n.v.harikrishna created CASSANDRA-21770:
-------------------------------------------
Summary: Add node-centric interface for identifying replica peers
Key: CASSANDRA-21770
URL: https://issues.apache.org/jira/browse/CASSANDRA-21770
Project: Apache Cassandra
Issue Type: Improvement
Components: Feature/Virtual Tables, Tool/nodetool
Reporter: n.v.harikrishna
Assignee: n.v.harikrishna
*Problem*
Operators often need to bounce, upgrade or replace nodes one after another. To
keep data available, they must not take down a node and the nodes that hold
copies of the same data (its replica peers) at the same time.
The tools Cassandra has today, such as nodetool describering and nodetool
getendpoints, are range-centric or key-centric. They answer "which nodes own
this range or key?" Operators need the node-centric answer: "which nodes share
data with this node?" To get it now, they have to put it together themselves
from token ranges, keyspaces and replication settings. This is slow and easy to
get wrong, especially while nodes are joining or leaving the cluster.
*Proposed solution*
Add a node-centric view of replica peers: for each node, list the other nodes
that share at least one token range with it. Peers are grouped by datacenter.
Reads and writes are listed separately, because they can differ while nodes are
joining or leaving. The data comes from the cluster metadata. This view is
exposed through:
- A new nodetool command: _nodetool placement peers_. It can filter by node
(-n), datacenter (-dc) and keyspace (-k / -x). It prints YAML or JSON.
- A new JMX operation in StorageServiceMBean.
- A new virtual table, _system_views.replica_peers_, which shows the peers of
the local node.
_Sample output:_
{noformat}
$ nodetool placement peers
version: 0
format_version: 1
epoch: 9
keyspaces: [ks1, system_auth, system_cluster_metadata, system_distributed,
system_traces]
endpoints: {1: '127.0.0.1:7005', 2: '127.0.0.3:7005', 3: '127.0.0.2:7005'}
peers: 1:
read: datacenter1: [2, 3]
write: datacenter1: [2, 3]
2:
read: datacenter1: [1, 3]
write: datacenter1: [1, 3]
3:
read: datacenter1: [1, 2]
write: datacenter1: [1, 2]
catch_up: SKIPPED
cms: INITIALIZED
queried_from: 127.0.0.1:7005
{noformat}
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
