[ 
https://issues.apache.org/jira/browse/CASSANDRA-17679?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17574964#comment-17574964
 ] 

Josh McKenzie commented on CASSANDRA-17679:
-------------------------------------------

{quote}Also I would not mention ZCS as a reason to disable this. ZCS applies 
pretty narrowly right now since it only works for whole files. So for the most 
part only works if you are using LCS, which is not the default compaction 
strategy.
{quote}
Streaming in general is faster in 4.0 even without taking ZCS into account so 
again, somewhat mitigated. Also, data resurrection is Bad; until someone steps 
forward to fix the feature I don't think we should enable it by default even if 
we haven't tackled the "bootstrapping large nodes is a pain" problem. Unless 
you and others are advocating for the counter position that resurrection in 
some scenarios is acceptable? And/or the scenarios enumerated on the original 
ticket aren't actually an issue?
{quote}We already have the "-Dcassandra.reset_bootstrap_progress" flag which 
disables resuming
{quote}
While we could better document the flag, having our default behavior of the 
database be "may resurrect deleted data" is pretty unfortunate.

> Make resumable bootstrap feature optional
> -----------------------------------------
>
>                 Key: CASSANDRA-17679
>                 URL: https://issues.apache.org/jira/browse/CASSANDRA-17679
>             Project: Cassandra
>          Issue Type: New Feature
>          Components: Consistency/Streaming
>            Reporter: Josh McKenzie
>            Assignee: Josh McKenzie
>            Priority: Normal
>             Fix For: 4.x
>
>          Time Spent: 10m
>  Remaining Estimate: 0h
>
> From the patch I'm working on:
> {code}
> # In certain environments, operators may want to disable resumable bootstrap 
> in order to avoid potential correctness
> # violations or data loss scenarios. Largely this centers around nodes going 
> down during bootstrap, tombstones being
> # written, and potential races with repair. By default we leave this on as 
> it's been enabled for quite some time,
> # however the option to disable it is more palatable now that we have zero 
> copy streaming as that greatly accelerates
> # bootstraps. This defaults to true.
> # resumable_bootstrap_enabled: true
> {code}
> Not really a great fit for guardrails as it's less a "feature to be toggled 
> on and off" and more a subset of a specific feature that in certain 
> circumstances can lead to issues.



--
This message was sent by Atlassian Jira
(v8.20.10#820010)

---------------------------------------------------------------------
To unsubscribe, e-mail: commits-unsubscr...@cassandra.apache.org
For additional commands, e-mail: commits-h...@cassandra.apache.org

Reply via email to