Feature request: make it feasible to use the Cloud SQL Proxy in batch Jobs and CronJobs [70746902]

Assigned

Feature Request

Status Update

No update yet.

Description

ab...@abevoelker.com

created issue #1

Dec 16, 2017 05:43PM

GKE requires me to use the Cloud SQL Proxy sidecar container to access my Cloud SQL instance. This is a hot mess when it comes to Jobs and CronJobs because the pods do not terminate correctly:

https://github.com/kubernetes/kubernetes/issues/25908

https://github.com/GoogleCloudPlatform/cloudsql-proxy/issues/128

Please make it easier to write Jobs/CronJobs that need access to Cloud SQL.

Alternatively how about eliminating the need for the sidecar container to access the Cloud SQL instance (because right about now I'm thinking it's simpler to pay for a different managed Postgres solution like ElephantSQL than to write really hacky k8s specs).

Comments

kn...@google.com <kn...@google.com> #2Dec 17, 2017 04:10PM

Assigned to gc...@google.com.

Thank you for submitting this feature request. I already notified our engineering team about this case. Future update(s) will be posted here.

na...@gmail.com <na...@gmail.com> #3Jan 9, 2018 02:34PM

Hi. Do you have any updates on this issue? We're moving from AWS to GCP and we stumbled upon this problem. Thanks.

Message last modified on Jan 9, 2018 02:34PM

[Deleted User] <[Deleted User]> #4Jan 23, 2018 11:35PM

Hi, same issue here. Any updates on this ?

[Deleted User] <[Deleted User]> #5Jan 30, 2018 07:18AM

We're also moving away from AWS to GCP and stumbled upon this problem. Please help to look into this. Thanks!

ma...@soundtracktor.com <ma...@soundtracktor.com> #6Feb 15, 2018 04:49PM

Same issue here

[Deleted User] <[Deleted User]> #7Feb 15, 2018 08:28PM

Facing the same issue

rs...@parkerici.org <rs...@parkerici.org> #8Feb 28, 2018 10:39PM

Facing the same issue here too.

[Deleted User] <[Deleted User]> #9Mar 23, 2018 05:58PM

For what it's worth you don't have to run the cloudsql in a sidecar format, it works fine as a separate deploy at least for postgresql. It will bind to external ports and you can expose it as a service in k8s. I've never encountered throughput or connection issues.

We actually run it along with pgbouncer in a pod as a deployment with anti-affinity and just have all our apps connect to either depending on if they need prepared statements or not.

je...@greensync.com.au <je...@greensync.com.au> #10Apr 9, 2018 03:48AM

I have a helm chart that deploys everything needed for the app and I was hoping to run migrations as a job with a 'pre-install' hook that could migrate the DB before rest of the deployment.

Having to install cloudsql proxies as a seperate deployment doesn't seem ideal..

I could download the cloudsql-proxy libraries and start the proxy manually.

to...@baucloud.com <to...@baucloud.com> #11Apr 9, 2018 04:17AM

I am a bit surprised, as Google advertises the whole sidecar pattern (especially in context with Cloud SQL) even on conferences (e.g. Google Cloud Summit in Munich 2017)

And then one realizes, that it has significant drawbacks...

[Deleted User] <[Deleted User]> #12Oct 22, 2018 10:50AM

How is still not resolved? Google Support still recommend the sidecar sqlproxy, but this does not work well for jobs and does not terminate correctly,
only solution would be to use CloudSQL Private IP's support once it's stable.

za...@bzdyl.net <za...@bzdyl.net> #13Dec 18, 2018 12:03PM

CloudSQL now supports private IPs:

https://cloud.google.com/sql/docs/mysql/private-ip. With this it's possible to create a VPC peering between CloudSQL instance VPC and your GKE cluster network so your pods can connect to CloudSQL instance directly using the private IP.

ab...@abevoelker.com <ab...@abevoelker.com> #14Jan 14, 2019 05:04PM

Yep I've been using private networking since November and it's been great. I don't see a way to close my own issue but I believe private networking / VPC has solved my issue.

[Deleted User] <[Deleted User]> #15Mar 4, 2019 09:53AM

This is also an issue for Jobs/Pods that use Istio Proxy injection. We're encountering this issue when running Helm Tests (uses Pods that run to completion) and with batch Jobs using Cloud SQL Proxy and Istio Proxy.