In the most recent release, I made a category error when filtering statistics for the shutdown() code path.
A community member was quick to point out the error and file a PR to fix it, so we got a hotfix out before most people had even discovered the new release.
This has been bugging me ever since because it feels like a footgun. Burying this assumption down in the aggregate() code makes it a bit of tribal knowledge instead of something people can discover for themselves.
So I'm thinking out loud about whether we should be setting a default aggregator in metric.js, gauge.js, histogram.js, etc such that a person sitting at a breakpoint trying to figure out why their graphs aren't graphing has a better chance of catching a problem such as the wrong aggregator being picked for their stat, instead of leaving it to collection time. My feeling is that the average user will look in the vicinity of when a stat is recorded rather than when it is scraped.
In the most recent release, I made a category error when filtering statistics for the shutdown() code path.
A community member was quick to point out the error and file a PR to fix it, so we got a hotfix out before most people had even discovered the new release.
This has been bugging me ever since because it feels like a footgun. Burying this assumption down in the aggregate() code makes it a bit of tribal knowledge instead of something people can discover for themselves.
So I'm thinking out loud about whether we should be setting a default aggregator in metric.js, gauge.js, histogram.js, etc such that a person sitting at a breakpoint trying to figure out why their graphs aren't graphing has a better chance of catching a problem such as the wrong aggregator being picked for their stat, instead of leaving it to collection time. My feeling is that the average user will look in the vicinity of when a stat is recorded rather than when it is scraped.