[
https://issues.apache.org/jira/browse/CASSANDRA-14694?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16611015#comment-16611015
]
Benedict commented on CASSANDRA-14694:
--------------------------------------
While we're at this patch, could we re-bikeshed the naming of
speculative_write_threshold?
I had hoped we would call it transient_write_threshold, since the writes - once
they happen - are no longer speculative. They're just writes. Admittedly,
they're happening 'speculatively' in the expectation we may not reach
consistency without them (even though we might).
Perhaps we should have a separate statistic for transient writes that includes
those we may not through 'speculation' but because the failure detector informs
us we need it to reach consistency, and retain the speculative_write.
Though these repair writes are also speculative, so we might at least need an
extra component, e.g. speculative_transient_write and speculative_rr_write
I thought this would be a good spot to generally discuss our terminology around
these kinds of actions, anyway.
> add latency sample for speculative read repair writes
> -----------------------------------------------------
>
> Key: CASSANDRA-14694
> URL: https://issues.apache.org/jira/browse/CASSANDRA-14694
> Project: Cassandra
> Issue Type: Bug
> Reporter: Blake Eggleston
> Assignee: Blake Eggleston
> Priority: Major
> Fix For: 4.0
>
>
> Speculative read repair mutations shouldn't use read latencies to determine
> when to send a speculative mutation. It should have it's own value based on
> mutation latencies.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
---------------------------------------------------------------------
To unsubscribe, e-mail: commits-unsubscr...@cassandra.apache.org
For additional commands, e-mail: commits-h...@cassandra.apache.org
