roots: filter out obsolete extents from extent refs

When an extent ref is modified, all of the refs in the same metadata
page get the same transid in the TREE_SEARCH_V2 header.  All of the
extents are rescanned by later subvol scans.  This causes up to 80%
overhead due to redundant reads of the same extents.

A proper fix for this requires extent-based scanning instead of
extent-ref-based scanning.  Until that happens, filter out new references
to old extents.

Signed-off-by: Zygo Blaxell <bees@furryterror.org>

src/bees-roots.cc

@@ -742,13 +742,24 @@ BeesCrawl::fetch_extents()
 		if (gen < get_state().m_min_transid) {
 			BEESCOUNT(crawl_gen_low);
 			++count_low;
-			// We probably want (need?) to scan these anyway.
-			// continue;
+			// We want (need?) to scan these anyway?
+			// The header generation refers to the transid
+			// of the metadata page holding the current ref.
+			// This includes anything else in that page that
+			// happened to be modified, regardless of how
+			// old it is.
+			// The file_extent_generation refers to the
+			// transid of the extent item's page, which is
+			// a different approximation of what we want.
+			// Combine both of these filters to minimize
+			// the number of times we unnecessarily re-read
+			// an extent.
+			continue;
 		}
 		if (gen > get_state().m_max_transid) {
 			BEESCOUNT(crawl_gen_high);
 			++count_high;
-			// This shouldn't ever happen
+			// This shouldn't ever happen...and so far, doesn't.
 			// continue;
 		}