Fix a bug which would drop samples in remote-write if the sending flow stalled
for longer than it takes to write one "WAL segment". How long this takes depends on the size
of your Prometheus; as a rough guide with 10 million series it is about 2-3 minutes.
* [BUGFIX] Remote-write: stop dropping samples in catch-up #14446
## 2.53.0 / 2024-06-16
## 2.53.0 / 2024-06-16
This release changes the default for GOGC, the Go runtime control for the trade-off between excess memory use and CPU usage. We have found that Prometheus operates with minimal additional CPU usage, but greatly reduced memory by adjusting the upstream Go default from 100 to 75.
This release changes the default for GOGC, the Go runtime control for the trade-off between excess memory use and CPU usage. We have found that Prometheus operates with minimal additional CPU usage, but greatly reduced memory by adjusting the upstream Go default from 100 to 75.