Collect cloud metadata

[felixbarny]

see elastic/apm#290

see also current spec doc: https://github.com/elastic/apm/blob/3f224a9cbacb196bbb196b0868d9fa90a0bbaf6c/specs/agents/metadata.md#cloud-provider-metadata

[JCMais]

How would someone go about implementing this?

From what I can see the metadata itself will need to be added to

apm-agent-nodejs/lib/config.js

Lines 239 to 249 in cde2dde

	// metadata
	agentName: 'nodejs',
	agentVersion: version,
	serviceName: conf.serviceName,
	serviceNodeName: conf.serviceNodeName,
	serviceVersion: conf.serviceVersion,
	frameworkName: conf.frameworkName,
	frameworkVersion: conf.frameworkVersion,
	globalLabels: maybePairsToObject(conf.globalLabels),
	hostname: conf.hostname,
	environment: conf.environment,

(which means adding it to the apm-nodejs-http-client package)

net.createConnection() is probably the fastest way to check if the metadata server exists, if it does a request can be made to retrieve the information. This is basically what was done on the reference implementation: https://github.com/basepi/apm-agent-python/blob/f13110156e57eec99134a7f9342243416b969698/elasticapm/utils/cloud.py#L79-L115

But the biggest "issue" so far, is that the config generation is synchronous.

A possible solution to the above that does not require a big refactor would be to create the "cloud metadata fetcher" as a function that accepts a callback and then pass that to ElasticAPMHttpClient. Writing to the stream could then be deferred until the callback has been called: https://github.com/elastic/apm-nodejs-http-client/blob/1526dedc26392bd92f20b5079f848c7e77567e7c/index.js#L496-L500

Something like:

    // All requests to the APM Server must start with a metadata object
    const encodeMetadataAndWriteStream = () => {
      if (!client._encodedMetadata) {
        client._encodedMetadata = client._encode({ metadata: client._conf.metadata }, Client.encoding.METADATA)
      }

      stream.write(client._encodedMetadata)
    }

    const { metadata } = client._conf

    if (metadata._cloudFetcher && !metadata.cloud) {
      metadata._cloudFetcher((cloud, _error) => {
        // do something with error?
        metadata.cloud = cloud
        encodeMetadataAndWriteStream()
      })
    } else {
      encodeMetadataAndWriteStream()
    }

[felixbarny]

That sounds like a good plan and is in line with what the spec says:

Any intake API requests to the APM server should be delayed until this metadata is available.

Any events that are created before the metadata is ready would be buffered in the meantime.

[JCMais]

I've started working on this at #1816

[astorm]

With #1937 and elastic/apm-nodejs-http-client#129 landed this work is done, and we be available after the next http client and agent release.