Post

Full opentelemetry support for all signals (logs, metrics, traces) and auto instrumentation of popular libraries that doesn't flood with noise. Exceptions are just trace events with a certain type :) But, I don't know why you'd put yourself in such pain implementing this ;)

I'm assuming you use dependencies like those listed in opentelemetry.io/docs/languag.... Do you like the experience from Elixir's side? I've often wondered if it wouldn't be nicer for me to report from my app with Elixir's telemetry, and the library translates that to OpenTelemetry stuff.

I did, and then didn't, because of various issues or deprecations or flexibility. The phoenix and cowboy ones are too noisy, the liveview one was useless. The oban one polluted Ecto iirc, and at the time Datadog was using deprecated keys, so I had to roll my own. No tail sampling made things hard

I heard there was a talk about this recently. 🤔 www.youtube.com/watch?v=irQi... (still not public, but... relevant)

I'd appreciate "OpenException" so that you can easily report to Sentry/Bugsnag and have that logic bundled up and only have adapters to actually send the event. But I don't think the need is too high. I'd appreciate an elixir focussed tool, friend of mine once built one.

Not sure how big the need is, granted that we have opentelemetry and friends.

Do you like the current integration of Elixir apps with opentelemetry right now? (I mean the Elixir-side of adding open telemetry)

A small but outsized-impact feature I've not seen mentioned yet: NR's "notice deployments" feature, with support for uploading a link to a diff/PR on a git host (and other arbitrary metadata)

Interesting! How would you use that feature? Do you add any specific metadata? And what does it allow you to do?

As @nezteb.net points out, the docs on this feature sucks now, used to be a lot better. Will fit an explanation in a 🧵:

I tried searching "notice deployments" but could only find: docs.newrelic.com/docs/change-... Is there some other link I should be looking at? 😄

Support for development deployments that would not require me to setup any accounts or do any fancy setup. I want just be able to run observability stack locally. I was even planning on writing something like that with just in-memory storage, so I can test and use it in development as well.

Speaking for my company's needs: - Alerting via Slack the first time a particular crash is observed in prod - Alerting via Slack when the overall rate of errors reaches x/min

- Better grouping of errors than the big player give (they both group things that absolutely are not related, and fail to group things that a human can tell are obviously the same) - Hosting in our own infra would be a huge plus, because we wouldn't have to worry about PII in backtraces

Things we do not really need: - Creating or assigning tickets from a crash (we trigger that via Slack) - Uptime monitoring (we get that from a bunch of other tools) - Response time monitoring (we get that from Grafana, and to be honest if something's not triggering DB timeouts, we just don't worry)