From: Samuel Just <sam.just@inktank.com>
Date: Wed, 1 May 2013 21:59:08 +0000 (-0700)
Subject: OSD: load_pgs() should fill in start_split honestly
X-Git-Tag: v0.61~19^2
X-Git-Url: http://git-server-git.apps.pok.os.sepia.ceph.com/?a=commitdiff_plain;h=f4982268f76c44d7d258bf4702437d4ee3c8e11d;p=ceph.git

OSD: load_pgs() should fill in start_split honestly

In load_pgs(), we previously called assigned children starting
at the loaded pg created between its stored epoch and the current
osdmap to have that pg as their parent.  This is not correct, some
of the children may have been split in subsequent epochs from children
split in earlier epochs.  Instead, do each map individually.

Signed-off-by: Samuel Just <sam.just@inktank.com>
---

diff --git a/src/osd/OSD.cc b/src/osd/OSD.cc
index d4b75d88c10c..e25c472d7828 100644
--- a/src/osd/OSD.cc
+++ b/src/osd/OSD.cc
@@ -1744,12 +1744,32 @@ void OSD::load_pgs()
       assert(i->second.empty());
     }
 
-    set<pg_t> split_pgs;
+    // First, check whether we can avoid this potentially expensive check
     if (osdmap->have_pg_pool(pg->info.pgid.pool()) &&
-	pg->info.pgid.is_split(pg->get_osdmap()->get_pg_num(pg->info.pgid.pool()),
-			       osdmap->get_pg_num(pg->info.pgid.pool()),
-			       &split_pgs)) {
-      service.start_split(pg->info.pgid, split_pgs);
+	pg->info.pgid.is_split(
+	  pg->get_osdmap()->get_pg_num(pg->info.pgid.pool()),
+	  osdmap->get_pg_num(pg->info.pgid.pool()),
+	  0)) {
+      // Ok, a split happened, so we need to walk the osdmaps
+      set<pg_t> new_pgs; // pgs to scan on each map
+      new_pgs.insert(pg->info.pgid);
+      for (epoch_t e = pg->get_osdmap()->get_epoch() + 1;
+	   e <= osdmap->get_epoch();
+	   ++e) {
+	OSDMapRef curmap(get_map(e-1));
+	OSDMapRef nextmap(get_map(e));
+	set<pg_t> even_newer_pgs; // pgs added in this loop
+	for (set<pg_t>::iterator i = new_pgs.begin(); i != new_pgs.end(); ++i) {
+	  set<pg_t> split_pgs;
+	  if (i->is_split(curmap->get_pg_num(i->pool()),
+			  nextmap->get_pg_num(i->pool()),
+			  &split_pgs)) {
+	    service.start_split(*i, split_pgs);
+	    even_newer_pgs.insert(split_pgs.begin(), split_pgs.end());
+	  }
+	}
+	new_pgs.insert(even_newer_pgs.begin(), even_newer_pgs.end());
+      }
     }
 
     pg->reg_next_scrub();