docs: mention that max-workers is capped by the number of partitions

This commit is contained in:
Vincent Bernat
2025-10-15 12:11:30 +02:00
parent 98d1028850
commit bd4a349fba
2 changed files with 10 additions and 5 deletions

View File

@@ -157,7 +157,9 @@ keys are accepted:
- `fetch-max-wait-time` defines the maximum time to wait for the minimum - `fetch-max-wait-time` defines the maximum time to wait for the minimum
number of bytes to become available. number of bytes to become available.
- `min-workers` defines the minimum number of Kafka workers to use. - `min-workers` defines the minimum number of Kafka workers to use.
- `max-workers` defines the maximum number of Kafka workers to use. - `max-workers` defines the maximum number of Kafka workers to use (it should
not be more than the number of partitions for the topic, as defined in
`kafka``num-partitions`)
- `worker-increase-rate-limit` defines the duration before increasing the - `worker-increase-rate-limit` defines the duration before increasing the
number of workers. number of workers.
- `worker-decrease-rate-limit` defines the duration before decreasing the - `worker-decrease-rate-limit` defines the duration before decreasing the

View File

@@ -443,10 +443,13 @@ akvorado_inlet_flow_input_udp_buffer_size_bytes{listener=":2055",worker="2"} 212
### Outlet ### Outlet
The outlet is expected to automatically scale the number of workers to ensure that the The outlet is expected to automatically scale the number of workers to ensure
data is delivered efficiently to ClickHouse. Increasing the maximum number of that the data is delivered efficiently to ClickHouse. Increasing the maximum
Kafka workers (`max-workers`) past the default value of 8 may put more pressure number of Kafka workers (`max-workers`) past the default value of 8 may put more
on ClickHouse. However, you can increase `maximum-batch-size`. pressure on ClickHouse. However, you can increase `maximum-batch-size`.
Moreover, there cannot be more workers than the number of partitions for the
topic. This part is configurable on the orchestrator
(`kafka``topic-configuration``num-partitions`).
The BMP routing component may have some challenges with scaling, especially when peers The BMP routing component may have some challenges with scaling, especially when peers
disappear, as it requires cleaning up many entries in the routing tables. BMP disappear, as it requires cleaning up many entries in the routing tables. BMP