mds: add throttle for trimming MDCache

This is necessary when the MDS cache size decreases by a significant amount.
For example, when stopping a large MDS or when the operator makes a large cache
size reduction.

Fixes: http://tracker.ceph.com/issues/37723
Signed-off-by: Patrick Donnelly <pdonnell@redhat.com>
(cherry picked from commit 7bf2f31abf069b9bafc52bc1478292bb8a6c71b8)

Conflicts:
	PendingReleaseNotes
	src/common/options.cc
	src/mds/MDCache.cc
	src/mds/MDCache.h

diff --git a/PendingReleaseNotes b/PendingReleaseNotes
index 05f7f44ae7ee9cffe97d055a548ea61678f9ed51..88add63e24449280802c664960cb68f91fc4df3a 100644 (file)
--- a/PendingReleaseNotes
+++ b/PendingReleaseNotes
@@ -12,6 +12,10 @@
   validated and providing a keyring with an invalid caps string to, e.g.,
   "ceph auth add" will result in an error.
 
+* The MDS cache trimming is now throttled. Dropping the MDS cache
+  via the `ceph tell mds.<foo> cache drop` command or large reductions in the
+  cache size will no longer cause service unavailability.
+
 >= 12.1.2
 ---------
 * When running 'df' on a CephFS filesystem comprising exactly one data pool,

diff --git a/src/common/options.cc b/src/common/options.cc
index d452fc73faec0f96839aa9d7a045078a780a271c..57468a332088f73ae4847d9a2a1deca52c3f4f91 100644 (file)
--- a/src/common/options.cc
+++ b/src/common/options.cc
@@ -6098,6 +6098,14 @@ std::vector<Option> get_mds_options() {
     .set_default(.7)
     .set_description(""),
 
+    Option("mds_cache_trim_decay_rate", Option::TYPE_FLOAT, Option::LEVEL_ADVANCED)
+    .set_default(1)
+    .set_description("decay rate for trimming MDS cache throttle"),
+
+    Option("mds_cache_trim_threshold", Option::TYPE_UINT, Option::LEVEL_ADVANCED)
+    .set_default(64_K)
+    .set_description("threshold for number of dentries that can be trimmed"),
+
     Option("mds_max_file_recover", Option::TYPE_UINT, Option::LEVEL_ADVANCED)
     .set_default(32)
     .set_description(""),

diff --git a/src/mds/MDCache.cc b/src/mds/MDCache.cc
index 099f28a0bc831f0434863775c5eaa4b04dcddb20..f36ba40002d3f730262c7ff188b576ba3077ec02 100644 (file)
--- a/src/mds/MDCache.cc
+++ b/src/mds/MDCache.cc
@@ -174,7 +174,8 @@ MDCache::MDCache(MDSRank *m, PurgeQueue &purge_queue_) :
   filer(m->objecter, m->finisher),
   exceeded_size_limit(false),
   recovery_queue(m),
-  stray_manager(m, purge_queue_)
+  stray_manager(m, purge_queue_),
+  trim_counter(ceph_clock_now(), g_conf->get_val<double>("mds_cache_trim_decay_rate"))
 {
   migrator.reset(new Migrator(mds, this));
   root = NULL;
@@ -244,6 +245,9 @@ void MDCache::handle_conf_change(const struct md_config_t *conf,
     cache_health_threshold = g_conf->get_val<double>("mds_health_cache_threshold");
   if (changed.count("mds_cache_mid"))
     lru.lru_set_midpoint(g_conf->get_val<double>("mds_cache_mid"));
+  if (changed.count("mds_cache_trim_decay_rate")) {
+    trim_counter = DecayCounter(ceph_clock_now(), g_conf->get_val<double>("mds_cache_trim_decay_rate"));
+  }
 
   migrator->handle_conf_change(conf, changed, mdsmap);
   mds->balancer->handle_conf_change(conf, changed, mdsmap);
@@ -6494,12 +6498,14 @@ void MDCache::start_recovered_truncates()
 // ================================================================================
 // cache trimming
 
-void MDCache::trim_lru(uint64_t count, map<mds_rank_t, MCacheExpire*> &expiremap)
+std::pair<bool, uint64_t> MDCache::trim_lru(uint64_t count, map<mds_rank_t, MCacheExpire*> &expiremap)
 {
   bool is_standby_replay = mds->is_standby_replay();
   std::vector<CDentry *> unexpirables;
   uint64_t trimmed = 0;
 
+  auto trim_threshold = g_conf->get_val<uint64_t>("mds_cache_trim_threshold");
+
   dout(7) << "trim_lru trimming " << count
           << " items from LRU"
           << " size=" << lru.lru_get_size()
@@ -6508,7 +6514,11 @@ void MDCache::trim_lru(uint64_t count, map<mds_rank_t, MCacheExpire*> &expiremap
           << " pinned=" << lru.lru_get_num_pinned()
           << dendl;
 
-  for (;;) {
+  const uint64_t trim_counter_start = trim_counter.get(ceph_clock_now());
+  bool throttled = false;
+  while (1) {
+    throttled |= trim_counter_start+trimmed >= trim_threshold;
+    if (throttled) break;
     CDentry *dn = static_cast<CDentry*>(bottom_lru.lru_expire());
     if (!dn)
       break;
@@ -6525,7 +6535,9 @@ void MDCache::trim_lru(uint64_t count, map<mds_rank_t, MCacheExpire*> &expiremap
   unexpirables.clear();
 
   // trim dentries from the LRU until count is reached
-  while (cache_toofull() || count > 0) {
+  while (!throttled && (cache_toofull() || count > 0)) {
+    throttled |= trim_counter_start+trimmed >= trim_threshold;
+    if (throttled) break;
     CDentry *dn = static_cast<CDentry*>(lru.lru_expire());
     if (!dn) {
       break;
@@ -6540,6 +6552,7 @@ void MDCache::trim_lru(uint64_t count, map<mds_rank_t, MCacheExpire*> &expiremap
       if (count > 0) count--;
     }
   }
+  trim_counter.hit(ceph_clock_now(), trimmed);
 
   for (auto &dn : unexpirables) {
     lru.lru_insert_mid(dn);
@@ -6547,6 +6560,7 @@ void MDCache::trim_lru(uint64_t count, map<mds_rank_t, MCacheExpire*> &expiremap
   unexpirables.clear();
 
   dout(7) << "trim_lru trimmed " << trimmed << " items" << dendl;
+  return std::pair<bool, uint64_t>(throttled, trimmed);
 }
 
 /*
@@ -6555,7 +6569,7 @@ void MDCache::trim_lru(uint64_t count, map<mds_rank_t, MCacheExpire*> &expiremap
  *
  * @param count is number of dentries to try to expire
  */
-bool MDCache::trim(uint64_t count)
+std::pair<bool, uint64_t> MDCache::trim(uint64_t count)
 {
   uint64_t used = cache_size();
   uint64_t limit = cache_memory_limit;
@@ -6569,7 +6583,8 @@ bool MDCache::trim(uint64_t count)
   // process delayed eval_stray()
   stray_manager.advance_delayed();
 
-  trim_lru(count, expiremap);
+  auto result = trim_lru(count, expiremap);
+  auto& trimmed = result.second;
 
   // trim non-auth, non-bound subtrees
   for (auto p = subtrees.begin(); p != subtrees.end();) {
@@ -6585,6 +6600,7 @@ bool MDCache::trim(uint64_t count)
          continue;
 
        migrator->export_empty_import(dir);
+        ++trimmed;
       }
     } else {
       if (!diri->is_auth()) {
@@ -6601,6 +6617,7 @@ bool MDCache::trim(uint64_t count)
            rejoin_ack_gather.count(dir->get_dir_auth().first))
          continue;
        trim_dirfrag(dir, 0, expiremap);
+        ++trimmed;
       }
     }
   }
@@ -6611,11 +6628,15 @@ bool MDCache::trim(uint64_t count)
     root->get_dirfrags(ls);
     for (list<CDir*>::iterator p = ls.begin(); p != ls.end(); ++p) {
       CDir *dir = *p;
-      if (dir->get_num_ref() == 1)  // subtree pin
+      if (dir->get_num_ref() == 1) { // subtree pin
        trim_dirfrag(dir, 0, expiremap);
+        ++trimmed;
+      }
     }
-    if (root->get_num_ref() == 0)
+    if (root->get_num_ref() == 0) {
       trim_inode(0, root, 0, expiremap);
+      ++trimmed;
+    }
   }
 
   std::set<mds_rank_t> stopping;
@@ -6638,11 +6659,15 @@ bool MDCache::trim(uint64_t count)
       list<CDir*> ls;
       mdsdir_in->get_dirfrags(ls);
       for (auto dir : ls) {
-       if (dir->get_num_ref() == 1)  // subtree pin
+       if (dir->get_num_ref() == 1) {  // subtree pin
          trim_dirfrag(dir, dir, expiremap);
+          ++trimmed;
+        }
       }
-      if (mdsdir_in->get_num_ref() == 0)
+      if (mdsdir_in->get_num_ref() == 0) {
        trim_inode(NULL, mdsdir_in, NULL, expiremap);
+        ++trimmed;
+      }
     } else {
       dout(20) << __func__ << ": some unexpirable contents in mdsdir" << dendl;
     }
@@ -6656,6 +6681,7 @@ bool MDCache::trim(uint64_t count)
         dout(20) << __func__ << ": maybe trimming base: " << *(*p) << dendl;
         if ((*p)->get_num_ref() == 0) {
           trim_inode(NULL, *p, NULL, expiremap);
+          ++trimmed;
         }
       }
     }
@@ -6664,7 +6690,7 @@ bool MDCache::trim(uint64_t count)
   // send any expire messages
   send_expire_messages(expiremap);
 
-  return true;
+  return result;
 }
 
 void MDCache::send_expire_messages(map<mds_rank_t, MCacheExpire*>& expiremap)

diff --git a/src/mds/MDCache.h b/src/mds/MDCache.h
index e2b9a92ab406da8e17cdb288ce4194a6e355a5f2..b791363bf4f5bf722a25424c6c09f2af72a27a80 100644 (file)
--- a/src/mds/MDCache.h
+++ b/src/mds/MDCache.h
@@ -19,6 +19,7 @@
 
 #include <boost/utility/string_view.hpp>
 
+#include "common/DecayCounter.h"
 #include "include/types.h"
 #include "include/filepath.h"
 #include "include/elist.h"
@@ -719,9 +720,9 @@ public:
   size_t get_cache_size() { return lru.lru_get_size(); }
 
   // trimming
-  bool trim(uint64_t count=0);
+  std::pair<bool, uint64_t> trim(uint64_t count=0);
 private:
-  void trim_lru(uint64_t count, map<mds_rank_t, MCacheExpire*>& expiremap);
+  std::pair<bool, uint64_t> trim_lru(uint64_t count, map<mds_rank_t, MCacheExpire*>& expiremap);
   bool trim_dentry(CDentry *dn, map<mds_rank_t, MCacheExpire*>& expiremap);
   void trim_dirfrag(CDir *dir, CDir *con,
                    map<mds_rank_t, MCacheExpire*>& expiremap);
@@ -1157,6 +1158,10 @@ private:
                                LogSegment *ls, bufferlist *rollback=NULL);
   void finish_uncommitted_fragment(dirfrag_t basedirfrag, int op);
   void rollback_uncommitted_fragment(dirfrag_t basedirfrag, list<frag_t>& old_frags);
+
+
+  DecayCounter trim_counter;
+
 public:
   void wait_for_uncommitted_fragment(dirfrag_t dirfrag, MDSInternalContextBase *c) {
     assert(uncommitted_fragments.count(dirfrag));

diff --git a/src/mds/MDSDaemon.cc b/src/mds/MDSDaemon.cc
index d0a414a3c9b48f2cf0b850092a53a66f6688a75b..8fbaa2f2ac14966f532eacd337077d22d1a37f59 100644 (file)
--- a/src/mds/MDSDaemon.cc
+++ b/src/mds/MDSDaemon.cc
@@ -373,6 +373,7 @@ const char** MDSDaemon::get_tracked_conf_keys() const
     "mds_health_cache_threshold",
     "mds_cache_mid",
     "mds_dump_cache_threshold_formatter",
+    "mds_cache_trim_decay_rate",
     "mds_dump_cache_threshold_file",
     // MDBalancer
     "mds_bal_fragment_interval",