mantle: re-design and fixes

- fix legacy encoding in mds map and add balancer to dumps
- fix blocking rados read and remove temporary files
- fix beacon message spamming

Signed-off-by: Michael Sevilla <mikesevilla3@gmail.com>

doc/cephfs/mantle.rst

@@ -51,12 +51,12 @@ Mantle with `vstart.sh`
 
 ::
 
-    ./vstart.sh -n -l
+    cd build
+    ../src/vstart.sh -n -l
     for i in a b c; do 
-      ./ceph --admin-daemon out/mds.$i.asok config set debug_ms 0
-      ./ceph --admin-daemon out/mds.$i.asok config set debug_mds 0
-      ./ceph --admin-daemon out/mds.$i.asok config set debug_mds_balancer 2
-      ./ceph --admin-daemon out/mds.$i.asok config set mds_beacon_grace 1500
+      bin/ceph --admin-daemon out/mds.$i.asok config set debug_ms 0
+      bin/ceph --admin-daemon out/mds.$i.asok config set debug_mds 2
+      bin/ceph --admin-daemon out/mds.$i.asok config set mds_beacon_grace 1500
     done
 
 
@@ -64,23 +64,23 @@ Mantle with `vstart.sh`
 
 ::
 
-    ./rados put --pool=cephfs_metadata_a greedyspill.lua mds/balancers/greedyspill.lua
+    bin/rados put --pool=cephfs_metadata_a greedyspill.lua ../src/mds/balancers/greedyspill.lua
 
 
 3. Activate Mantle:
 
 ::
 
-    ./ceph mds set allow_multimds true --yes-i-really-mean-it
-    ./ceph mds set max_mds 5
-    ./ceph mds set balancer greedyspill.lua
+    bin/ceph mds set allow_multimds true --yes-i-really-mean-it
+    bin/ceph mds set max_mds 5
+    bin/ceph fs set cephfs_a balancer greedyspill.lua
 
 
 4. Mount CephFS in another window:
 
 ::
 
-     ./ceph-fuse /cephfs -o allow_other &
+     bin/ceph-fuse /cephfs -o allow_other &
      tail -f out/mds.a.log
 
 
@@ -89,8 +89,7 @@ Mantle with `vstart.sh`
    last MDS tries to check the load of its neighbor, which does not exist.
 
 5. Run a simple benchmark. In our case, we use the Docker mdtest image to
-   create load. Assuming that CephFS is mounted in the first container, we can
-   share the mount and run 3 clients using: 
+   create load:
 
 ::
 
@@ -213,6 +212,27 @@ jargon: a dummy value is pushed onto the stack and the next iterator replaces
 the top of the stack with a (k, v) pair. After reading each value, pop that
 value but keep the key for the next call to `lua_next`. 
 
+Reading from RADOS
+~~~~~~~~~~~~~~~~~~
+
+All MDSs will read balancing code from RADOS when the balancer version changes
+in the MDS Map. The balancer pulls the Lua code from RADOS synchronously. We do
+this with a timeout: if the asynchronous read does not come back within half
+the balancing tick interval the operation is cancelled and a Connection Timeout
+error is returned. By default, the balancing tick interval is 10 seconds, so
+Mantle will use a 5 second second timeout. This design allows Mantle to
+immediately return an error if anything RADOS-related goes wrong.
+
+We use this implementation because we do not want to do a blocking OSD read
+from inside the global MDS lock. Doing so would bring down the MDS cluster if
+any of the OSDs are not responsive -- this is tested in the ceph-qa-suite by
+setting all OSDs to down/out and making sure the MDS cluster stays active.
+
+One approach would be to asynchronously fire the read when handling the MDS Map
+and fill in the Lua code in the background. We cannot do this because the MDS
+does not support daemon-local fallbacks and the balancer assumes that all MDSs
+come to the same decision at the same time (e.g., importers, exporters, etc.).
+
 Debugging
 ~~~~~~~~
 
@@ -228,6 +248,10 @@ It is implemented by passing a function that wraps the `dout` logging framework
 (`dout_wrapper`) to Lua with the `lua_register()` primitive. The Lua code is
 actually calling the `dout` function in C++.
 
+Warning and Info messages are centralized using the clog/Beacon. Successful
+messages are only sent on version changes by the first MDS to avoid spamming
+the `ceph -w` utility. These messages are used for the integration tests.
+
 Testing
 ~~~~~~
 

src/mds/MDBalancer.cc

@@ -171,24 +171,45 @@ mds_load_t MDBalancer::get_load(utime_t now)
   return load;
 }
 
-int MDBalancer::localize_balancer(string const balancer)
+/*
+ * Read synchronously from RADOS using a timeout. We cannot do daemon-local
+ * fallbacks (i.e. kick off async read when we are processing the map and
+ * check status when we get here) with the way the mds is structured.
+ */
+int MDBalancer::localize_balancer()
 {
-  int64_t pool_id = mds->mdsmap->get_metadata_pool();
-  string fname = "/tmp/" + balancer;
+  /* reset everything */
+  bool ack = false;
+  int r = 0;
+  bufferlist lua_src;
+  Mutex lock("lock");
+  Cond cond;
 
-  dout(15) << "looking for balancer=" << balancer << " in RADOS pool_id=" << pool_id << dendl;
-  object_t oid = object_t(balancer);
-  object_locator_t oloc(pool_id);
-  bufferlist data;
-  C_SaferCond waiter;
-  mds->objecter->read(oid, oloc, 0, 0, CEPH_NOSNAP, &data, 0, &waiter);
-  int r = waiter.wait();
-  if (r == 0) {
-    dout(15) << "write data from RADOS into fname=" << fname << " data=" << data.c_str() << dendl;
-    data.write_file(fname.c_str());
-  } else {
-    dout(0) << "tick could not find balancer " << balancer
-            << " in RADOS: " << cpp_strerror(r) << dendl;
+  /* we assume that balancer is in the metadata pool */
+  object_t oid = object_t(mds->mdsmap->get_balancer());
+  object_locator_t oloc(mds->mdsmap->get_metadata_pool());
+  ceph_tid_t tid = mds->objecter->read(oid, oloc, 0, 0, CEPH_NOSNAP, &lua_src, 0,
+                                       new C_SafeCond(&lock, &cond, &ack, &r));
+  dout(15) << "launched non-blocking read tid=" << tid
+           << " oid=" << oid << " oloc=" << oloc << dendl;
+
+  /* timeout: if we waste half our time waiting for RADOS, then abort! */
+  double t = ceph_clock_now(g_ceph_context) + g_conf->mds_bal_interval/2;
+  utime_t timeout;
+  timeout.set_from_double(t);
+  lock.Lock();
+  int ret_t = cond.WaitUntil(lock, timeout);
+  lock.Unlock();
+
+  /* success: store the balancer in memory and set the version. */
+  if (!r) {
+    if (ret_t == ETIMEDOUT) {
+      mds->objecter->op_cancel(tid, -ECANCELED);
+      return -ETIMEDOUT;
+    }
+    bal_code.assign(lua_src.to_str());
+    bal_version.assign(oid.name);
+    dout(0) << "localized balancer, bal_code=" << bal_code << dendl;
   }
   return r;
 }
@@ -301,16 +322,14 @@ void MDBalancer::handle_heartbeat(MHeartbeat *m)
       // let's go!
       //export_empties();  // no!
 
-      int r = mantle_prep_rebalance();
-      if (!r) {
-        mds->clog->info() << "mantle succeeded; "
-                          << "balancer=" << mds->mdsmap->get_balancer();
-        return;
+      /* avoid spamming ceph -w if user does not turn mantle on */
+      if (mds->mdsmap->get_balancer() != "") {
+        int r = mantle_prep_rebalance();
+        if (!r) return;
+	mds->clog->warn() << "using old balancer; mantle failed for "
+                          << "balancer=" << mds->mdsmap->get_balancer()
+                          << " : " << cpp_strerror(r);
       }
-
-      mds->clog->warn() << "mantle failed (falling back to original balancer); "
-                        << "balancer=" << mds->mdsmap->get_balancer()
-                        << " : " << cpp_strerror(r);
       prep_rebalance(m->get_beat());
     }
   }
@@ -629,13 +648,16 @@ void MDBalancer::prep_rebalance(int beat)
 
 int MDBalancer::mantle_prep_rebalance()
 {
-  /* pull metadata balancer from RADOS */
-  string const balancer = mds->mdsmap->get_balancer();
-  if (balancer == "" || localize_balancer(balancer))
-    return -ENOENT;
-  ifstream f("/tmp/" + balancer);
-  string script((istreambuf_iterator<char>(f)),
-                 istreambuf_iterator<char>());
+  /* refresh balancer if it has changed */
+  if (bal_version != mds->mdsmap->get_balancer()) {
+    bal_version.assign("");
+    int r = localize_balancer();
+    if (r) return r;
+
+    /* only spam the cluster log from 1 mds on version changes */
+    if (mds->get_nodeid() == 0)
+      mds->clog->info() << "mantle balancer version changed: " << bal_version;
+  }
 
   /* prepare for balancing */
   int cluster_size = mds->get_mds_map()->get_num_in_mds();
@@ -654,17 +676,16 @@ int MDBalancer::mantle_prep_rebalance()
     std::pair < map<mds_rank_t, mds_load_t>::iterator, bool > r(mds_load.insert(val));
     mds_load_t &load(r.first->second);
 
-    metrics[i].insert(make_pair("auth.meta_load", load.auth.meta_load()));
-    metrics[i].insert(make_pair("all.meta_load", load.all.meta_load()));
-    metrics[i].insert(make_pair("req_rate", load.req_rate));
-    metrics[i].insert(make_pair("queue_len", load.queue_len));
-    metrics[i].insert(make_pair("cpu_load_avg", load.cpu_load_avg));
+    metrics[i] = {{"auth.meta_load", load.auth.meta_load()},
+                  {"all.meta_load", load.all.meta_load()},
+                  {"req_rate", load.req_rate},
+                  {"queue_len", load.queue_len},
+                  {"cpu_load_avg", load.cpu_load_avg}};
   }
 
   /* execute the balancer */
-  Mantle *mantle = new Mantle();
-  int ret = mantle->balance(script, mds->get_nodeid(), metrics, my_targets);
-  delete mantle;
+  Mantle mantle;
+  int ret = mantle.balance(bal_code, mds->get_nodeid(), metrics, my_targets);
   dout(2) << " mantle decided that new targets=" << my_targets << dendl;
 
   /* mantle doesn't know about cluster size, so check target len here */

src/mds/MDBalancer.h

@@ -24,6 +24,7 @@ using std::map;
 
 #include "include/types.h"
 #include "common/Clock.h"
+#include "common/Cond.h"
 #include "CInode.h"
 
 
@@ -44,6 +45,8 @@ class MDBalancer {
 
   int last_epoch_under;  
   int last_epoch_over; 
+  string bal_code;
+  string bal_version;
 
   utime_t last_heartbeat;
   utime_t last_fragment;
@@ -88,7 +91,7 @@ public:
 
   int proc_message(Message *m);
   
-  int localize_balancer(string const balancer);
+  int localize_balancer();
   void send_heartbeat();
   void handle_heartbeat(MHeartbeat *m);
 

src/mds/MDSMap.cc

@@ -184,6 +184,7 @@ void MDSMap::dump(Formatter *f) const
   f->dump_int("metadata_pool", metadata_pool);
   f->dump_bool("enabled", enabled);
   f->dump_string("fs_name", fs_name);
+  f->dump_string("balancer", balancer);
 }
 
 void MDSMap::generate_test_instances(list<MDSMap*>& ls)
@@ -226,6 +227,7 @@ void MDSMap::print(ostream& out) const
   out << "data_pools\t" << data_pools << "\n";
   out << "metadata_pool\t" << metadata_pool << "\n";
   out << "inline_data\t" << (inline_data_enabled ? "enabled" : "disabled") << "\n";
+  out << "balancer\t" << balancer << "\n";
 
   multimap< pair<mds_rank_t, unsigned>, mds_gid_t > foo;
   for (const auto &p : mds_info) {
@@ -487,7 +489,6 @@ void MDSMap::encode(bufferlist& bl, uint64_t features) const
     ::encode(session_autoclose, bl);
     ::encode(max_file_size, bl);
     ::encode(max_mds, bl);
-    ::encode(balancer, bl);
     __u32 n = mds_info.size();
     ::encode(n, bl);
     for (map<mds_gid_t, mds_info_t>::const_iterator i = mds_info.begin();
@@ -516,7 +517,6 @@ void MDSMap::encode(bufferlist& bl, uint64_t features) const
     ::encode(session_autoclose, bl);
     ::encode(max_file_size, bl);
     ::encode(max_mds, bl);
-    ::encode(balancer, bl);
     __u32 n = mds_info.size();
     ::encode(n, bl);
     for (map<mds_gid_t, mds_info_t>::const_iterator i = mds_info.begin();
@@ -544,7 +544,7 @@ void MDSMap::encode(bufferlist& bl, uint64_t features) const
     return;
   }
 
-  ENCODE_START(6, 4, bl);
+  ENCODE_START(5, 4, bl);
   ::encode(epoch, bl);
   ::encode(flags, bl);
   ::encode(last_failure, bl);
@@ -553,13 +553,12 @@ void MDSMap::encode(bufferlist& bl, uint64_t features) const
   ::encode(session_autoclose, bl);
   ::encode(max_file_size, bl);
   ::encode(max_mds, bl);
-  ::encode(balancer, bl);
   ::encode(mds_info, bl, features);
   ::encode(data_pools, bl);
   ::encode(cas_pool, bl);
 
   // kclient ignores everything from here
-  __u16 ev = 10;
+  __u16 ev = 11;
   ::encode(ev, bl);
   ::encode(compat, bl);
   ::encode(metadata_pool, bl);
@@ -578,6 +577,7 @@ void MDSMap::encode(bufferlist& bl, uint64_t features) const
   ::encode(enabled, bl);
   ::encode(fs_name, bl);
   ::encode(damaged, bl);
+  ::encode(balancer, bl);
   ENCODE_FINISH(bl);
 }
 
@@ -586,7 +586,7 @@ void MDSMap::decode(bufferlist::iterator& p)
   std::map<mds_rank_t,int32_t> inc;  // Legacy field, parse and drop
 
   cached_up_features = 0;
-  DECODE_START_LEGACY_COMPAT_LEN_16(6, 4, 4, p);
+  DECODE_START_LEGACY_COMPAT_LEN_16(5, 4, 4, p);
   ::decode(epoch, p);
   ::decode(flags, p);
   ::decode(last_failure, p);
@@ -595,7 +595,6 @@ void MDSMap::decode(bufferlist::iterator& p)
   ::decode(session_autoclose, p);
   ::decode(max_file_size, p);
   ::decode(max_mds, p);
-  ::decode(balancer, p);
   ::decode(mds_info, p);
   if (struct_v < 3) {
     __u32 n;
@@ -683,6 +682,10 @@ void MDSMap::decode(bufferlist::iterator& p)
   if (ev >= 9) {
     ::decode(damaged, p);
   }
+
+  if (ev >= 11) {
+  ::decode(balancer, p);
+  }
   DECODE_FINISH(p);
 }
 

src/mds/MDSMap.h

@@ -193,7 +193,7 @@ protected:
    */
 
   mds_rank_t max_mds; /* The maximum number of active MDSes. Also, the maximum rank. */
-  string balancer; /* The name and version of the metadata load balancer. */
+  string balancer;    /* The name/version of the mantle balancer (i.e. the rados obj name) */
 
   std::set<mds_rank_t> in;              // currently defined cluster
 
@@ -290,7 +290,7 @@ public:
   mds_rank_t get_max_mds() const { return max_mds; }
   void set_max_mds(mds_rank_t m) { max_mds = m; }
 
-  std::string get_balancer() const { return balancer; }
+  const std::string get_balancer() const { return balancer; }
   void set_balancer(std::string val) { balancer.assign(val); }
 
   mds_rank_t get_tableserver() const { return tableserver; }

src/mds/Mantle.cc

@@ -1,3 +1,17 @@
+// -*- mode:C++; tab-width:8; c-basic-offset:2; indent-tabs-mode:t -*-
+// vim: ts=8 sw=2 smarttab
+/*
+ * Ceph - scalable distributed file system
+ *
+ * Copyright (C) 2016 Michael Sevilla <mikesevilla3@gmail.com>
+ *
+ * This is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License version 2.1, as published by the Free Software
+ * Foundation.  See file COPYING.
+ *
+ */
+
 #include "mdstypes.h"
 #include "MDSRank.h"
 #include "Mantle.h"

src/mds/Mantle.h

@@ -1,23 +1,28 @@
+// -*- mode:C++; tab-width:8; c-basic-offset:2; indent-tabs-mode:t -*-
+// vim: ts=8 sw=2 smarttab
+/*
+ * Ceph - scalable distributed file system
+ *
+ * Copyright (C) 2016 Michael Sevilla <mikesevilla3@gmail.com>
+ *
+ * This is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License version 2.1, as published by the Free Software
+ * Foundation.  See file COPYING.
+ *
+ */
+
+#ifndef CEPH_MANTLE_H
+#define CEPH_MANTLE_H
+
 #include <lua.hpp>
 #include <list>
 #include <map>
-using std::list;
-using std::map;
 
 #include "include/types.h"
 #include "common/Clock.h"
 #include "CInode.h"
 
-
-
-class MDSRank;
-class Message;
-class MHeartbeat;
-class CInode;
-class CDir;
-class Messenger;
-class MonClient;
-
 class Mantle {
   protected:
     lua_State *L;
@@ -32,3 +37,5 @@ class Mantle {
                 vector < map<string, double> > metrics,
                 map<mds_rank_t,double> &my_targets);
 };
+
+#endif

src/mon/MonCommands.h

@@ -330,7 +330,7 @@ COMMAND_WITH_FLAG("mds set_max_mds " \
 	"set max MDS index", "mds", "rw", "cli,rest", FLAG(DEPRECATED))
 COMMAND_WITH_FLAG("mds set " \
 	"name=var,type=CephChoices,strings=max_mds|max_file_size"
-	"|allow_new_snaps|inline_data|allow_multimds|allow_dirfrags|balancer " \
+	"|allow_new_snaps|inline_data|allow_multimds|allow_dirfrags " \
 	"name=val,type=CephString "					\
 	"name=confirm,type=CephString,req=false",			\
 	"set mds parameter <var> to <val>", "mds", "rw", "cli,rest", FLAG(DEPRECATED))
@@ -398,7 +398,7 @@ COMMAND("fs get name=fs_name,type=CephString", \
 COMMAND("fs set " \
 	"name=fs_name,type=CephString " \
 	"name=var,type=CephChoices,strings=max_mds|max_file_size"
-        "|allow_new_snaps|inline_data|cluster_down|allow_multimds|allow_dirfrags " \
+        "|allow_new_snaps|inline_data|cluster_down|allow_multimds|allow_dirfrags|balancer " \
 	"name=val,type=CephString "					\
 	"name=confirm,type=CephString,req=false",			\
 	"set mds parameter <var> to <val>", "mds", "rw", "cli,rest")