Draft: software/rapid-cdn: keep the master partition running until kedifa is ready
Kedifa needs two runs before a new slave has a real key download URL. The first run
publishes an URL that ends with NotReadyYet. Kedifa corrects itself, because its promise
promise-kedifa-auth-ready stays red until every reservation file exists.
The master partition does not correct itself. The master partition reads the URLs from
kedifa one time in each run, and sends the URLs to the frontend. Nothing asks slapgrid to
process the master partition again when kedifa publishes new URLs. If all the promises of
the master partition are green, slapgrid writes the .timestamp file and stops to process
the partition. The frontend then keeps NotReadyYet for a slave that has a key, and the
frontend promise promise-key-download-url-ready stays red.
This merge request adds a promise to the master partition for the slave URLs. The master
partition already has the same promise for its own key, in three places in
instance-master.cfg.in. While the promise is red, slapgrid processes the master partition
again, and the partition sends the correct URLs to the frontend.
Entry 1.0.506 of software/rapid-cdn/CHANGES.rst gives the same reason for the frontend
partition: "The delay left the partition unprocessed wherever no SlapOS Master re-requests
it, so the slave key download URLs never resolved …"
The 2 extra lines in each recorded file list hold the 2 new promise plugins.
Draft for two reasons:
-
a test run must confirm the correction -
the failure rate of test_domain_reuse_after_destructiononmaster, without this promise, is not measured