The title "Single Multi Data Centers Configuration Table" is confusing. Single seems like copy-paste leftover
Also, "endpoint_snitch" seems irrelevant to nodes having both internal and external addresses.
"If the node has two physical network interfaces in a multi-datacenter installation:"
the documentation here seems partial
here's also the listen_interface and listen_interface_prefer_ipv6 config params
and respectively, rpc_interface and rpc_interface_prefer_ipv6 for the client facing interface (related to rpc_address and broadcast_rpc_address.
change "storage_port or ssl_storage_port" -> "storage_port and/or ssl_storage_port" since some inter-node encryption settings require both to be opened.
This is not specific to having 2 physical interfaces.
It is applicable to single interface as well.
In "For a production system, it is recommended to use a DC-aware snitch", change "recommended" to "required".
"Decide the name of the rack/dc": swap the order. The DC is always specified before the RACK.
"Choose the data center name carefully. It is not possible to rename a data center later"
"On each node, edit the scylla.yaml file to configure the parameters listed below"
This contains just a partial list of configuration parameters.
In particular is does not describe how listen_address and broadcast_address are related and how they are used, and similarly for rpc_address and broadcast_rpc_address.
It doesn't mention compression and encryption (maybe it is in the enterprise version?)
We should probably have a more comprehensive networking configuration guide and refer to that form both places.
"In the cassandra-rackdc.properties file, edit the rack and data center information"
ditto. swap "data center" and "rack" in the sentence.
"To save bandwidth, add the prefer_local=true parameter. Scylla will use the node private (local) IP address when the nodes are in the same data center."
This doesn't save bandwidth per-se. But rather it could save WAN bandwidth and networking costs in the cloud.
tzach
commented
3 months ago
I would like to report an issue on page https://opensource.docs.scylladb.com/branch-6.0/operating-scylla/procedures/cluster-management/create-cluster-multidc
@elcallio please review this page as well.
Problem
The title "Single Multi Data Centers Configuration Table" is confusing.
Singleseems like copy-paste leftover"If the node has two physical network interfaces in a multi-datacenter installation:"
listen_interfaceandlisten_interface_prefer_ipv6config paramsrpc_interfaceandrpc_interface_prefer_ipv6for the client facing interface (related torpc_addressandbroadcast_rpc_address.change "storage_port or ssl_storage_port" -> "storage_port and/or ssl_storage_port" since some inter-node encryption settings require both to be opened.
In "For a production system, it is recommended to use a DC-aware snitch", change "recommended" to "required".
"Decide the name of the rack/dc": swap the order. The DC is always specified before the RACK.
"Choose the data center name carefully. It is not possible to rename a data center later"
"On each node, edit the scylla.yaml file to configure the parameters listed below"
listen_addressandbroadcast_addressare related and how they are used, and similarly forrpc_addressandbroadcast_rpc_address."In the cassandra-rackdc.properties file, edit the rack and data center information"
"To save bandwidth, add the prefer_local=true parameter. Scylla will use the node private (local) IP address when the nodes are in the same data center."
Suggest a fix