You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
@kyessenov Thank you for the info! That makes sense, although there seems to be a discrepancy between Prometheus' base units (in seconds), and the semantic conventions in the http metrics by OpenTelemetry (in milliseconds).
It seems like Istio is adhering to the OpenTelemetry convention, so I'll close this issue for now.
but as long as the unit metadata is in there, then I think this is still correct and there's nothing to fix:
The OpenMetrics UNIT metadata, if present, MUST be converted to the unit of the OTLP metric. After trimming type-specific suffixes, such as _total for counters, the unit MUST be trimmed from the suffix as well, if the metric suffix matches the unit. The unit SHOULD be translated from Prometheus conventions to OpenTelemetry conventions
Bug Description
The
istio_request_duration_milliseconds
metric isn't adhering to the Prometheus standard practice for using seconds as the base unit for time: https://prometheus.io/docs/practices/naming/#base-unitsSome libraries assume the buckets to be in seconds, which can cause unexpected behavior, like this one: pyrra-dev/pyrra#667
I plan to work around this issue by effectively relabeling the metrics with a Prometheus recording rule, but perhaps this metric needs to be in milliseconds for some reason?
Also, it looks like this metric used to be in seconds, but was changed to milliseconds during the upgrade to in-proxy telemetry (aka v2) from Mixer-based telemetry (aka v1):
https://istio.io/latest/about/faq/metrics-and-logs/
https://istio.io/v1.6/docs/reference/config/policy-and-telemetry/metrics/
Version
The text was updated successfully, but these errors were encountered: