Monitoring

Learn how to monitor your Fluent Bit data pipelines

Fluent Bit comes with built-it features to allow you to monitor the internals of your pipeline, connect to Prometheus and Grafana, Health checks and also connectors to use external services for such purposes:

• HTTP Server: JSON and Prometheus Exporter-style metrics

• Grafana Dashboards and Alerts

• Health Checks

• Calyptia Cloud: hosted service to monitor and visualize your pipelines

HTTP Server

Fluent Bit comes with a built-in HTTP Server that can be used to query internal information and monitor metrics of each running plugin.

The monitoring interface can be easily integrated with Prometheus since we support it native format.

Getting Started

To get started, the first step is to enable the HTTP Server from the configuration file:

[SERVICE]
    HTTP_Server  On
    HTTP_Listen  0.0.0.0
    HTTP_PORT    2020

[INPUT]
    Name cpu

[OUTPUT]
    Name  stdout
    Match *

the above configuration snippet will instruct Fluent Bit to start it HTTP Server on TCP Port 2020 and listening on all network interfaces:

$ bin/fluent-bit -c fluent-bit.conf
Fluent Bit v1.4.0
* Copyright (C) 2019-2020 The Fluent Bit Authors
* Copyright (C) 2015-2018 Treasure Data
* Fluent Bit is a CNCF sub-project under the umbrella of Fluentd
* https://fluentbit.io

[2020/03/10 19:08:24] [ info] [engine] started
[2020/03/10 19:08:24] [ info] [http_server] listen iface=0.0.0.0 tcp_port=2020

now with a simple curl command is enough to gather some information:

$ curl -s http://127.0.0.1:2020 | jq
{
  "fluent-bit": {
    "version": "0.13.0",
    "edition": "Community",
    "flags": [
      "FLB_HAVE_TLS",
      "FLB_HAVE_METRICS",
      "FLB_HAVE_SQLDB",
      "FLB_HAVE_TRACE",
      "FLB_HAVE_HTTP_SERVER",
      "FLB_HAVE_FLUSH_LIBCO",
      "FLB_HAVE_SYSTEMD",
      "FLB_HAVE_VALGRIND",
      "FLB_HAVE_FORK",
      "FLB_HAVE_PROXY_GO",
      "FLB_HAVE_REGEX",
      "FLB_HAVE_C_TLS",
      "FLB_HAVE_SETJMP",
      "FLB_HAVE_ACCEPT4",
      "FLB_HAVE_INOTIFY"
    ]
  }
}

Note that we are sending the curl command output to the jq program which helps to make the JSON data easy to read from the terminal. Fluent Bit doesn't aim to do JSON pretty-printing.

REST API Interface

Fluent Bit aims to expose useful interfaces for monitoring, as of Fluent Bit v0.14 the following end points are available:

URI	Description	Data Format
/	Fluent Bit build information	JSON
/api/v1/uptime	Get uptime information in seconds and human readable format	JSON
/api/v1/metrics	Internal metrics per loaded plugin	JSON
/api/v1/metrics/prometheus	Internal metrics per loaded plugin ready to be consumed by a Prometheus Server	Prometheus Text 0.0.4
/api/v1/storage	Get internal metrics of the storage layer / buffered data. This option is enabled only if in the SERVICE section the property storage.metrics has been enabled	JSON
/api/v1/health	Fluent Bit health check result	String
/api/v2/metrics	Internal metrics per loaded plugin	cmetrics text format
/api/v2/metrics/prometheus	Internal metrics per loaded plugin ready to be consumed by a Prometheus Server	Prometheus Text 0.0.4
/api/v2/reload	Execute hot reloading or get the status of hot reloading. For more details, please refer to the hot-reloading documentation.	JSON

Metric Descriptions

For v1 metrics

The following are detailed descriptions for the metrics outputted in prometheus format by /api/v1/metrics/prometheus.

The following definitions are key to understand:

• record: a single message collected from a source, such as a single long line in a file.

• chunk: Fluent Bit input plugin instances ingest log records and store them in chunks. A batch of records in a chunk are tracked together as a single unit; the Fluent Bit engine attempts to fit records into chunks of at most 2 MB, but the size can vary at runtime. Chunks are then sent to an output. An output plugin instance can either successfully send the full chunk to the destination and mark it as successful, or it can fail the chunk entirely if an unrecoverable error is encountered, or it can ask for the chunk to be retried.

Metric Name	Labels	Description	Type	Unit
fluentbit_input_bytes_total	name: the name or alias for the input instance	The number of bytes of log records that this input instance has successfully ingested	counter	bytes
fluentbit_input_records_total	name: the name or alias for the input instance	The number of log records this input has successfully ingested	counter	records
fluentbit_output_dropped_records_total	name: the name or alias for the output instance	The number of log records that have been dropped by the output. This means they met an unrecoverable error or retries expired for their chunk.	counter	records
fluentbit_output_errors_total	name: the name or alias for the output instance	The number of chunks that have faced an error (either unrecoverable or retriable). This is the number of times a chunk has failed, and does not correspond with the number of error messages you see in the Fluent Bit log output.	counter	chunks
fluentbit_output_proc_bytes_total	name: the name or alias for the output instance	The number of bytes of log records that this output instance has successfully sent. This is the total byte size of all unique chunks sent by this output. If a record is not sent due to some error, then it will not count towards this metric.	counter	bytes
fluentbit_output_proc_records_total	name: the name or alias for the output instance	The number of log records that this output instance has successfully sent. This is the total record count of all unique chunks sent by this output. If a record is not successfully sent, it does not count towards this metric.	counter	records
fluentbit_output_retried_records_total	name: the name or alias for the output instance	The number of log records that experienced a retry. Note that this is calculated at the chunk level, the count increased when an entire chunk is marked for retry. An output plugin may or may not perform multiple actions that generate many error messages when uploading a single chunk.	counter	records
fluentbit_output_retries_failed_total	name: the name or alias for the output instance	The number of times that retries expired for a chunk. Each plugin configures a Retry_Limit which applies to chunks. Once the Retry_Limit has been reached for a chunk it is discarded and this metric is incremented.	counter	chunks
fluentbit_output_retries_total	name: the name or alias for the output instance	The number of times this output instance requested a retry for a chunk.	counter	chunks
fluentbit_uptime		The number of seconds that Fluent Bit has been running.	counter	seconds
process_start_time_seconds		The Unix Epoch time stamp for when Fluent Bit started.	gauge	seconds

The following are detailed descriptions for the metrics outputted in JSON format by /api/v1/storage.

Metric Key	Description	Unit
chunks.total_chunks	The total number of chunks of records that Fluent Bit is currently buffering	chunks
chunks.mem_chunks	The total number of chunks that are buffered in memory at this time. Note that chunks can be both in memory and on the file system at the same time.	chunks
chunks.fs_chunks	The total number of chunks saved to the filesystem.	chunks
chunks.fs_chunks_up	A chunk is "up" if it is in memory. So this is the count of chunks that are both in filesystem and in memory.	chunks
chunks.fs_chunks_down	The count of chunks that are "down" and thus are only in the filesystem.	chunks
input_chunks.{plugin name}.status.overlimit	Is this input instance over its configured Mem_Buf_Limit?	boolean
input_chunks.{plugin name}.status.mem_size	The size of memory that this input is consuming to buffer logs in chunks.	bytes
input_chunks.{plugin name}.status.mem_limit	The buffer memory limit (Mem_Buf_Limit) that applies to this input plugin.	bytes
input_chunks.{plugin name}.chunks.total	The current total number of chunks owned by this input instance.	chunks
input_chunks.{plugin name}.chunks.up	The current number of chunks that are "up" in memory for this input. Chunks that are "up" will also be in the filesystem layer as well if filesystem storage is enabled.	chunks
input_chunks.{plugin name}.chunks.down	The current number of chunks that are "down" in the filesystem for this input.	chunks
input_chunks.{plugin name}.chunks.busy	"Busy" chunks are chunks that are being processed/sent by outputs and are not eligible to have new data appended.	chunks
input_chunks.{plugin name}.chunks.busy_size	The sum of the byte size of each chunk which is currently marked as busy.	bytes

For v2 metrics

The following are detailed descriptions for the metrics outputted in prometheus format by /api/v2/metrics/prometheus or /api/v2/metrics.

The following definitions are key to understand:

• record: a single message collected from a source, such as a single long line in a file.

• chunk: Fluent Bit input plugin instances ingest log records and store them in chunks. A batch of records in a chunk are tracked together as a single unit; the Fluent Bit engine attempts to fit records into chunks of at most 2 MB, but the size can vary at runtime. Chunks are then sent to an output. An output plugin instance can either successfully send the full chunk to the destination and mark it as successful, or it can fail the chunk entirely if an unrecoverable error is encountered, or it can ask for the chunk to be retried.

Metric Name	Labels	Description	Type	Unit
fluentbit_input_bytes_total	name: the name or alias for the input instance	The number of bytes of log records that this input instance has successfully ingested	counter	bytes
fluentbit_input_records_total	name: the name or alias for the input instance	The number of log records this input has successfully ingested	counter	records
fluentbit_filter_bytes_total	name: the name or alias for the filter instance	The number of bytes of log records that this filter instance has successfully ingested	counter	bytes
fluentbit_filter_records_total	name: the name or alias for the filter instance	The number of log records this filter has successfully ingested	counter	records
fluentbit_filter_added_records_total	name: the name or alias for the filter instance	The number of log records that have been added by the filter. This means they added into the data pipeline.	counter	records
fluentbit_filter_dropped_records_total	name: the name or alias for the filter instance	The number of log records that have been dropped by the filter. This means they removed from the data pipeline.	counter	records
fluentbit_output_dropped_records_total	name: the name or alias for the output instance	The number of log records that have been dropped by the output. This means they met an unrecoverable error or retries expired for their chunk.	counter	records
fluentbit_output_errors_total	name: the name or alias for the output instance	The number of chunks that have faced an error (either unrecoverable or retriable). This is the number of times a chunk has failed, and does not correspond with the number of error messages you see in the Fluent Bit log output.	counter	chunks
fluentbit_output_proc_bytes_total	name: the name or alias for the output instance	The number of bytes of log records that this output instance has successfully sent. This is the total byte size of all unique chunks sent by this output. If a record is not sent due to some error, then it will not count towards this metric.	counter	bytes
fluentbit_output_proc_records_total	name: the name or alias for the output instance	The number of log records that this output instance has successfully sent. This is the total record count of all unique chunks sent by this output. If a record is not successfully sent, it does not count towards this metric.	counter	records
fluentbit_output_retried_records_total	name: the name or alias for the output instance	The number of log records that experienced a retry. Note that this is calculated at the chunk level, the count increased when an entire chunk is marked for retry. An output plugin may or may not perform multiple actions that generate many error messages when uploading a single chunk.	counter	records
fluentbit_output_retries_failed_total	name: the name or alias for the output instance	The number of times that retries expired for a chunk. Each plugin configures a Retry_Limit which applies to chunks. Once the Retry_Limit has been reached for a chunk it is discarded and this metric is incremented.	counter	chunks
fluentbit_output_retries_total	name: the name or alias for the output instance	The number of times this output instance requested a retry for a chunk.	counter	chunks
fluentbit_uptime	hostname: the hostname on running fluent-bit	The number of seconds that Fluent Bit has been running.	counter	seconds
fluentbit_process_start_time_seconds	hostname: the hostname on running fluent-bit	The Unix Epoch time stamp for when Fluent Bit started.	gauge	seconds
fluentbit_build_info	hostname: the hostname, version: the version of fluent-bit, os: OS type	Build version information. The returned value is originated from initializing the Unix Epoch time stamp of config context.	gauge	seconds
fluentbit_hot_reloaded_times	hostname: the hostname on running fluent-bit	Collect the count of hot reloaded times.	gauge	seconds

The following are detailed descriptions for the metrics which is collected by storage layer.

Metric Name	Labels	Description	Type	Unit
fluentbit_input_chunks.storage_chunks	None	The total number of chunks of records that Fluent Bit is currently buffering	gauge	chunks
fluentbit_storage_mem_chunk	None	The total number of chunks that are buffered in memory at this time. Note that chunks can be both in memory and on the file system at the same time.	gauge	chunks
fluentbit_storage_fs_chunks	None	The total number of chunks saved to the filesystem.	gauge	chunks
fluentbit_storage_fs_chunks_up	None	A chunk is "up" if it is in memory. So this is the count of chunks that are both in filesystem and in memory.	gauge	chunks
fluentbit_storage_fs_chunks_down	None	The count of chunks that are "down" and thus are only in the filesystem.	gauge	chunks
fluentbit_storage_fs_chunks_busy	None	The total number of chunks are in a busy state.	gauge	chunks
fluentbit_storage_fs_chunks_busy_bytes	None	The total bytes of chunks are in a busy state.	gauge	bytes
fluentbit_input_storage_overlimit	name: the name or alias for the input instance	Is this input instance over its configured Mem_Buf_Limit?	gauge	boolean
fluentbit_input_storage_memory_bytes	name: the name or alias for the input instance	The size of memory that this input is consuming to buffer logs in chunks.	gauge	bytes
fluentbit_input_storage_chunks	name: the name or alias for the input instance	The current total number of chunks owned by this input instance.	gauge	chunks
fluentbit_input_storage_chunks_up	name: the name or alias for the input instance	The current number of chunks that are "up" in memory for this input. Chunks that are "up" will also be in the filesystem layer as well if filesystem storage is enabled.	gauge	chunks
fluentbit_input_storage_chunks_down	name: the name or alias for the input instance	The current number of chunks that are "down" in the filesystem for this input.	gauge	chunks
fluentbit_input_storage_chunks_busy	name: the name or alias for the input instance	"Busy" chunks are chunks that are being processed/sent by outputs and are not eligible to have new data appended.	gauge	chunks
fluentbit_input_storage_chunks_busy_bytes	name: the name or alias for the input instance	The sum of the byte size of each chunk which is currently marked as busy.	gauge	bytes
fluentbit_output_upstream_total_connections	name: the name or alias for the output instance	The sum of the connection count of each output plugins.	gauge	bytes
fluentbit_output_upstream_busy_connections	name: the name or alias for the output instance	The sum of the connection count in a busy state of each output plugins.	gauge	bytes

Uptime Example

Query the service uptime with the following command:

$ curl -s http://127.0.0.1:2020/api/v1/uptime | jq

it should print a similar output like this:

{
  "uptime_sec": 8950000,
  "uptime_hr": "Fluent Bit has been running:  103 days, 14 hours, 6 minutes and 40 seconds"
}

Metrics Examples

Query internal metrics in JSON format with the following command:

$ curl -s http://127.0.0.1:2020/api/v1/metrics | jq

it should print a similar output like this:

{
  "input": {
    "cpu.0": {
      "records": 8,
      "bytes": 2536
    }
  },
  "output": {
    "stdout.0": {
      "proc_records": 5,
      "proc_bytes": 1585,
      "errors": 0,
      "retries": 0,
      "retries_failed": 0
    }
  }
}

Metrics in Prometheus format

Query internal metrics in Prometheus Text 0.0.4 format:

$ curl -s http://127.0.0.1:2020/api/v1/metrics/prometheus

this time the same metrics will be in Prometheus format instead of JSON:

fluentbit_input_records_total{name="cpu.0"} 57 1509150350542
fluentbit_input_bytes_total{name="cpu.0"} 18069 1509150350542
fluentbit_output_proc_records_total{name="stdout.0"} 54 1509150350542
fluentbit_output_proc_bytes_total{name="stdout.0"} 17118 1509150350542
fluentbit_output_errors_total{name="stdout.0"} 0 1509150350542
fluentbit_output_retries_total{name="stdout.0"} 0 1509150350542
fluentbit_output_retries_failed_total{name="stdout.0"} 0 1509150350542

Configuring Aliases

By default configured plugins on runtime get an internal name in the format plugin_name.ID. For monitoring purposes, this can be confusing if many plugins of the same type were configured. To make a distinction each configured input or output section can get an alias that will be used as the parent name for the metric.

The following example set an alias to the INPUT section which is using the CPU input plugin:

[SERVICE]
    HTTP_Server  On
    HTTP_Listen  0.0.0.0
    HTTP_PORT    2020

[INPUT]
    Name  cpu
    Alias server1_cpu

[OUTPUT]
    Name  stdout
    Alias raw_output
    Match *

Now when querying the metrics we get the aliases in place instead of the plugin name:

{
  "input": {
    "server1_cpu": {
      "records": 8,
      "bytes": 2536
    }
  },
  "output": {
    "raw_output": {
      "proc_records": 5,
      "proc_bytes": 1585,
      "errors": 0,
      "retries": 0,
      "retries_failed": 0
    }
  }
}

Grafana Dashboard and Alerts

Fluent Bit's exposed prometheus style metrics can be leveraged to create dashboards and alerts.

The provided example dashboard is heavily inspired by Banzai Cloud's logging operator dashboard but with a few key differences such as the use of the instance label (see why here), stacked graphs and a focus on Fluent Bit metrics.

dashboard

Alerts

Sample alerts are available here.

Health Check for Fluent Bit

Fluent bit now supports four new configs to set up the health check.

Config Name	Description	Default Value
Health_Check	enable Health check feature	Off
HC_Errors_Count	the error count to meet the unhealthy requirement, this is a sum for all output plugins in a defined HC_Period, example for output error: [2022/02/16 10:44:10] [ warn] [engine] failed to flush chunk '1-1645008245.491540684.flb', retry in 7 seconds: task_id=0, input=forward.1 > output=cloudwatch_logs.3 (out_id=3)	5
HC_Retry_Failure_Count	the retry failure count to meet the unhealthy requirement, this is a sum for all output plugins in a defined HC_Period, example for retry failure: [2022/02/16 20:11:36] [ warn] [engine] chunk '1-1645042288.260516436.flb' cannot be retried: task_id=0, input=tcp.3 > output=cloudwatch_logs.1	5
HC_Period	The time period by second to count the error and retry failure data point	60

Note: Not every error log means an error nor be counted, the errors retry failures count only on specific errors which is the example in config table description

So the feature works as: Based on the HC_Period customer setup, if the real error number is over HC_Errors_Count or retry failure is over HC_Retry_Failure_Count, fluent bit will be considered as unhealthy. The health endpoint will return HTTP status 500 and String error. Otherwise it's healthy, will return HTTP status 200 and string ok

The equation is:

health status = (HC_Errors_Count > HC_Errors_Count config value) OR (HC_Retry_Failure_Count > HC_Retry_Failure_Count config value) IN the HC_Period interval

Note: the HC_Errors_Count and HC_Retry_Failure_Count only count for output plugins and count a sum for errors and retry failures from all output plugins which is running.

See the config example:

[SERVICE]
    HTTP_Server  On
    HTTP_Listen  0.0.0.0
    HTTP_PORT    2020
    Health_Check On 
    HC_Errors_Count 5 
    HC_Retry_Failure_Count 5 
    HC_Period 5 

[INPUT]
    Name  cpu

[OUTPUT]
    Name  stdout
    Match *

The command to call health endpoint

$ curl -s http://127.0.0.1:2020/api/v1/health

Based on the fluent bit status, the result will be:

• HTTP status 200 and "ok" in response to healthy status

• HTTP status 500 and "error" in response for unhealthy status

With the example config, the health status is determined by following equation:

Health status = (HC_Errors_Count > 5) OR (HC_Retry_Failure_Count > 5) IN 5 seconds

If (HC_Errors_Count > 5) OR (HC_Retry_Failure_Count > 5) IN 5 seconds is TRUE, then it's unhealthy.

If (HC_Errors_Count > 5) OR (HC_Retry_Failure_Count > 5) IN 5 seconds is FALSE, then it's healthy.

Calyptia

Calyptia is a hosted service that allows you to monitor your Fluent Bit agents including data flow, metrics and configurations.

Get Started with Calyptia Cloud

Register your Fluent Bit agent will take less than one minute, steps:

• Go to the calyptia core console and sign-in

• On the left menu click on settings and generate/copy your API key

In your Fluent Bit configuration file, append the following configuration section:

[CUSTOM]
    name     calyptia
    api_key  <YOUR_API_KEY>

Make sure to replace your API key in the configuration. After a few seconds upon restart your Fluent Bit agent, the Calyptia Cloud Dashboard will list your agent. Metrics will take around 30 seconds to shows up.

Contact Calyptia

If want to get in touch with Calyptia team, just send an email to [e-mail]