You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Update the API to fetch on-demand snapshots from the DB of
authorization state, bounding both the frequency with which snapshots
may be taken and also for how long a stale snapshot may be used.
A request will either succeed against a current snapshot (happy
fast path), or will fail. If it fails, the caller is asked to retry
after a given delay, and meanwhile the API will have typically
started to refresh the current snapshot.
If the request continues to fail against a snapshot which as taken
_after_ the issued-at time of the request token, only then is the error
considered permanently failed.
This means that requests are cheap, evaluted only by the agent-api,
but we *also* don't have caching artifacts that could result in false
errors when tasks or collections are being created or changed. Instead, we
incur a minor delay while a sufficiently-recent snapshot is taken.
A final change is that we now prefix match on shard ID templates and
journal name templates, rather than munging both into a more-approximate
prefix match over catalog names.
0 commit comments