Skip to content
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
Binary file added imgs/traces_head_sampling.png
Loading
Sorry, something went wrong. Reload?
Sorry, we cannot display this file.
Sorry, this file is invalid so it cannot be displayed.
Binary file added imgs/traces_tail_sampling.png
Loading
Sorry, something went wrong. Reload?
Sorry, we cannot display this file.
Sorry, this file is invalid so it cannot be displayed.
2 changes: 2 additions & 0 deletions pipeline/processors/README.md
Original file line number Diff line number Diff line change
Expand Up @@ -19,6 +19,8 @@ Fluent Bit offers the following processors:
- [OpenTelemetry Envelope](opentelemetry-envelope.md): Transform logs into an
OpenTelemetry-compatible format.
- [SQL](sql.md): Use SQL queries to extract log content.
- [Traces](traces.md): Trace sampling designed with a pluggable architecture,
allowing easy extension to support multiple sampling strategies and backends.
- [Filters](filters.md): Any filter can be used as a processor.

## Features
Expand Down
377 changes: 377 additions & 0 deletions pipeline/processors/traces.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,377 @@
# Traces

The _Traces_ sampling processor is designed with a pluggable architecture, allowing easy extension to support multiple sampling strategies and backends. It provides you with the ability to apply head or tail sampling to incoming trace telemetry data.

Available samplers:

- `probabilistic` (head sampling)
- `tail` (tail sampling)

Conditions:

- `latency`
- `span_count`
- `status_code`
- `string_attribute`
- `numeric_attribute`
- `boolean_attribute`
- `trace_state`

## Configuration Parameters

The processor does not provide any extra configuration parameter, it can be used directly in your _processors_ Yaml directive.

## Traces types

Traces have both a name and a type with the following possible settings:

| Key | Possible values |
| :----- | :---------------------: |
| `name` | `sampling` |
| `type` | `probabilistic`, `tail` |

## Head sampling

In this example, head sampling will be used to process a smaller percentage of the overall ingested traces and spans. This is done by setting up the pipeline to ingest on the OpenTelemetry defined port as shown below using the OpenTelemetry Protocol (OTLP). The processor section defines traces for head sampling and the sampling percentage defining the total ingested traces and spans to be forwarded to the defined output plugins.

![](/imgs/traces_head_sampling.png)

| Sampling settings | Description |
| :-------------------- | :------------------------------------------------------------------------------------------------------------------ |
| `sampling_percentage` | This sets the probability of sampling trace, can be between 0-100%. For example, 40 samples 40% of traces randomly. |

**fluent-bit.yaml**

```yaml
service:
flush: 1
log_level: info
hot_reload: on

pipeline:
inputs:
- name: opentelemetry
port: 4318

processors:
traces:
# Head sampling of traces (percentage)
- name: sampling
type: probabilistic
sampling_settings:
sampling_percentage: 40

outputs:
- name: stdout
match: "*"
```

With this head sampling configuration, a sample set of ingested traces will randomly send 40% of the total traces to the standard output.

## Tail sampling

Tail sampling is used to obtain a more selective and fine grained control over the collection of traces and spans without collecting everything. Below is an example showing the process is a combination of waiting on making a sampling decision together followed by configuration defined conditions to determine the spans to be sampled.

![](/imgs/traces_tail_sampling.png)

The following samplings settings are available with their default values:

| Sampling settings | Description | Default value |
| :---------------- | :------------------------------------------------------------------------------------------------------------------------- | :-----------: |
| `decision_wait` | Specifies how long to buffer spans before making a sampling decision, allowing full trace evaluation. | 30s |
| `max_traces` | Specifies the maximum number of traces that can be held in memory. When the limit is reached, the oldest trace is deleted. | |

The tail-based sampler supports various conditionals to sample traces if their spans meet a specific condition.

### Condition: latency

This condition samples traces based on span duration. It uses `threshold_ms_low` to capture short traces and `threshold_ms_high` for long traces.

| Condition settings | Description | Default value |
| :------------------ | :------------------------------------------------------------------------------------------- | :-----------: |
| `threshold_ms_low` | Specifies the lower latency threshold. Traces with a duration <= this value will be sampled. | 0 |
| `threshold_ms_high` | Specifies the upper latency threshold. Traces with a duration >= this value will be sampled. | 0 |

**fluent-bit.yaml**

```yaml
service:
flush: 1
log_level: info
hot_reload: on

pipeline:
inputs:
- name: opentelemetry
port: 4318

processors:
traces:
# Tail sampling of traces (latency)
- name: sampling
type: tail
sampling_settings:
decision_wait: 5s
conditions:
- type: latency
threshold_ms_high: 200
threshold_ms_high: 3000

outputs:
- name: stdout
match: "*"
```

This tail-based sampling configuration waits 5 seconds before making a decision. It samples traces based on latency, capturing short traces of 200ms or less and long traces of 3000ms or more. Traces between 200ms and 3000ms are not sampled unless another condition applies.

### Condition: span_count

This condition samples traces that have specific span counts defined in a configurable range. It uses `min_spans` and `max_spans` to specify the number of spans a trace can have to be sampled.

| Condition settings | Description | Default value |
| :----------------- | :--------------------------------------------------------------------- | :-----------: |
| `max_spans` | Specifies the minimum number of spans a trace must have to be sampled. | |
| `min_spans` | Specifies the maximum number of spans a trace can have to be sampled. | |

**fluent-bit.yaml**

```yaml
service:
flush: 1
log_level: info
hot_reload: on

pipeline:
inputs:
- name: opentelemetry
port: 4318

processors:
traces:
# Tail sampling of traces (span_count)
- name: sampling
type: tail
sampling_settings:
decision_wait: 5s
conditions:
- type: span_count
min_spans: 3
max_spans: 5

outputs:
- name: stdout
match: "*"
```

This tail-based sampling configuration waits 5 seconds before making a decision. It samples traces based on having a minimum of 3 spans and a maximum of 5 spans. Traces with less than 3 and more than 5 spans are not sampled unless another condition applies.

### Condition: status_code

This condition samples traces based on span status codes (`OK`, `ERROR`, `UNSET`).

| Condition settings | Description | Default value |
| :----------------- | :--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | :-----------: |
| `status_codes` | Defines an array of span status codes (`OK`, `ERROR`, `UNSET`) to filter traces. Traces are sampled if any span matches a listed status code. For example, `status_codes: [ERROR, UNSET]` captures traces with errors or unset statuses. | |

**fluent-bit.yaml**

```yaml
service:
flush: 1
log_level: info
hot_reload: on

pipeline:
inputs:
- name: opentelemetry
port: 4318

processors:
traces:
# Tail sampling of traces (status_code)
- name: sampling
type: tail
sampling_settings:
decision_wait: 5s
conditions:
- type: status_code
status_codes: [ERROR]

outputs:
- name: stdout
match: "*"
```

With this tail-based sampling configuration, a sample set of ingested traces will select only the spans with status codes marked as `ERROR` to the standard output.

### Condition: string_attribute

This conditional allows traces to be sampled based on specific span or resource attributes. Users can define key-value filters (e.g., http.method=POST) to selectively capture relevant traces.

| Condition settings | Description | Default value |
| :----------------- | :---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | :-----------: |
| `key` | Specifies the span or resource attribute to match (e.g., "service.name"). | |
| `values` | Defines an array of accepted values for the attribute. A trace is sampled if any span contains a matching key-value pair: `["payment-processing"]` | |
| `match_type` | Defines how attributes are compared: `strict` ensures exact value matching, while `exists` checks if the attribute is present regardless of its value (note that string type is enforced) | `strict` |

**fluent-bit.yaml**

```yaml
service:
flush: 1
log_level: info
hot_reload: on

pipeline:
inputs:
- name: opentelemetry
port: 4318

processors:
traces:
# Tail sampling of traces (string_attribute)
- name: sampling
type: tail
sampling_settings:
decision_wait: 2s
conditions:
- type: string_attribute
match_type: strict
key: "http.method"
values: ["GET"]
- type: string_attribute
match_type: exists
key: "service.name"

outputs:
- name: stdout
match: "*"
```

This tail-based sampling configuration waits 2 seconds before making a decision. It samples traces based on string matching key value pairs. Traces are sampled if the key `http.method` is set to `GET` or if spans or resources have a key `service.name`.

### Condition: numeric_attribute

This condition samples traces based on numeric attribute values of a defined key where users can configure minimum and maximum thresholds.

| Condition settings | Description | Default value |
| :----------------- | :---------------------------------------------------------------------------------------------------------------------------------------------------- | :-----------: |
| `key` | Specifies the span or resource attribute to match (e.g., "service.name"). | |
| `min_value` | The minimum inclusive value for the numeric attribute. Traces with values >= the `min_value` are sampled. | |
| `max_value` | The maximum inclusive value for the numeric attribute. Traces with values <= the `max_value` are sampled. | |
| `match_type` | This defines how attribute values are evaluated: `strict` matches exact values, `exists` checks if the attribute is present, regardless of its value. | `strict` |

**fluent-bit.yaml**

```yaml
service:
flush: 1
log_level: info
hot_reload: on

pipeline:
inputs:
- name: opentelemetry
port: 4318

processors:
traces:
# Tail sampling of traces (status_code)
- name: sampling
type: tail
sampling_settings:
decision_wait: 5s
conditions:
- type: numeric_attribute
key: "http.status_code"
min_value: 400
max_value: 504

outputs:
- name: stdout
match: "*"
```

With this tail-based sampling configuration, a sample set of ingested traces will select only the spans with a key `http.status code` with numeric values between 400 and 504 inclusive.

### Condition: boolean_attribute

This condition samples traces based on a boolean attribute value of a defined key. This allows for selection of traces based on flags such as error indicators or debug modes.

| Condition settings | Description | Default value |
| :----------------- | :------------------------------------------------------------------------ | :-----------: |
| `key` | Specifies the span or resource attribute to match (e.g., "service.name"). | |
| `value` | Expected boolean value: `true` or `false` | |

**fluent-bit.yaml**

```yaml
service:
flush: 1
log_level: info
hot_reload: on

pipeline:
inputs:
- name: opentelemetry
port: 4318

processors:
traces:
# Tail sampling of traces (boolean_attribute)
- name: sampling
type: tail
sampling_settings:
decision_wait: 2s
conditions:
- type: boolean_attribute
key: "user.logged"
value: false

outputs:
- name: stdout
match: "*"
```

This tail-based sampling configuration waits 2 seconds before making a decision. It samples traces that do not have the key `user.logged` set to true. Traces are sampled if the key `user.logged` is set to `true`.

### Condition: trace_state

This condition samples traces based on metadata stored int he W3C `trace_state` field.

| Condition settings | Description | Default value |
| :----------------- | :------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ | :-----------: |
| `values` | Defines a list of key, value pairs to match against the `trace_state`. A trace is sampled if any of the specified values exist in the `trace_state` field. Matching follows OR logic, meaning at least one value must be present for sampling to occur. | |

**fluent-bit.yaml**

```yaml
service:
flush: 1
log_level: info
hot_reload: on

pipeline:
inputs:
- name: opentelemetry
port: 4318

processors:
traces:
# Tail sampling of traces (trace_state)
- name: sampling
type: tail
sampling_settings:
decision_wait: 2s
conditions:
- type: trace_state
values: [debug=false, priority=high]

outputs:
- name: stdout
match: "*"
```

This tail-based sampling configuration waits 2 seconds before making a decision. It samples traces that do not have the key `user.logged` set to true. Traces are sampled if the key `user.logged` is set to `true`.

For more details about further processing, read the [Content Modifier](../processors/content-modifier.md) processor documentation.