[
https://issues.apache.org/jira/browse/CASSANDRA-12106?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Geoffrey Yu updated CASSANDRA-12106:
------------------------------------
Fix Version/s: 4.x
Status: Patch Available (was: Open)
I’ve attached a patch that implements this. There are a lot of changes, so I
thought I’d highlight the high level approach I took to make it easier to
understand what is going on.
This patch will let us blacklist any particular CQL partition, scoped by
keyspace and table. Any reads/writes to a blacklisted partition will be
rejected, and the client will receive a Read/WriteRejectedException
accordingly. The mechanism for blacklisting a partition is exposed through a
node tool command.
The approach to implementing this is to perform the rejection at the data
replica level. This is so each node only needs to be aware of blacklisted
partitions for ranges that it owns, allowing this to scale to larger clusters.
The blacklist is stored in a new table under the {{system_distributed}}
keyspace. Each node then maintains an in memory cache to store blacklist
entries corresponding to its token ranges.
For single partition reads and writes, we reject the request as long as one
replica responds with a rejection.
For partition range reads, we reject the request if there is a blacklisted
partition within the range.
CAS writes are rejected by the data nodes only on the prepare/promise step, and
potentially when the coordinator performs the read before the propose/accept
step. If the write proceeds past these places, the mutation will be allowed to
be applied.
A mutation in a batch log that is rejected will not be considered a "failure"
in the replay. This means if all mutations were either applied successfully or
rejected, then the replay would be considered successful and the batch log
would be deleted.
Mutations that are rejected are not hinted. Any hints that are rejected when
they are attempted to be applied will still be considered "successful" and
deleted.
There are also changes included to help cover cache consistency when a node
starts up, undergoes a range movement, and if it is decommissioned.
> Add ability to blacklist a CQL partition so all requests are ignored
> --------------------------------------------------------------------
>
> Key: CASSANDRA-12106
> URL: https://issues.apache.org/jira/browse/CASSANDRA-12106
> Project: Cassandra
> Issue Type: New Feature
> Reporter: Geoffrey Yu
> Assignee: Geoffrey Yu
> Priority: Minor
> Fix For: 4.x
>
> Attachments: 12106-trunk.txt
>
>
> Sometimes reads/writes to a given partition may cause problems due to the
> data present. It would be useful to have a manual way to blacklist such
> partitions so all read and write requests to them are rejected.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
