[
https://issues.apache.org/jira/browse/CASSANDRA-20197?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Dmitry Konstantinov updated CASSANDRA-20197:
--------------------------------------------
Description:
Background: It is a quite frequent issue when by mistake for a Cassandra
cluster an incremental repair is triggered instead of a full one (incremental
as a default option helps with it). As a result we have SSTables marked within
non-zero "Repaired At" value and they are isolated by compaction logic to a
separate SSTable pool. If incremental repair is not executed regularly after it
we have issues like uncleaned tombstones sitting forever within the table even
after a major compaction.
The current way to solve the issue is to use
[sstablerepairedset|https://cassandra.apache.org/doc/stable/cassandra/tools/sstable/sstablerepairedset.html]
command but we have to stop Cassandra node to run it, so to recover from the
issue described before we have to do a rolling restart of Cassandra nodes - it
is time consuming and affects performance due to a warmup period after a
restart.
The improvement is about introducing an nodetool command which can ask
Cassandra node to reset "Repaired At" value for local SSTables of a table in
runtime without an node shutdown.
A command can look like:
{code:java}
nodetool set-unrepaired some_keyspace.some_table
nodetool set-unrepaired some_keyspace{code}
was:
Background: It is a quite frequent issue when by mistake for a Cassandra
cluster and incremental repair is triggered instead of a full one (incremental
as a default option helps with it). As a result we have SSTable marked within
non-zero "Repaired At" value and the are separated by compaction logic to a
separate SSTable pool. If incremental repair is not executed regularly after it
we have issues like uncleaned tombstones sitting forever within the table even
after a major compaction.
The current way to solve the issue is to use
[sstablerepairedset|https://cassandra.apache.org/doc/stable/cassandra/tools/sstable/sstablerepairedset.html]
command but we have to stop Cassandra node to run it, so to recover from the
issue described before we have to do a rolling restart of Cassandra nodes - it
is time consuming and affects performance due to a warmup period after a
restart.
The improvement is about introducing an nodetool command which can ask
Cassandra node to reset "Repaired At" value for local SSTables of a table in
runtime without an node shutdown.
A command can look like:
{code:java}
nodetool set-unrepaired some_keyspace.some_table
nodetool set-unrepaired some_keyspace{code}
> Provide the ability to reset Repaired At metadata without Cassandra node
> shutdown
> ---------------------------------------------------------------------------------
>
> Key: CASSANDRA-20197
> URL: https://issues.apache.org/jira/browse/CASSANDRA-20197
> Project: Apache Cassandra
> Issue Type: Improvement
> Components: Consistency/Repair
> Reporter: Dmitry Konstantinov
> Priority: Normal
>
> Background: It is a quite frequent issue when by mistake for a Cassandra
> cluster an incremental repair is triggered instead of a full one (incremental
> as a default option helps with it). As a result we have SSTables marked
> within non-zero "Repaired At" value and they are isolated by compaction logic
> to a separate SSTable pool. If incremental repair is not executed regularly
> after it we have issues like uncleaned tombstones sitting forever within the
> table even after a major compaction.
> The current way to solve the issue is to use
> [sstablerepairedset|https://cassandra.apache.org/doc/stable/cassandra/tools/sstable/sstablerepairedset.html]
> command but we have to stop Cassandra node to run it, so to recover from the
> issue described before we have to do a rolling restart of Cassandra nodes -
> it is time consuming and affects performance due to a warmup period after a
> restart.
> The improvement is about introducing an nodetool command which can ask
> Cassandra node to reset "Repaired At" value for local SSTables of a table in
> runtime without an node shutdown.
> A command can look like:
> {code:java}
> nodetool set-unrepaired some_keyspace.some_table
> nodetool set-unrepaired some_keyspace{code}
>
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
