Garry Hill
Garry Hill
Our current "shape_count" metric is just total defined shapes. This needs to be divided by consumer status so we'd have - defined: total shapes in status table - running: shapes...
We don't necessarily need to launch a corresponding consumer process for a newly defined shape. We can snapshot and then lazily launch the consumer on first transaction, as for previously...
Consumer processes should instead be started by the first transaction that includes their shape. New shapes should still launch consumer processes as now (cf #2976)
If we update LRU culling to only include online shapes, then we also need some kind of culling of offline shapes. This needs to delete the shape, including the data,...
If we're managing server load then we need to take into account only running shapes when deciding which to cull. Should we delete the shape as now or merely force...
When we eventually transition to unloading inactive shape consumers, we'll need to know if a consumer is inactive and for how long. With a transition from active -> hibernated ->...
- Subqueries aren't currently restored from disk on startup anyway, so there's no need to think about lazily loading them. - Need to ensure that subqueries touching existing but not...
Currently it is the termination of a consumer process that triggers removal of a shape from the ShapeLogCollector filters. In a world where there may not be a consumer for...
Since our stack is taking a long time to restart we get into a loop. Taking the stack down and bringing it back takes > 1min so we end up...
**Versions** - Electric: main **Bug description** If you configure a db with host `127.0.0.1` the connection resolver treats it as a hostname and tries to connect over ipv6. **Expected behavior**...