Risk of not doing repair

["olek.stasiak@gmail.com" <ol...@gmail.com>]

Hello,
I'm facing bug https://issues.apache.org/jira/browse/CASSANDRA-6277.
After migration to 2.0.2 I can't perform repair on my cluster (six
nodes). Repair on the biggest CF breaks with error described in Jira.
I know, that probably there is a solution in repository, but it's not
included in any release. I can estimate, that 2.0.3 with this fix will
be released in december. If it's not really neccessary, i would avoid
building unstable version of cass from sources and install it in prod
environ, I would rather use rpm-based distribution to keep system in
consistent state.
So this is my question:  What is the risk for me concerned with not
doing repair for a month, assuming that gc_grace is 10days? Should I
really worry? Maybe I should use repo version of cass?
best regards
Olek

Re: Risk of not doing repair

[Robert Coli <rc...@eventbrite.com>]

On Thu, Nov 14, 2013 at 6:25 AM, olek.stasiak@gmail.com <
olek.stasiak@gmail.com> wrote:

> After migration to 2.0.2 I can't perform repair on my cluster (six
> nodes).

...

>  If it's not really neccessary, i would avoid
> building unstable version of cass from sources and install it in prod
> environ


You've already installed an unstable version of cassandra in prod, moving
up to an unreleased version is unlikely to make things that much less
stable.


> So this is my question:  What is the risk for me concerned with not
> doing repair for a month, assuming that gc_grace is 10days? Should I
> really worry? Maybe I should use repo version of cass?
>

Do you do delete or CQL3-delete like operations?

If so, you have a risk of exposure to zombie data.

You should probably increase your gc_grace_seconds to 34 days anyway, so
why not use this experience as an opportunity to do so?

https://issues.apache.org/jira/browse/CASSANDRA-5850

=Rob