mon: cache osd epochs

The monitor may get a series of messages from the OSD that prompt it to
send incremental maps (pg_temp updates, failures, probably more).  Avoid
sending the same incremental maps twice by keeping a cache of what epochs
we think the OSDs have.

This reduces monitor load, especially when the mon is a bit behind and is
getting a stream of delayed messages, and the work associated with sending
the inc maps prevents it from catching up.

Signed-off-by: Sage Weil <sage@inktank.com>
Reviewed-by: Greg Farnum <greg@inktank.com>

diff --git a/src/mon/OSDMonitor.cc b/src/mon/OSDMonitor.cc
index 60e0f2c1b39b08babcc171041594a3d6b176aba0..e8a277a7b01dba338c97eb8b5ed5b0d5ac324759 100644 (file)
--- a/src/mon/OSDMonitor.cc
+++ b/src/mon/OSDMonitor.cc
@@ -156,13 +156,24 @@ void OSDMonitor::update_from_paxos()
   if (!t.empty())
     mon->store->apply_transaction(t);
 
-  // populate down -> out map
-  for (int o = 0; o < osdmap.get_max_osd(); o++)
-    if (osdmap.is_down(o) && osdmap.is_in(o) &&
-       down_pending_out.count(o) == 0) {
-      dout(10) << " adding osd." << o << " to down_pending_out map" << dendl;
-      down_pending_out[o] = ceph_clock_now(g_ceph_context);
+  for (int o = 0; o < osdmap.get_max_osd(); o++) {
+    if (osdmap.is_down(o)) {
+      // invalidate osd_epoch cache
+      osd_epoch.erase(o);
+
+      // populate down -> out map
+      if (osdmap.is_in(o) &&
+         down_pending_out.count(o) == 0) {
+       dout(10) << " adding osd." << o << " to down_pending_out map" << dendl;
+       down_pending_out[o] = ceph_clock_now(g_ceph_context);
+      }
     }
+  }
+  // blow away any osd_epoch items beyond max_osd
+  map<int,epoch_t>::iterator p = osd_epoch.upper_bound(osdmap.get_max_osd());
+  while (p != osd_epoch.end()) {
+    osd_epoch.erase(p++);
+  }
 
   if (mon->is_leader()) {
     // kick pgmon, make sure it's seen the latest map
@@ -1495,7 +1506,21 @@ void OSDMonitor::send_full(PaxosServiceMessage *m)
 void OSDMonitor::send_incremental(PaxosServiceMessage *req, epoch_t first)
 {
   dout(5) << "send_incremental [" << first << ".." << osdmap.get_epoch() << "]"
-         << " to " << req->get_orig_source_inst() << dendl;
+         << " to " << req->get_orig_source_inst()
+         << dendl;
+
+  int osd = -1;
+  if (req->get_source().is_osd()) {
+    osd = req->get_source().num();
+    map<int,epoch_t>::iterator p = osd_epoch.find(osd);
+    if (p != osd_epoch.end()) {
+      dout(10) << " osd." << osd << " should have epoch " << p->second << dendl;
+      first = p->second + 1;
+      if (first > osdmap.get_epoch())
+       return;
+    }
+  }
+
   if (first < get_first_committed()) {
     first = get_first_committed();
     bufferlist bl;
@@ -1511,6 +1536,9 @@ void OSDMonitor::send_incremental(PaxosServiceMessage *req, epoch_t first)
     m->newest_map = osdmap.get_epoch();
     m->maps[first] = bl;
     mon->send_reply(req, m);
+
+    if (osd >= 0)
+      osd_epoch[osd] = osdmap.get_epoch();
     return;
   }
 
@@ -1521,6 +1549,9 @@ void OSDMonitor::send_incremental(PaxosServiceMessage *req, epoch_t first)
   m->oldest_map = get_first_committed();
   m->newest_map = osdmap.get_epoch();
   mon->send_reply(req, m);
+
+  if (osd >= 0)
+    osd_epoch[osd] = last;
 }
 
 void OSDMonitor::send_incremental(epoch_t first, entity_inst_t& dest, bool onetime)

diff --git a/src/mon/OSDMonitor.h b/src/mon/OSDMonitor.h
index 036aed5ffd3729c9fe3cd5a9363418704346b781..0034bb0baca4bbb4ea8e5585f7e54d7b8b4a1501 100644 (file)
--- a/src/mon/OSDMonitor.h
+++ b/src/mon/OSDMonitor.h
@@ -123,6 +123,12 @@ private:
 
   map<int,double> osd_weight;
 
+  /*
+   * cache what epochs we think osds have.  this is purely
+   * optimization to try to avoid sending the same inc maps twice.
+   */
+  map<int,epoch_t> osd_epoch;
+
   void check_failures(utime_t now);
   bool check_failure(utime_t now, int target_osd, failure_info_t& fi);