Bosh scale up process


Sundarajan Srinivasan
 

I find that bosh restarts all existing process while increasing the number of instances in a job. Do we have any approach so to achieve 0% down time while doing a scale up/down job ?

Thanks
Sundar


Ronak Banka
 

Hi Sundar,

If you are referring to cf release jobs than they already have drain
scripts which take care of clean shutdown and restart . so if you have
sufficient amount of instances running , to keep components in HA then
there shouldn't be any issue.

Thanks
Ronak

On Thu, Aug 18, 2016 at 4:57 PM, Sundarajan Srinivasan <
sundarajan.s(a)gmail.com> wrote:

I find that bosh restarts all existing process while increasing the number
of instances in a job. Do we have any approach so to achieve 0% down time
while doing a scale up/down job ?

Thanks
Sundar


Sundarajan Srinivasan
 

Thanks Ronak.

I didn't mean cf releases. I am deploying databases in bosh. So when i try to increase the instance number in a deployment yaml file and redeploy it, bosh restarts the instances and does canary update.

I am trying to understand if there will be set of procedure so that i will avoid this restarts during canary update.

Thanks
Sundar


Fabio.Berchtold@...
 

Hi Sundar,

I think it depends on how the boshrelease you are using is set up.
If you have to change anything in the manifest / specs, then bosh will update all the old instances too because of these changes. Like if your database boshrelease does peer discovery through bosh-links or via IPs that are defined as spec.
If on the other hand you have a boshrelease where the peer discovery is done through some external means like etcd or consul, and no manifest / specs changes happen when you just increase the number of instances, then it will not update the existing vm instances, since it knows there is nothing to do.
I just confirmed it with our company-internal postgres boshrelease, while adding another slave node bosh did not update/touch the existing instances.

Regards
Fabio