osd: augment osd heartbeat peers with neighbors and randoms, to up some min

- always include our neighbors to ensure we have a fully-connected
  graph
- include some random neighbors to get at least some min number of peers.

Signed-off-by: Sage Weil <sage@inktank.com>

diff --git a/src/common/config_opts.h b/src/common/config_opts.h
index 56a46aec0c0c20e20d0e16f73094a6cd015d6f2a..01e8b5a5a15d462ffcdc3f8363d40c8bfe7ad12c 100644 (file)
--- a/src/common/config_opts.h
+++ b/src/common/config_opts.h
@@ -408,6 +408,7 @@ OPTION(osd_age_time, OPT_INT, 0)
 OPTION(osd_heartbeat_addr, OPT_ADDR, entity_addr_t())
 OPTION(osd_heartbeat_interval, OPT_INT, 6)       // (seconds) how often we ping peers
 OPTION(osd_heartbeat_grace, OPT_INT, 20)         // (seconds) how long before we decide a peer has failed
+OPTION(osd_heartbeat_min_peers, OPT_INT, 10)     // minimum number of peers
 OPTION(osd_mon_heartbeat_interval, OPT_INT, 30)  // (seconds) how often to ping monitor if no peers
 OPTION(osd_mon_report_interval_max, OPT_INT, 120)
 OPTION(osd_mon_report_interval_min, OPT_INT, 5)  // pg stats, failures, up_thru, boot.

diff --git a/src/osd/OSD.cc b/src/osd/OSD.cc
index 4bbcd06243e9d9114483a7316e364bd674a3fe6f..cf08418262b92f2d58ac27f0594ebec25c761caa 100644 (file)
--- a/src/osd/OSD.cc
+++ b/src/osd/OSD.cc
@@ -2292,6 +2292,8 @@ void OSD::maybe_update_heartbeat_peers()
     return;
   heartbeat_need_update = false;
 
+  dout(10) << "maybe_update_heartbeat_peers updating" << dendl;
+
   heartbeat_epoch = osdmap->get_epoch();
 
   // build heartbeat from set
@@ -2314,25 +2316,66 @@ void OSD::maybe_update_heartbeat_peers()
     pg->heartbeat_peer_lock.Unlock();
   }
 
-  map<int,HeartbeatInfo>::iterator p = heartbeat_peers.begin();
-  while (p != heartbeat_peers.end()) {
+  // include next and previous up osds to ensure we have a fully-connected set
+  set<int> want, extras;
+  int next = osdmap->get_next_up_osd_after(whoami);
+  if (next >= 0)
+    want.insert(next);
+  int prev = osdmap->get_previous_up_osd_before(whoami);
+  if (prev >= 0)
+    want.insert(prev);
+
+  for (set<int>::iterator p = want.begin(); p != want.end(); ++p) {
+    dout(10) << " adding neighbor peer osd." << *p << dendl;
+    extras.insert(*p);
+    _add_heartbeat_peer(*p);
+  }
+
+  // identify extras
+  for (map<int,HeartbeatInfo>::iterator p = heartbeat_peers.begin();
+       p != heartbeat_peers.end();
+       ++p) {
     if (p->second.epoch < osdmap->get_epoch()) {
-      dout(20) << " removing heartbeat peer osd." << p->first
-              << " " << p->second.con_back->get_peer_addr()
-              << " " << (p->second.con_front ? p->second.con_front->get_peer_addr() : entity_addr_t())
-              << dendl;
-      hbclient_messenger->mark_down(p->second.con_back);
-      p->second.con_back->put();
-      if (p->second.con_front) {
-       hbclient_messenger->mark_down(p->second.con_front);
-       p->second.con_front->put();
-      }
-      heartbeat_peers.erase(p++);
-    } else {
-      ++p;
+      extras.insert(p->first);
+    }
+  }
+
+  // too few?
+  int start = osdmap->get_next_up_osd_after(whoami);
+  for (int n = start; n >= 0; ) {
+    if ((int)heartbeat_peers.size() >= g_conf->osd_heartbeat_min_peers)
+      break;
+    if (!extras.count(n) && !want.count(n) && n != whoami) {
+      dout(10) << " adding random peer osd." << n << dendl;
+      extras.insert(n);
+      _add_heartbeat_peer(n);
+    }
+    n = osdmap->get_next_up_osd_after(n);
+    if (n == start)
+      break;  // came full circle; stop
+  }
+
+  // too many?
+  for (set<int>::iterator p = extras.begin();
+       (int)heartbeat_peers.size() > g_conf->osd_heartbeat_min_peers && p != extras.end();
+       ++p) {
+    if (want.count(*p))
+      continue;
+    map<int,HeartbeatInfo>::iterator q = heartbeat_peers.find(*p);
+    dout(20) << " removing heartbeat peer osd." << q->first
+            << " " << q->second.con_back->get_peer_addr()
+            << " " << (q->second.con_front ? q->second.con_front->get_peer_addr() : entity_addr_t())
+            << dendl;
+    hbclient_messenger->mark_down(q->second.con_back);
+    q->second.con_back->put();
+    if (q->second.con_front) {
+      hbclient_messenger->mark_down(q->second.con_front);
+      q->second.con_front->put();
     }
+    heartbeat_peers.erase(q);
   }
-  dout(10) << "maybe_update_heartbeat_peers " << heartbeat_peers.size() << " peers" << dendl;
+
+  dout(10) << "maybe_update_heartbeat_peers " << heartbeat_peers.size() << " peers, extras " << extras << dendl;
 }
 
 void OSD::reset_heartbeat_peers()

diff --git a/src/osd/OSDMap.h b/src/osd/OSDMap.h
index deebc376a916c73bfbd25ee655355b3a435cfb93..0a00c40e23b3e16861c6dc053ec77fb935805988 100644 (file)
--- a/src/osd/OSDMap.h
+++ b/src/osd/OSDMap.h
@@ -404,6 +404,30 @@ private:
     return -1;
   }
 
+  int get_next_up_osd_after(int n) const {
+    for (int i = n + 1; i != n; ++i) {
+      if (i >= get_max_osd())
+       i = 0;
+      if (i == n)
+       break;
+      if (is_up(i))
+       return i;
+    }
+    return -1;
+  }
+
+  int get_previous_up_osd_before(int n) const {
+    for (int i = n - 1; i != n; --i) {
+      if (i < 0)
+       i = get_max_osd() - 1;
+      if (i == n)
+       break;
+      if (is_up(i))
+       return i;
+    }
+    return -1;
+  }
+
   /**
    * get feature bits required by the current structure
    *