mirror of
https://github.com/neondatabase/neon.git
synced 2026-01-15 01:12:56 +00:00
760a48207d
From #6037 on, until this patch, if the client opens the connection but doesn't send a `PagestreamFeMessage` within the first 10ms, we'd close the connection because `self.timeline_cancelled()` returns. It returns because `self.shard_timelines` is still empty at that point: it gets filled lazily within the handlers for the incoming messages. Changes ------- The question is: if we can't check for timeline cancellation, what else do we need to be cancellable for? `tenant.cancel` is also a bad choice because the `tenant` (shard) we pick at the top of handle_pagerequests might indeed go away over the course of the connection lifetime, but other shards may still be there. The correct solution, I think, is to be responsive to task_mgr cancellation, because the connection handler runs in a task_mgr task and it is already the current canonical way how we shut down a tenant's / timelin's page_service connections (see `Tenant::shutdown` / `Timeline::shutdown`). So, rename the function and make it sensitive to task_mgr cancellation.