[
https://issues.apache.org/jira/browse/CASSANDRA-8185?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14184951#comment-14184951
]
Stuart Gunter commented on CASSANDRA-8185:
------------------------------------------
Thanks for your response, Aleksey.
However, I disagree that this is the same issue. Having a TTL that is set on
all columns of a row during an UPDATE is not what I'm concerned about in this
issue. Regardless of whether it was intentional or not, the semantics of the
UPDATE statement did change from 2.0.5 to 2.0.6 and has a breaking effect on
applications. I don't think this is something that should happen in a patch
release.
Conceptually, the changes meant that {{SET(X) - X = null}} rather than {{SET(X)
- X = the empty set}}. This change is inconsistent with what the programmer
expects, and rightfully so.
The implication of this change is that the primary key has no inherent value
and is merely a pointer to some other valuable data (contained in other columns
of the row). Here's an example:
Let's consider a have a web crawler that tags URLs with a set of values. In
2.0.5, it would be possible for me to have a row in my table that contains a
URL without any tags. When I upgrade to 2.0.6, this is no longer possible.
Unless a URL has _at least one_ tag, I cannot even store it (proven by another
test added to the repo in the description). This completely de-values the data
within the PK and relegates it to nothing more than a meaningless pointer. But
in this scenario, the URL itself _is a value_. Data that would be retained in
2.0.5 will be lost in 2.0.6!
I understand that you see similarities between this issue and CASSANDRA-6782,
but I'm concerned that they're not the same thing and that this should be
investigated further. Would it be possible to get another opinion on this?
> Removing the last item from a collection removes the entire row
> ---------------------------------------------------------------
>
> Key: CASSANDRA-8185
> URL: https://issues.apache.org/jira/browse/CASSANDRA-8185
> Project: Cassandra
> Issue Type: Bug
> Components: Core
> Environment: Mac OS X Yosemite
> JDK 8u25
> Reporter: Stuart Gunter
> Priority: Minor
>
> When removing the last item from a list or set (and I assume the same applies
> to maps), the entire row is deleted rather than just that last item.
> I've only tested this in cases where the collection column is the only non-PK
> column, but it is definitely a change between v2.0.5 and v2.0.6. I've looked
> through the 2.0.6 release notes and issues and can't find any item that might
> describe this, so I assume it's not an intended change. If such an issue
> exists, I've probably just missed it, so apologies in advance.
> I've created a very simple project to reproduce the issue. If you clone the
> repo and run it in two different modes (the README explains how to do this
> with preconfigured Maven profiles), you will see the tests that pass on
> v2.0.5 and fail on v2.0.6.
> Repo: https://github.com/stuartgunter/cassandra-2.0.6-bug
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
