Investigate the response delay when scaling down to a new pod

[heruan]

When scaling the app down, new requests are redirected to a new pod and the first request takes a noticeable time to be server (5-7 seconds). We need to investigate why this happens: is it because of how the load-balancing is configured or because that is the first request to the Flow instance on the new pod and it requires some time to be initialized?

[mcollovati]

One of the culprits of slowness seems to be PwaRegistry initialization, specifically the icon's generation loop

com.vaadin.flow.server.PwaRegistry       : PwaRegistry initialization took 4480ms

I wonder if Flow can build icons during production build 🤔

[mcollovati]

Flow related issue
vaadin/flow#14606

[mcollovati]

For an application with PWA enabled, adding a http readiness probe pointing at /sw-runtime-resources-precache.js could help to work around the issue.