mirror of
https://github.com/neondatabase/neon.git
synced 2026-05-16 04:30:38 +00:00
2f0677be26
In #10063 we will switch BlobWriter, which underlies delta and image layer writers, to use the owned buffers IO buffered writer. That buffered writer implements double-buffering by virtue of a background task that performs the flushing -- it owns the VirtualFile and both DeltaLayerWriter and ImageLayerWriter are mere clients to it. The implication is that it's no longer true that dropping these client objects guarantees that all IO activity is complete. We must wait for the flush task to exit. In preparation for that new world, this PR moves the cleanup to a short-lived task that is spawned from the Drop impl, and adds appropriate gate guard holdings to hook it into the Timeline lifecycle. We must (theoretically) worry that there will be a retry inbetween Drop completing and the spawned task completing. It could collide on the randomly generated temporary file name. We avoid this by switching to a global monotonic counter. Refs - extracted from https://github.com/neondatabase/neon/pull/10063 - epic https://github.com/neondatabase/neon/issues/9868