News
Fluent Bit v2.0.0 is the stable release!, new changes on this version:
Logs, Metrics, and Traces
Fluent Bit has always been agnostic of the data that its processes and moves around; one of the major use cases has been around Log Processing. Recently we introduced functionality to integrate Metrics and now with Fluent Bit v2, we are formally announcing the support for Traces.
As a vendor-agnostic solution, Fluent Bit natively integrates with protocols and other ecosystems, so having Metrics and Traces support means that we fully integrate with systems like Prometheus and OpenTelemetry.
Metrics
As part of the metrics collectors (inputs) and outputs supported for Metrics, allow us to collect and deliver metrics in a smooth way, a common example of these components are:
| Input | Description |
|---|---|
| node_exporter_metrics | A plugin that implements a subset of host metrics collection like the known external tool Prometheus Node Exporter. |
| nginx_metrics | Scrape Nginx metrics endpoints. It supports the OSS and Nginx+ enterprise editions. |
| windows_exporter_metrics | Collect system metrics from a Windows system, for it support CPU collection. |
| fluentbit_metrics | Collect internal Fluent Bit metrics and ingest them into the pipeline. |
| Output | Description |
|---|---|
| prometheus_exporter | Expose metrics in an HTTP endpoint in Prometheus text format. This mechanism is commonly used when you want to scrape metrics available by Fluent Bit by other a third part solution like Prometheus Server. |
| prometheus_remote_write | Deliver metrics to a remote endpoint by using the Prometheus Remote Write protocol. |
| splunk | New support for Splunk Metrics (HEC) |
| Influxdb | Send metrics to InfluxDB database |
One of the biggest addition on this are is the support for OpenTelemetry metrics, in the input and output side, more details below.
OpenTelemetry
OpenTelemetry is a framework, spec and protocol to unify the aspects of Telemetry collection and delivery. One of the major use cases in production is for Traces; Metrics and Logs support has been recently added.
In Fluent Bit, we are announcing full support for OpenTelemetry, where now we can receive and send telemetry data by using OpenTelemetry protocol:
OpenTelemetry input plugin
This new plugin, can receive OpenTelemetry Logs, Metrics and Traces, all of them are supported through this new implementation that allow easily integration with applications instrumented with OpenTelemetry SDKs.
OpenTelemetry output plugin
This new plugin allows to deliver any Log, Metric or Trace collected to a remote endpoint that supports OpenTelemetry, it could be any vendor platform or the OpenTelemetry collector.
Performance
The new threaded mechanism allows input plugins to run in a separate thread which helps to desaturate the main pipeline and move certain loads to a different CPU core. This feature must be enabled manually in the configuration of each input plugin by adding the threaded property, e.g:
[INPUT]
name tail
path /var/log/containers/*.log
tag kube.*
threaded on
Full OpenSSL support / deprecate mbedTLS
With this new version, all the network transport layer that needs security enabled will use OpenSSL instead of mbedTLS: networking & crypto functions. From now on, mbedTLS is officially deprecated.
All interfaces and plugins that were using mbedTLS has been ported to use the new crypto and networking layer we have implemented.
TLS for input plugins
Using Fluent Bit as an aggregator is a common use case, but one component was missing: TLS support for ingestion (Transport Layer Security), which for many users it was a big blocker; the same problem has been reported by many Fluentd users looking forward to migrating to Fluent Bit.
In this new version, we are announcing full-native TLS capabilities for input plugins.
Plugins Changes
Loki (output)
In previous versions, the Loki project did not support out-of-order records, for performance reasons, Fluent Bit delivers messages in parallel which could cause exceptions on the transmission since some messages might not arrive in the order expected by Loki; to solve this problem, we used to force our Loki connector to just send one Chunk at a time, so the order was preserved.
Since newer versions of Loki no longer have this restriction, we are removing the restriction too from our side and allowing concurrency again, which is a boost in performance. No changes are needed.
Note: this improvement might be a breaking change if you are using an older version of Loki. If you are using **Loki >= 2.4 ** you are good to go.
To learn more about this Loki restriction and the enhancements, you can refer to the following blog post from Grafana team:
https://grafana.com/blog/2021/12/03/new-feature-in-loki-2.4-no-more-ordering-constraint/
Splunk + Metrics (output)
The current Splunk output connector now supports the delivery of Metrics. No changes in your configuration are needed, if you send metrics records to a Splunk output plugin, the data will be converted and delivered as expected by the Splunk HEC endpoint.
Developers Experience
There are many cases where a user would like to extend the functionality that Fluent Bit provides, usually for specific business reasons.
For customizations, we currently support plugins that can be written in C, filters in Lua, or output plugins in Golang. But understanding our user’s needs and the restrictions of this limited functionality, we are going beyond that.
Fluent Bit v2 comes with support to integrate more plugin types with Golang and WebAssembly. The following table describes the supported languages by plugin type.
| Language | Input / Source | Filter | Output / Destination |
|---|---|---|---|
| C | Yes | Yes | Yes |
| Golang | Yes | —- | Yes |
| WebAssembly | Yes | Yes | —- |
TAP
One of the common questions from our users has been “how can I see the data flowing through a pipeline?”, usually, this comes up when there is a need to perform troubleshooting without stopping the agent.
TAP is a very advanced functionality that allows inspecting the data that flows through a pipeline, it can be enabled on runtime by just using a simple HTTP call.
More details about this functionality can be seen here in the official documentation:
Internal Metrics
Historically, a common way to extract internal metrics from Fluent Bit has been by enabling the built-in HTTP service which exposes the following endpoints in the REST API:
| Endpoint | Description |
|---|---|
| /api/v1/metrics | JSON metrics for records and byte sizes associated. |
| /api/v1/metrics/prometheus | Same metrics as /api/v1/metrics but formatted in Prometheus Text format.
|
| /api/v1/storage | JSON metrics for the storage component which expose metrics for Chunks. |
One of the restrictions of this interface, is that those metrics are not part of the pipeline, which means, you can only access them remotely. But what about if you want to send your metrics to a destination like Prometheus, InfluxDB or other ?.
During the release cycle of the previous series of Fluent Bit, we introduced a new mechanism to collect and process internal metrics through a new input plugin plugin called fluentbit_metrics. This new mechanism allows to expose or send the metrics to a destination in many ways like:
- Prometheus Exporter
- Prometheus Remote Write
- InfluxDB
- OpenTelemetry
- Standard output interface (stdout)
New Storage Metrics
As mentioned, the Storage metrics originally were exposed in a different endpoint which forced the user to scrape another endpoint and also faced the restriction that storage metrics were only exposed in JSON.
Starting from Fluent Bit v2, fluentbit_metrics exposes seven new metrics for the Storage layer:
| Metrics name | Description |
|---|---|
fluentbit_input_storage_overlimit
| It takes a value of 1 if the input plugin instance is over the limit imposed by a mem_buf_limit configuration. Otherwise is just set to 0. |
fluentbit_input_storage_memory_bytes
| Ttotal number of bytes used by the chunks which are up in memory. |
fluentbit_input_storage_chunks
| Total number of chunks currently associated to the input plugin instance. This includes chunks in an 'up' and 'down' state. |
fluentbit_input_storage_chunks_up
| Total number of chunks that are 'up' in memory generated by the input plugin instance. |
fluentbit_input_storage_chunks_down
| Total number of chunks that are in a 'down' state. When the input plugin has enabled the filesystem storage type, not all chunks are loaded 'up' in memory, some of them will be on disk only and not loaded in memory until a later time. This metric represents the number of chunks on that state. |
fluentbit_input_storage_chunks_busy
| Total number of chunks that are in a busy state. If a chunk is being used by an output plugin, likely under a flush operation, the chunk is marked as busy. |
fluentbit_input_storage_chunks_busy_bytes
| Total number of bytes being used by chunks in a busy state. |
Memory Ring buffer
The memory-ring-buffer mechanism is an optional buffering strategy for input plugins that relies only in memory. When this is enabled, the chunks management behaves as a ring buffer of a fixed size were setting up mem_buf_limit is mandatory to specify that limit.
When the chunks are enqueued, and mem_buf_limit is reached, instead of pausing the input plugin as normally would happen with the default memory strategy, this new mechanism removes the oldest Chunks from the queue until to make sure there is enough space for the new incoming data.
The configuration of this new strategy is as follows:
[INPUT]
name tail
path /var/log/*
# new configuration options for memory ring buffer
storage.type memrb
mem_buf_limit 20M
NOTE: by default Fluent Bit tries to enqueue Chunks of size around 2M, this is an internal size that cannot be changed, but there are scenarios where the Chunk size can be greater than that. So mem_buf_limit should be no less than 20M.
Metrics exposed
In addition to the ring-buffer functionality, this PR implements an extension of metrics to count the number of dropped bytes and dropped chunks, e.g:
# HELP fluentbit_input_memrb_dropped_chunks Number of memrb dropped chunks.
# TYPE fluentbit_input_memrb_dropped_chunks counter
fluentbit_input_memrb_dropped_chunks{name="dummy.0"} 22
# HELP fluentbit_input_memrb_dropped_bytes Number of memrb dropped bytes.
# TYPE fluentbit_input_memrb_dropped_bytes counter
fluentbit_input_memrb_dropped_bytes{name="dummy.0"} 572
The metrics above are an example of how they are retrieved when exposed through the Prometheus Exporter output plugin. The following configuration was used:
[INPUT]
name dummy
# new configuration options for memory ring buffer
storage.type memrb
mem_buf_limit 20M
[INPUT]
name fluentbit_metrics
tag metrics
[OUTPUT]
name prometheus_exporter
match metrics
[OUTPUT]
name http
match dummy*
host 127.0.0.1
port 9999
Deprecating td-agent-bit
In the early days, Fluent Bit packages were called td-agent-bit, since it was the distribution of Fluent Bit by Treasure Data, the original sponsor of this project, but that’s sponsorship is no longer active.
TD Agent Bit was widely adopted by the community and during the v1.9 release cycle, we informed the community that this package name will be deprecated, now with v2, we are finalizing that transition.
From now on, all Fluent Bit packages and container images follow the project name fluent-bit as it should be.
We encourage users to move to Fluent Bit v2 and adopt the new package name. Note that repositories keep being the same.