Repair schedules for new clusters

[Ashic Mahtab <as...@live.com>]

Hi All,My previous cassandra clusters had moderate loads, and I'd simply schedule full repairs at different times in the week (but on the same day). That seemed to work ok, but was redundant. In my current project, I'm going to need to care about repair times a lot more, and was wondering what would be the best way to go about it. I have a few questions around this:
* This would be a brand new cluster, and as such, was wondering if I could simply turn on incremental repair from the get go.* I would then run nodetool repair -pr -par -inc once a week on every node at (roughly) the same time once a week. I'd do this with a cron job / external scheduler.* If I were to replace a node, or one rejoins after being absent for longer than the grace period, I'd run a full repair on that node.
Does this sound reasonable? Are there any pitfalls I should be aware of?
Thanks,Ashic. 		 	   		  

Re: Repair schedules for new clusters

[Ben Slater <be...@instaclustr.com>]

We’ve found with incremental repairs that more frequent repairs are
generally better. Our current standard for incremental repairs is once per
day. I imagine that the exact optimum frequency is dependant on the ratio
of reads to write in your cluster.

Turning on incremental repairs from the get-go works OK if your data load
is increment. If you do a big load before your first incremental repair
then it’s not much different to migrating to incremental repairs so worth
following the procedures for migration to avoid a big impact.

Cheers
Ben

On Tue, 17 May 2016 at 16:50 Ashic Mahtab <as...@live.com> wrote:

> Hi All,
> My previous cassandra clusters had moderate loads, and I'd simply schedule
> full repairs at different times in the week (but on the same day). That
> seemed to work ok, but was redundant. In my current project, I'm going to
> need to care about repair times a lot more, and was wondering what would be
> the best way to go about it. I have a few questions around this:
>
> * This would be a brand new cluster, and as such, was wondering if I could
> simply turn on incremental repair from the get go.
> * I would then run nodetool repair -pr -par -inc once a week on every node
> at (roughly) the same time once a week. I'd do this with a cron job /
> external scheduler.
> * If I were to replace a node, or one rejoins after being absent for
> longer than the grace period, I'd run a full repair on that node.
>
> Does this sound reasonable? Are there any pitfalls I should be aware of?
>
> Thanks,
> Ashic.
>
-- 
————————
Ben Slater
Chief Product Officer, Instaclustr
+61 437 929 798