William
William
The SHIFT_RANGE feature causes the queried time range to be rounded on the "current cadence". The current cadence is another way of saying how big the resulting bucket of results...
# Problem Quota limits are currently communicated through the `.limits` field in the MetricsResponse. This only includes information about what kind of quota was violated, while a user would mostly...
# Problem Querying Heroic with JSON can be verbose and hard to manually do. Heroic currently has its own query language, HQL, but we haven't kept it up to date...
Currently resource identifiers (RI's) are not filterable in visualization frontends like grafana. Since RI's are not indexed in elasticsearch, the filtering will have to happen after the metrics are fetched...
Similar to #245, we should have a Python client for interacting with Heroic. The initial implementation probably doesn't have to interact with the ingestion side of the pipeline at all,...
As pointed out by @dmichel1 in #497, our schemas are named with a `Spotify` prefix - eg `Spotify100` - but there is nothing specific to Spotify in it. We should...
For a new release, Heroic currently gets completely built twice. Travis builds and tests it, but doesn't publish the built jar. Then the Docker Hub triggers and does its own...
# Problem Kubernetes supports autoscaling through different sources. Right now using a metric from heroic to autoscale is cumbersome and involves shipping the metric to a second datastore that supports...
Metrics stored in Bigtable should be able to have a retention policy set via a column family garbage collection policy.
## Problem We want to make sampling a first class citizen rather than a side-effect of the first aggregation. This will address a number of problems: * Users are often...