Skip to content

Latest commit

 

History

History
408 lines (300 loc) · 13.6 KB

README.md

File metadata and controls

408 lines (300 loc) · 13.6 KB
Raw

fluent-plugin-protobuf-http

ci License: Apache GitHub release (latest by date) RubyGems Downloads

Overview

Fluentd HTTP input plugin for Protocol Buffers.

Features

  • Automatic compilation of .proto files located in proto_dir
  • Incoming Format: Binary or JSON (Content-Type: application/octet-stream or application/json)
  • Outgoing Format: Binary or JSON
  • Single and Batch message support
  • TLS Support with <transport> section and https:// URL protocol prefix.

For more details on TLS configuration, see this official example.

Installation

RubyGems

gem install fluent-plugin-protobuf-http

Bundler

Add the following line to your Gemfile:

gem 'fluent-plugin-protobuf-http'

And then execute:

bundle

Configuration

  • bind (string) (optional): The address to listen to.
    • Default: 0.0.0.0
  • port (integer) (optional): The port to listen to.
    • Default: 8080
  • proto_dir (string) (required): The directory path that contains the .proto files.
  • in_mode (enum) (optional): The mode of incoming (supported) events.
    • Modes: binary, json
    • Default: binary
  • out_mode (enum) (optional): The mode of outgoing (emitted) events.
    • Modes: binary, json
    • Default: binary
  • tag (string) (required): The tag for the event.

<transport> section (optional) (single)

  • protocol (enum) (optional):
    • Protocols: tcp, tls
    • Default: tcp
    • For more details, see this official configuration example.

Example

# Endpoints:
# - Single Message: http://ip:port/<tag>?msgtype=<msgtype>
# - Batch  Message: http://ip:port/<tag>?msgtype=<batch-msgtype>?batch=true

<source>
  @type       protobuf_http
  @id         protobuf_http_input

  bind        0.0.0.0
  port        8080
  tag         debug.test

  proto_dir   ~/fluent/protos
  in_mode     binary
  out_mode    json
</source>

Schemas (.proto files)

The prime use-case for this plugin is assumed to be event logging. So, always use self-contained .proto file(s) that do not import other .proto files. The names e.g. package, message, etc. must be unique and are treated as case-sensitive.

Consider this log.proto schema from protobuf-log-sample repository:

syntax = "proto3";

package service.logging;

import "google/protobuf/timestamp.proto";

message Log {
  message Context {
    google.protobuf.Timestamp timestamp = 1;
    string host_or_ip = 2;
    string service_name = 3;
    string user = 4;
  }

  enum Level {
    DEBUG = 0;
    INFO = 1;
    WARN = 2;
    ERROR = 3;
    FATAL = 4;
  }

  Context context = 1;
  Level level = 2;
  string message = 3;
}

The fully-qualified message type for Log is service.logging.Log. This message type is used as the value of msgtype query parameter in the URL. See URL section below for more on msgtype.

Single Message

The above schema will be used as-is for the single message.

Batch Message

For the batch message, the schema must be like this:

message Batch {
  string type = 1;
  repeated Log batch = 2;
}

IMPORTANT: The Batch message type is part of log.proto, it is not a separate file! You can choose any name for a batch message type.

Here is the complete log.proto file:

syntax = "proto3";

package service.logging;

import "google/protobuf/timestamp.proto";

message Log {
  message Context {
    google.protobuf.Timestamp timestamp = 1;
    string host_or_ip = 2;
    string service_name = 3;
    string user = 4;
  }

  enum Level {
    DEBUG = 0;
    INFO = 1;
    WARN = 2;
    ERROR = 3;
    FATAL = 4;
  }

  Context context = 1;
  Level level = 2;
  string message = 3;
}

message Batch {
  string type = 1;
  repeated Log batch = 2;
}

For batch processing, the plugin looks for special members type and batch. The type will indicate the message type of batch i.e. Log in this example.

The type of Batch is service.logging.Batch and it will be the value of msgtype in the URL query. The type of batch array is service.logging.Log and it will be the value of type.

The google.protobuf.Any type has not been used here deliberately. It stores the message type with each message resulting in an increase in size. Refer to protobuf-repeated-type-vs-any for a simple comparison. With the above approach, the type is stored only once for the whole batch.

Endpoint (URL)

For single message:

http://<ip>:<port>/<tag>?msgtype=<fully-qualified-message-type>

For batch message:

http://<ip>:<port>/<tag>?msgtype=<fully-qualified-message-type-for-batch>&batch=true

Without batch=true query parameter, the batch will be treated as a single message.

For example, for a log type service.logging.Log and its corresponding batch type service.logging.Batch, the URLs would be:

For single message:

http://localhost:8080/debug.test?msgtype=service.logging.Log

For batch message:

http://localhost:8080/debug.test?msgtype=service.logging.Batch&batch=true

NOTE: The values of query parameters (msgtype, batch) are case-sensitive!

Test Use-Case (curl)

For a simple use-case of incoming HTTP events and their routing to stdout may be configured like this:

fluent.conf:

<source>
  @type       protobuf_http
  @id         protobuf_http_input

  bind        0.0.0.0
  port        8080
  tag         debug.test

  proto_dir   ~/fluent/protos
  in_mode     binary
  out_mode    json
</source>

<match debug.test>
  @type       stdout
</match>

The incoming binary messages will be converted to JSON for further consumption.

Single Message Use-case

Test Parameters:

input file single message type
log.bin service.logging.Log

URL:

http://localhost:8080/debug.test?msgtype=service.logging.Log

curl command:

curl -X POST -H "Content-Type: application/octet-stream" \
    --data-binary "@/<path>/log.bin" \
    "http://localhost:8080/debug.test?msgtype=service.logging.Log"

fluentd logs (Observe JSON at the end):

2020-06-09 18:53:47 +0500 [info]: #0 [protobuf_http_input] [R] {binary} [127.0.0.1:41222, size: 86 bytes]
2020-06-09 18:53:47 +0500 [warn]: #0 [protobuf_http_input] 'batch' not found in 'query_string' [msgtype=service.logging.Log]
2020-06-09 18:53:47 +0500 [info]: #0 [protobuf_http_input] [S] {binary} [127.0.0.1:41222, msgtype: service.logging.Log, size: 86 bytes]
2020-06-09 18:53:47.025638542 +0500 debug.test: {"message":"{\"context\":{\"timestamp\":\"2020-06-01T16:24:19Z\",\"hostOrIp\":\"192.168.xxx.xxx\",\"serviceName\":\"test\",\"user\":\"test\"},\"level\":\"INFO\",\"message\":\"This is a test log generated by [./log.rb].\"}"}
2020-06-09 18:53:47 +0500 [info]: #0 [protobuf_http_input] [S] {json} [127.0.0.1:41222, msgtype: service.logging.Log, size: 183 bytes]

For generating sample Single messages, see https://github.com/iamAzeem/protobuf-log-sample.

Batch Message Use-case

Test Parameters:

input file batch message type batch internal type messages
logbatch2.bin service.logging.Batch service.logging.Log 2
logbatch5.bin service.logging.Batch service.logging.Log 5

URL:

http://localhost:8080/debug.test?msgtype=service.logging.Batch&batch=true

logbatch2.bin

curl command:

$ curl -X POST -H "Content-Type: application/octet-stream" \
    --data-binary "@/<path>/logbatch2.bin" \
    "http://localhost:8080/debug.test?msgtype=service.logging.Batch&batch=true"
{"status":"Batch received! [batch_type: service.logging.Log, batch_size: 2 messages]"}

fluentd logs:

2020-06-09 19:04:13 +0500 [info]: #0 [protobuf_http_input] [R] {binary} [127.0.0.1:41416, size: 207 bytes]
2020-06-09 19:04:13 +0500 [info]: #0 [protobuf_http_input] [B] {binary} [127.0.0.1:41416, msgtype: service.logging.Batch, size: 207 bytes]
2020-06-09 19:04:13 +0500 [info]: #0 [protobuf_http_input] [B] Emitting message stream/batch [batch_size: 2 messages]...
2020-06-09 19:04:13.546158307 +0500 debug.test: {"message":"{\"context\":{\"timestamp\":\"2020-06-08T17:27:05Z\",\"hostOrIp\":\"192.168.xxx.xxx\",\"serviceName\":\"test\",\"user\":\"test\"},\"level\":\"INFO\",\"message\":\"This is a test log generated by [./logbatch.rb].\"}"}
2020-06-09 19:04:13.546192659 +0500 debug.test: {"message":"{\"context\":{\"timestamp\":\"2020-06-08T17:27:05Z\",\"hostOrIp\":\"192.168.xxx.xxx\",\"serviceName\":\"test\",\"user\":\"test\"},\"level\":\"INFO\",\"message\":\"This is a test log generated by [./logbatch.rb].\"}"}
2020-06-09 19:04:13 +0500 [info]: #0 [protobuf_http_input] [B] {json} [127.0.0.1:41416, msgtype: service.logging.Batch] Batch received! [batch_type: service.logging.Log, batch_size: 2 messages]

logbatch5.bin

curl command:

$ curl -X POST -H "Content-Type: application/octet-stream" \
    --data-binary "@/<path>/logbatch5.bin" \
    "http://localhost:8080/debug.test?msgtype=service.logging.Batch&batch=true"
{"status":"Batch received! [batch_type: service.logging.Log, batch_size: 5 messages]"}

fluentd logs:

2020-06-09 19:07:09 +0500 [info]: #0 [protobuf_http_input] [R] {binary} [127.0.0.1:41552, size: 486 bytes]
2020-06-09 19:07:09 +0500 [info]: #0 [protobuf_http_input] [B] {binary} [127.0.0.1:41552, msgtype: service.logging.Batch, size: 486 bytes]
2020-06-09 19:07:09 +0500 [info]: #0 [protobuf_http_input] [B] Emitting message stream/batch [batch_size: 5 messages]...
2020-06-09 19:07:09.738057617 +0500 debug.test: {"message":"{\"context\":{\"timestamp\":\"2020-06-08T17:27:05Z\",\"hostOrIp\":\"192.168.xxx.xxx\",\"serviceName\":\"test\",\"user\":\"test\"},\"level\":\"INFO\",\"message\":\"This is a test log generated by [./logbatch.rb].\"}"}
2020-06-09 19:07:09.738131040 +0500 debug.test: {"message":"{\"context\":{\"timestamp\":\"2020-06-08T17:27:05Z\",\"hostOrIp\":\"192.168.xxx.xxx\",\"serviceName\":\"test\",\"user\":\"test\"},\"level\":\"INFO\",\"message\":\"This is a test log generated by [./logbatch.rb].\"}"}
2020-06-09 19:07:09.738144897 +0500 debug.test: {"message":"{\"context\":{\"timestamp\":\"2020-06-08T17:27:05Z\",\"hostOrIp\":\"192.168.xxx.xxx\",\"serviceName\":\"test\",\"user\":\"test\"},\"level\":\"INFO\",\"message\":\"This is a test log generated by [./logbatch.rb].\"}"}
2020-06-09 19:07:09.738155033 +0500 debug.test: {"message":"{\"context\":{\"timestamp\":\"2020-06-08T17:27:05Z\",\"hostOrIp\":\"192.168.xxx.xxx\",\"serviceName\":\"test\",\"user\":\"test\"},\"level\":\"INFO\",\"message\":\"This is a test log generated by [./logbatch.rb].\"}"}
2020-06-09 19:07:09.738165527 +0500 debug.test: {"message":"{\"context\":{\"timestamp\":\"2020-06-08T17:27:05Z\",\"hostOrIp\":\"192.168.xxx.xxx\",\"serviceName\":\"test\",\"user\":\"test\"},\"level\":\"INFO\",\"message\":\"This is a test log generated by [./logbatch.rb].\"}"}
2020-06-09 19:07:09 +0500 [info]: #0 [protobuf_http_input] [B] {json} [127.0.0.1:41552, msgtype: service.logging.Batch] Batch received! [batch_type: service.logging.Log, batch_size: 5 messages]

For sample Batch message generation, see this gist.

CI Workflow

The CI workflow sets up the prerequisites. It builds and installs the plugin, and then runs the automated tests.

To run tests locally, run:

bundle exec rake test

The test directory contains the tests and the input files.

The code coverage is printed at the end using simplecov.

Known Issues

  • This plugin internally uses the HTTP server plugin helper which has higher precedence for async-http over webrick. But, webrick is required to run this. In an environment where both are installed, async-http is automatically selected causing runtime issues. To make this work, you need to uninstall async-http i.e. gem uninstall async-http. See issue #10 for more details where this was identified in Docker containers where both gems were already installed.

Contribute

  • Fork the project.
  • Check out the latest main branch.
  • Create a feature or bugfix branch from main.
  • Commit and push your changes.
  • Make sure to add and run tests locally: bundle exec rake test.
  • Run Rubocop and fix the lint errors.
  • Submit the PR.

License

Apache 2.0