output: make it go zoom

[yaauie]

By declaring concurrency :shared, a single shared instance can be used by pipeline workers concurrently instead of the legacy sharing we get by default that shares the instance sequentially.

(labeled as a bug, since this plugin was designed with concurrent use in mind, we simply neglected to declare it)

What does this PR do?

Removes a bottleneck caused by Logstash selecting the "legacy" strategy, which prevented the single shared instance of this plugin from being used concurrently.

Why is it important/What is the impact to the user?

Make it go ZOOM! (improves throughput by allowing use of available resources)

Checklist

• My code follows the style guidelines of this project
• I have commented my code, particularly in hard-to-understand areas
• [ ] I have made corresponding changes to the documentation
• [ ] I have made corresponding change to the default configuration files (and/or docker env variables)
• I have added tests that prove my fix is effective or that my feature works

Author's Checklist

• [ ]

How to test this PR locally

• A pair of pipelines, upstream and downstream:
  • A downstream pipeline with a logstash input that receives events and discards them
  • An upstream pipeline with a java_generator input that floods the queue as quickly as possible and a logstash output that transmits the events to the downstream
  • Note: for simplicity's sake, we add ssl_enabled => false to both to obviate the need to set up certificates.

# pipelines.yml
- pipeline.id: upstream
  pipeline.workers: 16
  config.string: |-
    input { java_generator {} }
    output { logstash { hosts => ["localhost:9803"] ssl_enabled => false } }
- pipeline.id: downstream
  config.string: |-
    input { logstash { port => 9803 ssl_enabled => false } }
    output { sink {} }'

• Observe that without this patch, changing the number of workers on the upstream has little/no net effect on throughput (because the single shared logstash output is sequentially accessed by the workers)
• Observe that with this patch, throughput is higher when there is more than one worker in the pipeline

[jsvd]

LGTM