BgBufferSync(): clarification about reusable_buffers variable - Mailing list pgsql-hackers

I have been reading the source code of the BgWriter, and there is some
code in BgBufferSync() that I don't fully understand.

In BgBufferSync(), we have the following code:

    while (num_to_scan > 0 && reusable_buffers < upcoming_alloc_est)
    {
        int            sync_state = SyncOneBuffer(next_to_clean, true,
                                               wb_context);

        if (++next_to_clean >= NBuffers)
        {
            next_to_clean = 0;
            next_passes++;
        }
        num_to_scan--;

        if (sync_state & BUF_WRITTEN)
        {
            reusable_buffers++;
            if (++num_written >= bgwriter_lru_maxpages)
            {
                PendingBgWriterStats.maxwritten_clean++;
                break;
            }
        }
        else if (sync_state & BUF_REUSABLE)
            reusable_buffers++;
    }


In SyncOneBuffer(), we lock the bufHdr and then check if both the
refcount and usage_count are zero. If so, we mark the return value as
BUF_REUSABLE.
My understanding is that this means that the buffer could be reused
when am empty shared buffer is needed by a backend. However, in the
code above, we seem to track these in the reusable_buffers variable.
But that variable is always incremented when the buffer was written in
SyncOneBuffer() even though that buffer might have a non-zero refcount
or non-zero usage_count.

Initially, I thought this might be a bug, but then realized that the
reusable_buffers variable is used to make the BGWriter more or less
aggressive in reading pages, and now I'm unsure.

I discussed this with some of the PG committers at Microsoft, and we
all agree that the current code is a little confusing and it might
help if some more comments were added.

Does anybody know what the reusable_buffers variable really accounts
for, and whether the code as currently written is correct?

Thanks,

Marcel


-- 
Marcel van der Holst
mvdholst@gmail.com