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
Trying to use MetricsQL-only functions in Sloth results in an error. The actual promparse errors are not passed to the user, but if you wire up the promql parser manually, you can get them. Depending on the query, you might see, e.g.:
1:1: parse error: unknown function with name "histogram_share"
To get an le-labeled equivalent, you need to use the MetricsQL-specific function prometheus_buckets, so building a good latency SLI against anything using a victoriametrics-format histogram - including victoriametrics itself - seems to be impossible at the moment.
Another alternative suggested on in the LogQL issue would be to add a flag allowing validation to be skipped. Happy to work on either approach, and on making the parse errors more visible, if a fix would be welcomed.
The text was updated successfully, but these errors were encountered:
First, thanks for Sloth!
This issue is like #280 in concept, just for a different fork of PromQL - https://docs.victoriametrics.com/metricsql/
Trying to use MetricsQL-only functions in Sloth results in an error. The actual promparse errors are not passed to the user, but if you wire up the promql parser manually, you can get them. Depending on the query, you might see, e.g.:
The issue is exacerbated relative to LogQL as victoriametrics pushes its own custom histograms (a bit similar to the new prometheus native histogram) which use a
vmrange
label instead ofle
: https://valyala.medium.com/improving-histogram-usability-for-prometheus-and-grafana-bc7e5df0e350To get an
le
-labeled equivalent, you need to use the MetricsQL-specific functionprometheus_buckets
, so building a good latency SLI against anything using a victoriametrics-format histogram - including victoriametrics itself - seems to be impossible at the moment.Another alternative suggested on in the LogQL issue would be to add a flag allowing validation to be skipped. Happy to work on either approach, and on making the parse errors more visible, if a fix would be welcomed.
The text was updated successfully, but these errors were encountered: