This reverts commit d3bc298f.

since quite some time we do not report the pod as unhealthy any longer
if redis connection is gone. So in case the connection is lost, we want it
to reconnect.

Make sure to always fetch from redis once client (re-)connected.

julian.baeume authored 1 year ago and julian.baeume committed 1 year ago

Changelog:
- introduce two separate roles for middleware deployments:
    - ui-middleware: read from redis only, listen to version updates, can easily be scaled
    - updater: split out all code that writes to redis, no need to scale > 1

- Unit and Integration tests now use "inject" instead of starting a server
  and using supertest.
- Add Coverage Reports and "test:dev" mode. This enables a mocha --watch
  like workflow using nodemon, as --watch does not work with esm modules.
- Reorganized and renamed files and functions

Set cache control to no-store. The ui-middleware will then only look at the remote servers for the files and never in the local cache

richard.petersen authored 2 years ago and julian.baeume committed 2 years ago

Root cause: Due to a bunch of redis restarts, the bull-queues got unregistered and did not register again.
Solution: On every reconnect, readd the bull queue again.

This commit contains a fixup for the original commit 64f2780a

Root cause: If multiple connections are opened at the same time after an update, the ui-middleware will do multiple requests to get initial data. That went especially worse, when e2e-tests are launched against ui-middlewares with multiple clients connecting at the same time.
Soltion: Make sure, that a node already caches stuff without a response.

Root cause: configuration was only reloaded on those nodes, that did the version update. The other nodes might not notice that.
Solution: On propagated version changes, the nodes always check for config changes