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
Is your feature request related to a problem? Please describe.
When running Cortex with tenants ingesting at different rates, a single user may greatly increase their ingestion rate and cause distributors to get OOMKilled affecting ingestion for other users. Distributor should allow per-user throttling on ingestion to prevent noisy neighbors.
Describe the solution you'd like
Add a per-user token bucket in distributors to limit ingestion/throttle single users, similar to how store-gateway has token bucket throttling.
Describe alternatives you've considered
I've considered using distributor limits like max_inflight_push_requests to prevent distributors from getting into high CPU and memory situations overall but then distributors may drop requests from smaller users in the cluster if there is a large user taking up most of the ingestion room.
ingestion_rate limit is also per-user but it limits the number of overall samples, and not the number of requests so a user with a small batch size but high TPS could still cause problems.
The text was updated successfully, but these errors were encountered:
Is your feature request related to a problem? Please describe.
When running Cortex with tenants ingesting at different rates, a single user may greatly increase their ingestion rate and cause distributors to get OOMKilled affecting ingestion for other users. Distributor should allow per-user throttling on ingestion to prevent noisy neighbors.
Describe the solution you'd like
Add a per-user token bucket in distributors to limit ingestion/throttle single users, similar to how store-gateway has token bucket throttling.
Describe alternatives you've considered
I've considered using distributor limits like
max_inflight_push_requests
to prevent distributors from getting into high CPU and memory situations overall but then distributors may drop requests from smaller users in the cluster if there is a large user taking up most of the ingestion room.ingestion_rate
limit is also per-user but it limits the number of overall samples, and not the number of requests so a user with a small batch size but high TPS could still cause problems.The text was updated successfully, but these errors were encountered: