fluent-plugin-cloudwatch-logs

CloudWatch Logs Plugin for Fluentd

Requirements

fluent-plugin-cloudwatch-logs	fluentd	ruby
>= 0.5.0	>= 0.14.15	>= 2.1
<= 0.4.5	~> 0.12.0 *	>= 1.9

• May not support all future fluentd features

Installation

$ gem install fluent-plugin-cloudwatch-logs

Preparation

Create IAM user with a policy like the following:

{
  "Version": "2012-10-17",
  "Statement": [
    {
      "Effect": "Allow",
      "Action": [
        "logs:*",
        "s3:GetObject"
      ],
      "Resource": [
        "arn:aws:logs:us-east-1:*:*",
        "arn:aws:s3:::*"
      ]
    }
  ]
}

Set region and credentials:

$ export AWS_REGION=us-east-1
$ export AWS_ACCESS_KEY_ID="YOUR_ACCESS_KEY"
$ export AWS_SECRET_ACCESS_KEY="YOUR_SECRET_ACCESS_KEY"

Example

Start fluentd:

$ fluentd -c example/fluentd.conf

Send sample log to CloudWatch Logs:

$ echo '{"hello":"world"}' | fluent-cat test.cloudwatch_logs.out

Fetch sample log from CloudWatch Logs:

# stdout
2014-07-17 00:28:02 +0900 test.cloudwatch_logs.in: {"hello":"world"}

Configuration

out_cloudwatch_logs

<match tag>
  @type cloudwatch_logs
  log_group_name log-group-name
  log_stream_name log-stream-name
  auto_create_stream true
  #message_keys key1,key2,key3,...
  #max_message_length 32768
  #use_tag_as_group false
  #use_tag_as_stream false
  #include_time_key true
  #localtime true
  #log_group_name_key group_name_key
  #log_stream_name_key stream_name_key
  #remove_log_group_name_key true
  #remove_log_stream_name_key true
  #put_log_events_retry_wait 1s
  #put_log_events_retry_limit 17
  #put_log_events_disable_retry_limit false
  #endpoint http://localhost:5000/
  #json_handler json
  #log_rejected_request true
</match>

• auto_create_stream: to create log group and stream automatically. (defaults to false)
• concurrency: use to set the number of threads pushing data to CloudWatch. (default: 1)
• endpoint: use this parameter to connect to the local API endpoint (for testing)
• http_proxy: use to set an optional HTTP proxy
• include_time_key: include time key as part of the log entry (defaults to UTC)
• json_handler: name of the library to be used to handle JSON data. For now, supported libraries are json (default) and yajl.
• localtime: use localtime timezone for include_time_key output (overrides UTC default)
• log_group_aws_tags: set a hash with keys and values to tag the log group resource
• log_group_aws_tags_key: use specified field of records as AWS tags for the log group
• log_group_name: name of log group to store logs
• log_group_name_key: use specified field of records as log group name
• log_rejected_request: output rejected_log_events_info request log. (defaults to false)
• log_stream_name: name of log stream to store logs
• log_stream_name_key: use specified field of records as log stream name
• max_events_per_batch: maximum number of events to send at once (default 10000)
• max_message_length: maximum length of the message
• message_keys: keys to send messages as events
• put_log_events_disable_retry_limit: if true, put_log_events_retry_limit will be ignored
• put_log_events_retry_limit: maximum count of retry (if exceeding this, the events will be discarded)
• put_log_events_retry_wait: time before retrying PutLogEvents (retry interval increases exponentially like put_log_events_retry_wait * (2 ^ retry_count))
• remove_log_group_aws_tags_key: remove field specified by log_group_aws_tags_key
• remove_log_group_name_key: remove field specified by log_group_name_key
• remove_log_stream_name_key: remove field specified by log_stream_name_key
• remove_retention_in_days: remove field specified by retention_in_days
• retention_in_days: use to set the expiry time for log group when created with auto_create_stream. (default to no expiry)
• retention_in_days_key: use specified field of records as retention period
• use_tag_as_group: to use tag as a group name
• use_tag_as_stream: to use tag as a stream name

in_cloudwatch_logs

<source>
  @type cloudwatch_logs
  tag cloudwatch.in
  log_group_name group
  log_stream_name stream
  #use_log_stream_name_prefix true
  state_file /var/lib/fluent/group_stream.in.state
  #endpoint http://localhost:5000/
  #json_handler json
</source>

• aws_sts_role_arn: the role ARN to assume when using cross-account sts authentication
• aws_sts_session_name: the session name to use with sts authentication (default: fluentd)
• aws_use_sts: use AssumeRoleCredentials to authenticate, rather than the default credential hierarchy. See 'Cross-Account Operation' below for more detail.
• endpoint: use this parameter to connect to the local API endpoint (for testing)
• fetch_interval: time period in seconds between checking CloudWatch for new logs. (default: 60)
• http_proxy: use to set an optional HTTP proxy
• json_handler: name of the library to be used to handle JSON data. For now, supported libraries are json (default) and yajl.
• log_group_name: name of log group to fetch logs
• log_stream_name: name of log stream to fetch logs
• state_file: file to store current state (e.g. next_forward_token)
• tag: fluentd tag
• use_log_stream_name_prefix: to use log_stream_name as log stream name prefix (default false)
• use_todays_log_stream: use todays and yesterdays date as log stream name prefix (formatted YYYY/MM/DD). (default: false)
• use_aws_timestamp: get timestamp from Cloudwatch event for non json logs, otherwise fluentd will parse the log to get the timestamp (default false)

Test

Set credentials:

$ export AWS_REGION=us-east-1
$ export AWS_ACCESS_KEY_ID="YOUR_ACCESS_KEY"
$ export AWS_SECRET_ACCESS_KEY="YOUR_SECRET_KEY"

Run tests:

$ rake test

Or, If you do not want to use IAM roll or ENV(this is just like writing to configuration file) :

$ rake aws_key_id=YOUR_ACCESS_KEY aws_sec_key=YOUR_SECRET_KEY region=us-east-1 test

If you want to run the test suite against a mock server, set endpoint as below:

$ export endpoint='http://localhost:5000/'
$ rake test

Caution

• If an event message exceeds API limit (256KB), the event will be discarded.

Cross-Account Operation

In order to have an instance of this plugin running in one AWS account to fetch logs from another account cross-account IAM authentication is required. Whilst this can be accomplished by configuring specific instances of the plugin manually with credentials for the source account in question this is not desirable for a number of reasons.

In this case IAM can be used to allow the fluentd instance in one account ("A") to ingest Cloudwatch logs from another ("B") via the following mechanic:

• plugin instance running in account "A" has an IAM instance role assigned to the underlying EC2 instance
• The IAM instance role and associated policies permit the EC2 instance to assume a role in another account
• An IAM role in account "B" and associated policies allow read access to the Cloudwatch Logs service, as appropriate.

IAM Detail: Consuming Account "A"

• Create an IAM role cloudwatch
• Attach a policy to allow the role holder to assume another role (where ACCOUNT-B is substituted for the appropriate account number):

{
    "Version": "2012-10-17",
    "Statement": [
        {
            "Effect": "Allow",
            "Action": [
                "sts:*"
            ],
            "Resource": [
                "arn:aws:iam::ACCOUNT-B:role/fluentd"
            ]
        }
    ]
}

• Ensure the EC2 instance on which this plugin is executing as role cloudwatch as its assigned IAM instance role.

IAM Detail: Log Source Account "B"

• Create an IAM role fluentd
• Ensure the fluentd role as account "A" as a trusted entity:

{
  "Version": "2012-10-17",
  "Statement": [
    {
      "Effect": "Allow",
      "Principal": {
        "AWS": "arn:aws:iam::ACCOUNT-A:root"
      },
      "Action": "sts:AssumeRole"
    }
  ]
}

• Attach a policy:

{
    "Version": "2012-10-17",
    "Statement": [
        {
            "Effect": "Allow",
            "Action": [
                "logs:DescribeDestinations",
                "logs:DescribeExportTasks",
                "logs:DescribeLogGroups",
                "logs:DescribeLogStreams",
                "logs:DescribeMetricFilters",
                "logs:DescribeSubscriptionFilters",
                "logs:FilterLogEvents",
                "logs:GetLogEvents"
            ],
            "Resource": [
                "arn:aws:logs:eu-west-1:ACCOUNT-B:log-group:LOG_GROUP_NAME_FOR_CONSUMPTION:*"
            ]
        }
    ]
}

Configuring the plugin for STS authentication

<source>
  @type cloudwatch_logs
  region us-east-1      # You must supply a region
  aws_use_sts true
  aws_sts_role_arn arn:aws:iam::ACCOUNT-B:role/fluentd
  log_group_name LOG_GROUP_NAME_FOR_CONSUMPTION
  log_stream_name SOME_PREFIX
  use_log_stream_name_prefix true
  state_file /path/to/state_file
  format /(?<message>.+)/
</source>

Using build-in placeholders, but they don't replace placeholders with actual values, why?

Built-in placeholders use buffer metadata when replacing placeholders with actual values. So, you should specify buffer attributes what you want to replace placeholders with.

Using ${tag} placeholders, you should specify tag attributes in buffer:

<buffer tag>
  @type memory
</buffer>

Using %Y%m%d placeholders, you should specify time attributes in buffer:

<buffer time>
  @type memory
  timekey 3600
</buffer>

In more detail, please refer to the officilal document for built-in placeholders.

TODO

• out_cloudwatch_logs
  • if the data is too big for API, split into multiple requests
  • format
  • check data size
• in_cloudwatch_logs
  • format
  • fallback to start_time because next_token expires after 24 hours

Contributing

1. Fork it ( https://github.com/[my-github-username]/fluent-plugin-cloudwatch-logs/fork )
2. Create your feature branch (git checkout -b my-new-feature)
3. Commit your changes (git commit -am 'Add some feature')
4. Push to the branch (git push origin my-new-feature)
5. Create a new Pull Request