Accepted (related to ADR #0008)
cljdoc uses CircleCI as a sandbox to run analysis on projects as outlined in ADR #0008.
The process currenlty works like this:
- cljdoc queues a build
- CircleCI eventually runs the build
- cljdoc is notified via a webhook
This approach is nice because we don't need to constantly check the status of a build until it eventually finishes but it also has a few drawbacks:
- Testing anything involving webhooks in a local development environment is a serious pain
- Code related to running the analysis and storing it's result are spread accross multiple places making the code harder to follow and maintain
Move to a model where the respective analysis service (CircleCI or Local) exposes a blocking interface for running the analysis.
For CircleCI this blocking interface will simply poll the CircleCI build until
it reached lifecycle "finished".
Unsolved problems applying to both approaches:
Restarting the cljdoc server can cause a loss of state.
- Webhooks may arrive at a time the server is restarted/offline.
- Background processes running analysis may be terminated by restarting the service.
It may take longer until analysis is finished.
- We poll whether the analysis job completed every 5 seconds.
- So in consequence, loading the result of analysis may be delayed by up to 5 seconds from the moment the job finished.