Closed
Description
- [Bug] Interplay of Cryostat discovery ping period and harvester period results in reset intervals and timing skew #90 : timing skew can become total failure to upload periodic recordings when the discovery ping period is shorter than the upload period
- discovery ping signal causes Agent to deregister and re-register. The deregistration is handled internally as a full deregistration as if the Agent is shutting down. This cancels the periodic archive timer (which starts anew on re-registration) and also seems to prevent onexit uploads
This can probably be resolved by adding additional internal state to the Agent. Currently the lifecycle can go from REGISTERED (and PUBLISHED) to DEREGISTERED, and back again. Adding state transition chains like REGISTERED -> REFRESHING -> REGISTERED (on success) and REGISTERED - > REFRESHING -> DEREGISTERED (on failure) should allow more nuanced handling of the discovery ping signal so that these bad effects can be avoided until distinctly necessary.