a09465baee
* storage/remote: disable resharding during active retry backoffs Today, remote_write reshards based on pure throughput. This is problematic if throughput has been diminished because of HTTP 429s; increasing the number of shards due to backpressure will only exacerbate the problem. This commit disables resharding for twice the retry backoff, ensuring that resharding will never occur during an active backoff, and that resharding does not become enabled again until enough time has elapsed to allow any pending requests to be retried. Signed-off-by: Robert Fratto <robertfratto@gmail.com> * storage/remote: test that resharding is disabled on retry Signed-off-by: Robert Fratto <robertfratto@gmail.com> * storage/remote: address review feedback Signed-off-by: Robert Fratto <robertfratto@gmail.com> * storage/remote: track time where resharding initially got disabled This change introduces a second atomic int64 to roughly track when resharding got disabled. This int64 is only updated after updating the disabled timestamp if resharding was previously enabled. Signed-off-by: Robert Fratto <robertfratto@gmail.com> --------- Signed-off-by: Robert Fratto <robertfratto@gmail.com> |
||
|---|---|---|
| .. | ||
| azuread | ||
| otlptranslator | ||
| chunked.go | ||
| chunked_test.go | ||
| client.go | ||
| client_test.go | ||
| codec.go | ||
| codec_test.go | ||
| ewma.go | ||
| intern.go | ||
| intern_test.go | ||
| max_timestamp.go | ||
| metadata_watcher.go | ||
| metadata_watcher_test.go | ||
| queue_manager.go | ||
| queue_manager_test.go | ||
| read.go | ||
| read_handler.go | ||
| read_handler_test.go | ||
| read_test.go | ||
| storage.go | ||
| storage_test.go | ||
| write.go | ||
| write_handler.go | ||
| write_handler_test.go | ||
| write_test.go | ||