GitLab

/cc @tomo @romain @luke @jerome

I added a test that deploys a nested theia inside theia0, waits until theia1 is synced then checks that during processing theia0 did forward the frontend requests and theia1 did not.

This is a bit delicate because the proxy database is cloned from theia0 to theia1 before the cloned embedded instance is processed, so both versions contain the forwarded requests. Thankfully, every time a row is inserted the rowid changes, so by checking the rowids are identical we can be sure no forwarded frontend requests were inserted in the database after it was cloned.

Using the rowids is so that the test will still work once slapos.core!423 (merged) is merged. Until then the rowids are actually unnecessary because the history of all insertions is (unintentionally) preserved, so if theia1 did forward requests there would be more rows in the cloned database.

Note that inspecting the "external master", i.e. the proxy of the test node, is unpractical: since that proxy does not forward frontend requests, it falls back to faking the frontend allocation, and this does not leave a trace in the proxy database either way. The only sign that this happened is in the logs, so this makes it unpractical to determine whether theia1 did or did not forward frontend requests.