Breaking change in tracing setup, no more jaeger agent possible

[wkloucek]

Describe the bug

After #2372 it's not possible anymore to use jaeger agent. oCIS services now want to send traces directly to the jaeger collector.

This is the respective code path:

ocis/ocis-pkg/tracing/tracing.go

Lines 27 to 31 in d714cda

	exp, err := jaeger.New(
	jaeger.WithCollectorEndpoint(
	jaeger.WithEndpoint(collectorEndpoint),
	),
	)

This is a conflict to our recommendation on https://owncloud.dev/ocis/deployment/monitoring-tracing/ and the monitoring / tracing examples (https://github.com/owncloud-devops/monitoring-tracing-client, https://github.com/owncloud-devops/monitoring-tracing-server).

Especially for our fleet of continuously deployed oCIS examples, we are dependent on jaeger agent, since this supports adding tags:

https://github.com/owncloud-devops/monitoring-tracing-client/blob/176b5d268cd9d125a53dcd93dfd631226463e72a/docker-compose.yml#L53

These tags allow us to filter oCIS instances in our single jaeger instance (https://jaeger.infra.owncloud.works/search) where all oCIS demo instances are reporting to.

Steps to reproduce

Steps to reproduce the behavior:

1. start one of our oCIS deployment examples with the monitoring / tracing additions
   

   ocis/deployments/examples/ocis_traefik/monitoring_tracing/docker-compose-additions.yml

   Lines 5 to 8 in d714cda

   	ocis:
   	environment:
   	OCIS_TRACING_ENABLED: "true"
   	OCIS_TRACING_ENDPOINT: jaeger-agent:6831

2. start the monitoring / tracing client besides it (https://github.com/owncloud-devops/monitoring-tracing-client/blob/master/docker-compose.yml)

Expected behavior

oCIS sends traces to the jaeger agent, which forwards them to the jaeger collector and you can see them in the jaeger UI.

Actual behavior

oCIS says 2021/08/31 00:21:50 Post "": unsupported protocol scheme ""

[wkloucek]

@micbar @refs is this something for the fastlane or next sprint?

[refs]

mea culpa. Fastlane, I'll get to it now, as I believe it should be a quick one. Sorry for the regression, I assumed too much :)

[wkloucek]

mea culpa. Fastlane, I'll get to it now, as I believe it should be a quick one. Sorry for the regression, I assumed too much :)

No worries. Single oCIS deployments may also use the Collector directly, but since we have 10+ extensions / service creating traces, it also may make sense to always use an Agent!?

[refs]

it also may make sense to always use an Agent!?

I think you're right, as we should probably deploy an agent for example as a sidecar process with our services, whereas we would deploy only a collector, as its own service.

I would still keep creating an exporter from a collector endpoint if the config so requires it, but take precedence an agent. Opening a PR soon and we can move the discussion over there. Will link it here when I have it.

[exalate-issue-sync]

Alex Unger commented: PR in Reva: cs3org/reva#2029

PR in oCIS incoming.

[exalate-issue-sync]

Alex Unger commented: oCIS PR: #2457