Hi,
4830f1024325 made VACUUM's failsafe abandon its buffer access strategy
and use normal shared-buffer replacement. Commit 9256822608f3 later
converted VACUUM's first phase to use streaming read I/O.
Was it an oversight in the read stream conversion that the stream is
initialized with VACUUM's strategy, but is not updated when failsafe
clears vacrel->bstrategy? As a result, the active stream continues to
use the VACUUM ring for later heap reads.
This patch allows a read stream's strategy to be changed for future I/O.
VACUUM switches the active stream to a NULL strategy when failsafe
activates, without modifying in-flight I/O operations.
The stream calculates its pin limit and queue capacity when it is
created. It uses the original strategy. This patch deliberately changes
only buffer replacement for future reads; it does not resize the stream.
I am unsure whether a generic strategy setter is the right interface. A
new strategy must accommodate the stream's fixed maximum number of pins.
I tested this manually with a 600000-block unlogged table and a 256 kB
VACUUM buffer usage limit. A second session advanced the XID counter
after VACUUM began scanning, causing the 4 GB periodic check to enter
failsafe mode. On my test instance, without this patch, the table's
main fork retained exactly 32 buffers in pg_buffercache. With this
patch, it retained 30665 buffers after failsafe activation.
---
Regards,
Jingtang
