[
https://issues.apache.org/jira/browse/CASSANDRA-18093?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Branimir Lambov updated CASSANDRA-18093:
----------------------------------------
Description:
When the number of racks is the same as the replication factor, token
allocation should use `NoReplicationTokenAllocator`. Because `generatetokens`
caches the allocation strategy
[here|https://github.com/apache/cassandra/commit/2346ed8241022882e77433e283ab8ce3004d12b0#diff-a93e5fa817f4b6d64484e0c28391b76e1760f51860de7f4f036470766ff5090cR225],
it uses one that was created before all the racks are defined and does not
switch to the correct one.
This has the effect of constructing very poor allocations for the racks = RF
case, especially for the first couple of nodes per rack.
was:
When the number of racks is the same as the replication factor, token
allocation should use `NoReplicationTokenAllocator`. Because `generatetokens`
caches the allocation strategy
[here|https://github.com/apache/cassandra/commit/2346ed8241022882e77433e283ab8ce3004d12b0#diff-a93e5fa817f4b6d64484e0c28391b76e1760f51860de7f4f036470766ff5090cR225],
it uses one that was created before all the racks are defined and does not
switch to the correct one.
This has the effect of constructing very poor allocations for the racks = RF
case, especially for the first couple of nodes.
> `generatetokens` uses the wrong allocation strategy
> ---------------------------------------------------
>
> Key: CASSANDRA-18093
> URL: https://issues.apache.org/jira/browse/CASSANDRA-18093
> Project: Cassandra
> Issue Type: Bug
> Reporter: Branimir Lambov
> Priority: Normal
>
> When the number of racks is the same as the replication factor, token
> allocation should use `NoReplicationTokenAllocator`. Because `generatetokens`
> caches the allocation strategy
> [here|https://github.com/apache/cassandra/commit/2346ed8241022882e77433e283ab8ce3004d12b0#diff-a93e5fa817f4b6d64484e0c28391b76e1760f51860de7f4f036470766ff5090cR225],
> it uses one that was created before all the racks are defined and does not
> switch to the correct one.
> This has the effect of constructing very poor allocations for the racks = RF
> case, especially for the first couple of nodes per rack.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
