I have a query with respect to both SST and SEED, ...
# seeda
Ashishkumar Pandey
06/23/2021, 1:37 AMI have a query with respect to both SST and SEED, how should I go about running CLI commands pre and post deploy that use env vars from my codebase. Also, the post deploy commands need to be run based on the state of the deployment. I need this to basically run mongodb migrations pre deployment and rollback if the deployment fails and do nothing if the deployment succeeds.
f
Frank
06/23/2021, 3:29 AM
hmm.. post deploy doesn’t get triggered if the deploy fails.
a
Ashishkumar Pandey
06/23/2021, 3:29 AMso how would I go about rolling back my migration if a deployment fails?
f
Frank
06/23/2021, 3:34 AM
I can see there’s value in having a way to run something on deploy failure to clean stuff up.
Frank
06/23/2021, 3:35 AM
But I think migration rollbacks isn’t a such use case.
Frank
06/23/2021, 3:36 AM
Usually you should separate data migration from deployment, since the migration and deployment aren’t done atomicly.
Frank
06/23/2021, 3:37 AM
So either:
1. you migrate the data first, and ensure existing code works with the new data, and then deploy; or
2. you deploy first. After the deploy succeeds, you then migrate the data.
Frank
06/23/2021, 3:38 AM
I haven’t come across a use case where the two are tightly coupled.
a
Ashishkumar Pandey
06/23/2021, 3:39 AMI agree but I really don’t want to have a manual deploy step because missing it is too easy. Usually all monolith deployments have a script that deals with migrations.
f
Frank
06/23/2021, 3:41 AM
Right for sure.
Frank
06/23/2021, 3:43 AM
Here’s a recently discussion on running migrations https://serverless-stack.slack.com/archives/C01JG3B20RY/p1624329822495000
Frank
06/23/2021, 3:43 AM
For now, something you can try is running the data migration inside a custom resource.
a
Ashishkumar Pandey
06/23/2021, 3:43 AMYep, I’m in that thread.
Ashishkumar Pandey
06/23/2021, 3:46 AMOh! Interesting, any links to some docs or if you’ve a guide on how to use this? The prime features I need are to handle deployment failures and have access to env vars. If it’s a cdk construct then env vars shouldn’t be a problem but how would I go about dealing with deployment failures?
f
Frank
06/23/2021, 3:50 AM
I found a guide here on setting up data migration using custom resource - https://dev.to/aws-builders/handling-serverless-aurora-schema-migration-using-custom-resource-with-python-cdk-5fpc
Frank
06/23/2021, 3:51 AM
So when your stack gets updated, the custom resource is going to receive an
UpdateFrank
06/23/2021, 3:52 AM
And CF deployment failed, the custom resource is going to get called again. And based on the input, you can implement the rollback logic.
a
Ashishkumar Pandey
06/23/2021, 3:53 AMokay, thank you. I will go through this but I believe this a very common scenario and should not need effort and r&d to achieve. If you could maybe think of including this as a high-level construct within sst itself.
f
Frank
06/23/2021, 3:53 AM
The example doesn’t have rollback. Again, I haven’t come across that many ppl that does rollback to be honest.
a
Ashishkumar Pandey
06/23/2021, 3:54 AMYeah, I haven’t had to use rollback myself ever but knowing it is possible helps me sleep better. 😅
f
Frank
06/23/2021, 3:57 AM
Totally agreed! I actually just opened an issue for supporting DB migrations https://github.com/serverless-stack/serverless-stack/issues/483
a
Ashishkumar Pandey
06/23/2021, 3:57 AMWow! That’s great, thank you, I appreciate it. 🙏
f
Frank
06/23/2021, 4:08 AM
Just updated it with a proposal on what it might look like :)
a
Ashishkumar Pandey
06/23/2021, 4:09 AMAh! interesting indeed. 🙌