Feature request: Support reads to existing shapes even when connection to Postgres is lost
We could read existing shapes even when connection to Postgres is lost as we just need to read from disk. Currently it's not supported.
Since #2576 reads to existing shapes will be held until Postgres connectivity is reestablished. Before that, the reads would have errored with "the table identifier does not refer to an existing ETS table" due to the LsnTracker ets table not being available as the ShapeLogCollector is torn down with the other replication processes.
What would be a good semantic for live requests? They are not really live.
Potenitally "live" requests still get held for 20sec and return an "up-to-date" but also mark whether replication is active so the client can decide what to do?
we can have up-to-date and replication sate, or a new control-message. up-to-date is not really the most accurate name.
Alternatively, we can reject the live requests and give a proper status code, so that clients fall back to non-live requests.
Closing this as we have opened plenty of tasks to work towards this goal