
$task_$container_name
$taskfooo$container_name
However, the following are valid:
$task.$container_name
$task.resource.$container_name
$task.fooo.$container_name
And the following are valid since they only contain one template variable with nothing after it:
fooo$task
fooo____$task
fooo/bar$container_name
Fluent Bit has different input plugins (cpu, mem, disk, netif) to collect host resource usage metrics. cloudwatch_logs output plugin can be used to send these host metrics to CloudWatch in Embedded Metric Format (EMF). If data comes from any of the above mentioned input plugins, cloudwatch_logs output plugin will convert them to EMF format and sent to CloudWatch as JSON log. Additionally, if we set json/emf as the value of log_format config option, CloudWatch will extract custom metrics from embedded JSON payload.
Note: Right now, only cpu and mem metrics can be sent to CloudWatch.
For using the mem input plugin and sending memory usage metrics to CloudWatch, we can consider the following example config file. Here, we use the aws filter which adds ec2_instance_id and az (availability zone) to the log records. Later, in the output config section, we set ec2_instance_id as our metric dimension.
The following config will set two dimensions to all of our metrics- ec2_instance_id and az.
Amazon distributes a container image with Fluent Bit and these plugins.
github.com/aws/aws-for-fluent-bit
Our images are available in Amazon ECR Public Gallery. You can download images with different tags by following command:
For example, you can pull the image with latest version by:
If you see errors for image pull limits, try log into public ECR with your AWS credentials:
You can check the Amazon ECR Public official doc for more details
You can use our SSM Public Parameters to find the Amazon ECR image URI in your region:
For more see the AWS for Fluent Bit github repo.
| Key | Description |
|---|---|
region
The AWS region.
log_group_name
The name of the CloudWatch Log Group that you want log records sent to.
log_group_template
Template for Log Group name using Fluent Bit record_accessor syntax. This field is optional and if configured it overrides the log_group_name. If the template translation fails, an error is logged and the log_group_name (which is still required) is used instead. See the tutorial below for an example.
log_stream_name
The name of the CloudWatch Log Stream that you want log records sent to.
log_stream_prefix
Prefix for the Log Stream name. The tag is appended to the prefix to construct the full log stream name. Not compatible with the log_stream_name option.
log_stream_template
Template for Log Stream name using Fluent Bit record_accessor syntax. This field is optional and if configured it overrides the other log stream options. If the template translation fails, an error is logged and the log_stream_name or log_stream_prefix are used instead (and thus one of those fields is still required to be configured). See the tutorial below for an example.
log_key
By default, the whole log record will be sent to CloudWatch. If you specify a key name with this option, then only the value of that key will be sent to CloudWatch. For example, if you are using the Fluentd Docker log driver, you can specify log_key log and only the log message will be sent to CloudWatch.
log_format
An optional parameter that can be used to tell CloudWatch the format of the data. A value of json/emf enables CloudWatch to extract custom metrics embedded in a JSON payload. See the Embedded Metric Format.
role_arn
ARN of an IAM role to assume (for cross account access).
auto_create_group
Automatically create the log group. Valid values are "true" or "false" (case insensitive). Defaults to false.
log_retention_days
If set to a number greater than zero, and newly create log group's retention policy is set to this many days. Valid values are: [1, 3, 5, 7, 14, 30, 60, 90, 120, 150, 180, 365, 400, 545, 731, 1827, 3653]
endpoint
Specify a custom endpoint for the CloudWatch Logs API.
metric_namespace
An optional string representing the CloudWatch namespace for the metrics. See Metrics Tutorial section below for a full configuration.
metric_dimensions
A list of lists containing the dimension keys that will be applied to all metrics. The values within a dimension set MUST also be members on the root-node. For more information about dimensions, see Dimension and Dimensions. In the fluent-bit config, metric_dimensions is a comma and semicolon separated string. If you have only one list of dimensions, put the values as a comma separated string. If you want to put list of lists, use the list as semicolon separated strings. For example, if you set the value as 'dimension_1,dimension_2;dimension_3', we will convert it as [[dimension_1, dimension_2],[dimension_3]]
sts_endpoint
Specify a custom STS endpoint for the AWS STS API.
auto_retry_requests
Immediately retry failed requests to AWS services once. This option does not affect the normal Fluent Bit retry mechanism with backoff. Instead, it enables an immediate retry with no delay for networking errors, which may help improve throughput when there are transient/random networking issues. This option defaults to true.
external_id
Specify an external ID for the STS API, can be used with the role_arn parameter if your role requires an external ID.