Elixir response time missing when using Bandit

[sgessa]

Describe the bug

When using Bandit I am unable to see the Elixir response time in New Relic dashboard (I still see Postgres and total response time tho).
I started receiving crash reports that might be related:

KeyError: NewRelic.Transaction.Complete.report_transaction_event/1

key :host not found in: %{
  name: "Unknown/Unknown",
  pid: "#PID<0.155834.0>",
  priority: 0.799115,
  reductions: 21264818,
  status: 500,
  start_time: 1738663081850,
  guid: "213aea1b85585908",
  sampled: false,
  traceId: "d3f468594a6581f76f487d6998bb71c1",
  duration_s: 2.3400000000000002e-4,
  total_time_s: 0.0,
  memory_kb: 34.6640625,
  end_time: 1738663081850.234,
  duration_ms: 0.234,
  transactionType: :Web,
  process_spawns: -1,
  "nr.apdexPerfZone": "S",
  duration_us: 234
}

lib/new_relic/transaction/complete.ex in NewRelic.Transaction.Complete.report_transaction_event/1 at line 458
lib/new_relic/transaction/complete.ex in NewRelic.Transaction.Complete.run/2 at line 19
lib/new_relic/transaction/sidecar.ex in NewRelic.Transaction.Sidecar.handle_continue/2 at line 224
gen_server.erl in :gen_server.try_handle_continue/3 at line 1085
gen_server.erl in :gen_server.loop/7 at line 995
proc_lib.erl in :proc_lib.init_p_do_apply/3 at line 241

Environment

Erlang/OTP 26 [erts-14.2.5] [source] [64-bit] [smp:16:16] [ds:16:16:10] [async-threads:1] [jit:ns]
Elixir 1.17.3 (compiled with Erlang/OTP 26)

new_relic_agent 1.31.1

bandit 1.6.7

[binaryseed]

Thanks for the report. I don't think the two things are related. I have a hunch I'm already tracking down related to instrumenting bandit that might be related.

This error is interesting though, looks like the transition completed w/o setting a host attribute, which I'm not sure how that could happen since it always gets set when a web transaction is opened.

Are you doing anything out of the ordinary, perhaps trying to start a web transaction manually in your app?

[sgessa]

Nope, I am not doing anything manually..
I also found a post on elixir forums from someone having the same issue...

[sgessa]

Oh I'm calling NewRelic.ignore_transaction/0 in one my controller used for heartbeat, that's the only manual thing I do.
And the frequency of crash reports are similar to the interval of my heartbeat calls but it might just be a coincidence.

[binaryseed]

Will you share a link to that elixir forum post?

[sgessa]

Sorry the issue wasn't related after all.

This is what I see in NewRelic by the way, Elixir response time is showing with Cowboy then disappeared with Bandit.

[TylerWitt]

This error is interesting though, looks like the transition completed w/o setting a host attribute, which I'm not sure how that could happen since it always gets set when a web transaction is opened.

I think this could happen on ignored paths.

elixir_agent/lib/new_relic/telemetry/plug.ex

Line 79 in 82a0b1e

with :collect <- Transaction.Reporter.start_transaction(:web, path(meta, server)) do

start_transaction returns :ignore, so the attributes from the plug telemetry (including host) never get set.

[binaryseed]

That wouldn't cause the error... when you call NewRelic.ignore_transaction, we end the transaction right away, and don't call complete on it (since that is what would report the transaction)

[binaryseed]

I was able to reproduce the error locally and have a PR to fix it: #497

The response time metric is a separate issue that I aim to track down soon