[
https://issues.apache.org/jira/browse/CASSANDRA-13959?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16234068#comment-16234068
]
Jeff Jirsa edited comment on CASSANDRA-13959 at 11/1/17 1:35 PM:
-----------------------------------------------------------------
I didn't say nobody was running MVs at scale - I said nobody was doing it AND
looking for errors. I'm confident in that assertion because had they done so,
they would have found the 5 8099 correctness-impacting bugs I linked, which you
would certainly hit as you did range through a table (or view) to check the
data matched. Those bugs and ~8 years of intuition are all I need to vote in
favor of this.
If you're willing to assert that there exists a competent team running MVs in
multiple DCs with multiple racks, writing data into the table with nontrivial
volumes and non-trivial patterns (new rows, overwrites, TTLs, range deletes,
partition deletes), adding/removing nodes, growing the cluster, shrinking the
cluster, running full repair, running incremental repair, killing disks,
replacing instances, and then verifying that EVERY SINGLE WRITE *and* EVERY
SINGLE DELETE is present with the expected consistency guarantees, by all
means, say so and I'll vote to remove the flag in 4.0.
I'm sure people are running MVs at scale, I just don't believe that they're
actively hunting for bugs or trying to prove correctness. But you've got a
binding -1 now, if you really believe this is hurting users, you know what to
do.
was (Author: jjirsa):
I didn't say nobody was running MVs at scale - I said nobody was doing it AND
looking for errors. I'm confident in that assertion because had they done so,
they would have found the 5 8099 correctness-impacting bugs I linked, which you
would certainly hit as you did range through a table (or view) to check the
data matched. Those bugs and ~8 years of intuition are all I need to vote in
favor of this.
If you're willing to assert that there exists a competent team running MVs in
multiple DCs with multiple racks, writing data into the table with nontrivial
volumes and non-trivial patterns (new rows, overwrites, TTLs, range deletes,
partition deletes), adding/removing nodes, growing the cluster, shrinking the
cluster, running full repair, running incremental repair, killing disks,
replacing instances, and then verifying that EVERY SINGLE WRITE is present with
the expected consistency guarantees, by all means, say so and I'll vote to
remove the flag in 4.0.
I'm sure people are running MVs at scale, I just don't believe that they're
actively hunting for bugs or trying to prove correctness. But you've got a
binding -1 now, if you really believe this is hurting users, you know what to
do.
> Add yaml flag for disabling MVs, log warnings on creation
> ---------------------------------------------------------
>
> Key: CASSANDRA-13959
> URL: https://issues.apache.org/jira/browse/CASSANDRA-13959
> Project: Cassandra
> Issue Type: Bug
> Components: Materialized Views
> Reporter: Blake Eggleston
> Assignee: Blake Eggleston
> Priority: Minor
> Fix For: 3.0.16, 3.11.2, 4.0
>
>
> As discussed on dev@, we should give operators the option to disable
> materialized view creation, and log warnings when they're created.
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
---------------------------------------------------------------------
To unsubscribe, e-mail: commits-unsubscr...@cassandra.apache.org
For additional commands, e-mail: commits-h...@cassandra.apache.org
