middleware poc with compression middleware

[margaretjgu]

Next steps:

• Publish unstable release of @elastic/transport with middleware exports
• Test in elasticsearch-js to validate client integration

Example client usage:

import { Client } from '@elastic/elasticsearch'
import { MiddlewareEngine, CompressionMiddleware } from '@elastic/transport/middleware'

// Custom middleware for Kibana
class KibanaMiddleware implements Middleware {
  readonly name = 'kibana'
  readonly priority = 15

  onBeforeRequest = async (ctx) => {
    return {
      context: {
        request: {
          headers: {
            ...ctx.request.headers,
            'x-elastic-product-origin': 'kibana'
          }
        }
      }
    }
  }
}

const client = new Client({
  node: 'https://localhost:9200',
  middleware: [
    new CompressionMiddleware({ enabled: true }),
    new KibanaMiddleware()
  ]
})

[JoshMock]

oof, reminder that I want to refactor both libraries to properly support ESM. this pattern was a naive attempt to move that direction, but I'm almost certain there was a simpler way. 🙃

[JoshMock]

Suggested change

	onBeforeRequest = async (ctx: MiddlewareContext): Promise<MiddlewareResult | undefined> => {
	if (!this.shouldCompress(ctx)) {
	return undefined // void = no changes needed
	}
	onBeforeRequest = async (ctx: MiddlewareContext): Promise<MiddlewareResult | void> => {
	if (!this.shouldCompress(ctx)) {
	return // void = no changes needed
	}

void is a little clearer than undefined, IMO.

[JoshMock]

This is the bit we were worried would be a perf issue: deep-merging objects on a very hot code path could lead to a lot more garbage collection, which is a CPU hog.

[margaretjgu]

tried sth out here with benchmarked comparison b/w original code - original deep merge - optimized deep merge

#307