Re: [cf-dev] Which components can be HA?

classic Classic list List threaded Threaded
1 message Options
Reply | Threaded
Open this post in threaded view
|

Re: [cf-dev] Which components can be HA?

John Wong
Hi Dieu

Thank you for the answers. They are very helpful.

Regarding #4, you are right, I believe when I do CF deployment I get these short-lived VMs that compile different CF jobs.

Regarding #5, I think it is doppler in our latest deployment (v193, I know still behind the most current version). I think in very old CF version there used to be

loggreator_traffic
loggreator
syslog_loggreator

(as seen in the documentation mentions syslog loggreator).

So we probably don't need to worry about syslog then.


It seems like these are the one we can run >=2
NAT
DEA
UAA
HM9000
CC
Workers
Doppler
Log traffic controller
Gorouter
NFS (use s3 in our case)
Postgres (use RDS in our case)


These are the one not to run with > 1
collector
bosh
clock


Not sure:
stats server (metro agent?)


Thanks.

John

On Wed, May 6, 2015 at 2:27 AM, Dieu Cao <[hidden email]> wrote:
1) I'll ask our doc team to clarify the title of the section. 
It's not recommended to run more than 1 collector.  This component collects metrics from system components.  We use it in combination with Datadog to monitor the many components of cloud foundry.  This component is not strictly required for an HA system. 

2) HM9000 can have multiple active instances. No need for a standby mode.

3) The Cloud Controller clock periodically schedules Cloud Controller clean up tasks for app usage events, audit events, failed jobs, and more. Only single instance of this job is necessary.

4) Likely the job called api_workers is actually the cloud controller workers. These are not compilation vms.
Cloud Controller worker processes background tasks submitted via clients of the api.

5) I'm not sure what you mean by this. Do you mean loggregator? or doppler?

-Dieu
CF Runtime PM

On Tue, May 5, 2015 at 1:19 PM, John Wong <[hidden email]> wrote:
Hi


1) Why is collector listed 1 but in scalable process table?

2) How do you run a second Health Manager in standby mode if only 1 can run at any time?

3) Do we still need clock job? Is it also 1 instance?

4) I notice I have a job called api_workers, and I believe that's compilation machine. I run two of these 24x7, is that necessary? The doc said it is active if we need to compile things (say deploying a new release). Is that all? I don't think they handle application code compilation.

5) What about syslog? Can it have 2? I understand we have to choose what to be HA or not...I am not sure "the BOSH resurrector will recover the VM if it becomes non-responsive" convinces me because all of these jobs are deployed with BOSH but if BOSH is down I am facing some outage. I know Dr. Nic has some article regarding HA bosh.


Correct me if I am wrong.

Thanks.

John

_______________________________________________
cf-dev mailing list
[hidden email]
https://lists.cloudfoundry.org/mailman/listinfo/cf-dev




_______________________________________________
cf-dev mailing list
[hidden email]
https://lists.cloudfoundry.org/mailman/listinfo/cf-dev