chore: simplify deployment

[emileten]

⚠️ Checklist if your PR is changing anything else than documentation

• The manual deployment workflow ran successfully

• have all the cdk code in one app.py file

• add the vector app

• add a letter representation of datetime.datetime.utcnow() to the test stack ids in case (who knows 🤷‍♂️) several people try to test their PR at the same time. This will fail though if they trigger their workflow at the exact same second, but that's much less likely to happen ! I could use the branch name instead, but then I need to change the workflow again... edit: eventually decided to use the branch name _edit: second edit, branch name has slashes sometimes, which doesn't work with cloudformation, so I decided to use the short commit sha eventually.

[emileten]

workflow running https://github.com/developmentseed/eoapi-cdk/actions/runs/8004317622/job/21861453171

[emileten]

Fun times ! The workflow 'succeeded', but the stack wasn't actually destroyed (I checked on the AWS console). This is the bit that's causing me challenges :

      - name: Tear down any infrastructure
        if: always()
        run: |
          cd integration_tests/cdk
          # run this only if we find a 'cdk.out' directory, which means there might be things to tear down
          if [ -d "cdk.out" ]; then
            cd -
            source .deployment_venv/bin/activate
            cd integration_tests/cdk
            # see https://github.com/aws/aws-cdk/issues/24946
            # this didn't work : rm -f cdk.out/synth.lock
            # so we just duplicate the cdk output to cdk-destroy.out
            npx cdk destroy --output cdk-destroy.out --ci --all --force
          fi

This ran 'fine' in the above workflow. It didn't actually destroy the deployed stack, though :/.

Recap : I want the workflow to be resilient to a sudden cancel in the middle of the deployment ; i.e, I want the destruction to happen properly in these cases. The problem is that in these cases, cdk.out/synth.lock is not released by the deployment step, which probably runs in a docker container. This is a known issue..

People have been saying that removing the file manually rm cdk.out/synth.lock solves this, but it did not for me. So, I instead decided to have the synthesized output of cdk destroy be saved in a different directory, like above :

npx cdk destroy --output cdk-destroy.out --ci --all --force

This is probably why the stack doesn't actually get destroyed...

Edit : actually, it's probably not destroyed because I am using time stamps up the second to name stacks, and since cdk destroy runs a new synth from scratch, it uses a different time stamp and therefore a different stack name ... no wonder it doesn't destroy the initial stack 😆 I am going to use the branch name instead of the time stamp !

[vincentsarago]

🚀

[emileten]

successful workflow : https://github.com/developmentseed/eoapi-cdk/actions/runs/8010980629 merging