Huge gap in PLAN_LAYER traces

[trungda]

Hi team,

We are debugging a slow GetPage@LSN request and we are staring at a GET_VECTORED traces. The overall trace looks like this:

--GET_VECTORED (3.39s)
----PLAN_IO (3.39s)
----RECONSTRUCT (negligible)

Under PLAN_IO there are 137 PLAN_LAYER traces (lots of layers to visit which indicates compaction slowness). However, we noticed there is a huge gap (3.3s) between 2 PLAN_LAYER traces. I wonder how should we figure out what's happening between these two traces. We are using sidecar_task for get_vectored_concurrent_io.