Switch to a maintained RDS broker

[jmcarp]

There are several forks of both the 18f rds broker (https://github.com/cloudfoundry-community/rds-broker, https://github.com/18F/aws-broker) and the pivotal rds broker (https://github.com/cloudfoundry-community/pe-rds-broker, https://github.com/alphagov/paas-rds-broker), most of which are effectively abandoned--as far as I can tell, the only actively maintained broker is https://github.com/alphagov/paas-rds-broker. I think we should either allocate time to maintain the broker that we currently use or switch to (and contribute to) a broker that is already maintained, probably the alphagov broker.

[afeld]

From GDS in Slack:

We are currently using the broker you linked to (https://github.com/alphagov/paas-rds-broker), we are planning on sticking with it.

[afeld]

Therefore, I say we focus on the code in that repository, and potentially try to merge their changes upstream to @cloudfoundry-community. We should also try and consolidate/deprecate all of the various RDS broker repositories we've created, to reduce confusion.

[jmcarp]

We can deploy the alphagov rds broker from https://github.com/18f/cg-deploy-rds-broker. There are a few PRs in flight that I think we should get merged before offering this broker: https://github.com/alphagov/paas-rds-broker/pull/27 https://github.com/alphagov/paas-rds-broker/pull/29 https://github.com/alphagov/paas-rds-broker/pull/31. The alphagov folks have been very helpful so far, so hopefully it won't take long to get these PRs in shape and merged.

The alphagov broker doesn't include a shared plan. I was talking about this with @afeld and @jcscottiii, and we might be better off replacing the shared plan with a really cheap rds plan. Some motivations: it's hard to persuade users not to used the shared plan for production services; we aren't doing any monitoring around shared plans in case performance degrades or disks fill up; https://github.com/18F/aws-broker/issues/25. WDYT @mogul @dlapiduz ?

[jmcarp]

Update: I fleshed out the read replica PR against upstream. Now we need to either wait for upstream to review our PRs or merge them together in our own fork. WDYT @mogul? Also, bump on my question from above: can we jettison shared plans?

[jmcarp]

Marking blocked until we decide on those questions.

[mogul]

Set a date of the 8th to decide what to do and proceed

[mogul]

We decided to fork the GDS broker into the CF community and work on it there.

[mogul]

We also decided to delay work on this in favor of unblocking FEC with the fec-infra repository

[mogul]

I'm having a hard time understanding the overall status now. Have we got much left pending that they haven't incorporated?

On Wed, Jan 25, 2017 at 1:43 PM Joshua Carp notifications@github.com wrote:

Marking blocked until we decide on those questions.

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/18F/cg-agent-q/issues/10#issuecomment-275241928, or mute the thread https://github.com/notifications/unsubscribe-auth/AAC6kgUW2cM_pOH3rQPIkiHSqXnLZuBfks5rV8HqgaJpZM4KYNfY .

[cnelson]

Noting here that the Go broker we want to switch to was the cause of our manual snapshots quota almost being met. If we do switch to that broker, we should

1) Make sure our smoke tests are setting skip_final_snapshot to true so we don't create snapshots for each test run

2) Figure out if we should expose this setting to users as we could quickly run into our quota if every brokered database creates a manual snapshot when deleted.