Telemetry drop assertions

dropped: true passes when nothing reaches any exporter:

- name: health checks are dropped
  input: fixtures/healthz-span.json
  expect:
    - dropped: true

Use it for filters, sampling policies and anything else that should discard the fixture completely.

What "dropped" requiresCopy Anchor Link

A dropped assertion passes only if all of the following hold:

1. the Collector accepted the fixture (HTTP 2xx, no partialSuccess rejection);
2. the Collector stayed up for the whole window and exited cleanly when stopped;
3. the capture server rejected no requests;
4. zero items reached any exporter during the full timeout and the graceful shutdown.

Conditions 1–3 are checked for every test. When one fails, the test is an error (exit code 5), not a pass. A broken pipeline therefore can't look like a successful filter.

Shutdown matters: batch and tail_sampling flush buffered telemetry during graceful shutdown, and that output is counted. A dropped test can't pass just because a batch timeout was longer than the observation window.

On failure, the output shows what was exported, and where:

✗ telemetry is dropped
  Expected: nothing exported
  Actual:   1 span exported
  Received: otlp_http/backend: 1 span

dropped is not vacuousCopy Anchor Link

When the config changes, the same fixture is exported and the test fails. The sampling example proves this with two of its regression patches: removing the kube-probe user-agent rule, and adding an always_sample baseline policy. Each makes a dropped: true test fail.

RulesCopy Anchor Link

• dropped: false isn't supported. Assert where telemetry should arrive with exporter:.
• dropped can't be combined with exporter: or count:. To assert that only one exporter received nothing, use exporter: E with count: 0.
• dropped can't be combined with attribute: in the same assertion.

CaveatsCopy Anchor Link

• groupbytrace discards pending traces at shutdown. If its wait_duration (default 1 s) isn't shorter than the test timeout, a trace still waiting is discarded and dropped passes for the wrong reason. Tracelet warns when this applies.
• Windows. The Collector is killed rather than shut down, so buffered telemetry is never flushed. Keep batch timeouts and decision_wait well below the test timeout there.