From 2ae07a8ad7e874b5dbd180c614b0cb1462ab55e9 Mon Sep 17 00:00:00 2001
From: Yanhu Cao <jrcaoyanhu@jd.com>
Date: Tue, 16 Jun 2020 14:07:26 +0800
Subject: [PATCH] mds: reset heartbeat in EMetaBlob replay

Fixes: https://tracker.ceph.com/issues/46042
Signed-off-by: Yanhu Cao <jrcaoyanhu@jd.com>
(cherry picked from commit 07031a0de9654e029fc5063899c2d9394c7f4668)
---
 src/mds/journal.cc | 40 ++++++++++++++++++++++++++++++++++++++++
 1 file changed, 40 insertions(+)

diff --git a/src/mds/journal.cc b/src/mds/journal.cc
index b9ef6acab8990..3eb24af2fed3b 100644
--- a/src/mds/journal.cc
+++ b/src/mds/journal.cc
@@ -1133,6 +1133,7 @@ void EMetaBlob::replay(MDSRank *mds, LogSegment *logseg, MDSlaveUpdate *slaveup)
   set<CInode*> linked;
 
   // walk through my dirs (in order!)
+  int count = 0;
   for (const auto& lp : lump_order) {
     dout(10) << "EMetaBlob.replay dir " << lp << dendl;
     dirlump &lump = lump_map[lp];
@@ -1284,6 +1285,9 @@ void EMetaBlob::replay(MDSRank *mds, LogSegment *logseg, MDSlaveUpdate *slaveup)
       else
 	in->state_clear(CInode::STATE_AUTH);
       ceph_assert(g_conf()->mds_kill_journal_replay_at != 2);
+
+      if (!(++count % 1000))
+        mds->heartbeat_reset();
     }
 
     // remote dentries
@@ -1315,6 +1319,9 @@ void EMetaBlob::replay(MDSRank *mds, LogSegment *logseg, MDSlaveUpdate *slaveup)
       }
       if (lump.is_importing())
 	dn->state_set(CDentry::STATE_AUTH);
+
+      if (!(++count % 1000))
+        mds->heartbeat_reset();
     }
 
     // null dentries
@@ -1350,6 +1357,9 @@ void EMetaBlob::replay(MDSRank *mds, LogSegment *logseg, MDSlaveUpdate *slaveup)
 
       // Make null dentries the first things we trim
       dout(10) << "EMetaBlob.replay pushing to bottom of lru " << *dn << dendl;
+
+      if (!(++count % 1000))
+        mds->heartbeat_reset();
     }
   }
 
@@ -1380,6 +1390,9 @@ void EMetaBlob::replay(MDSRank *mds, LogSegment *logseg, MDSlaveUpdate *slaveup)
 	    slaveup->olddirs.insert(dir->inode);
 	  else
 	    dir->state_set(CDir::STATE_AUTH);
+
+          if (!(++count % 1000))
+            mds->heartbeat_reset();
 	}
       }
 
@@ -1409,6 +1422,9 @@ void EMetaBlob::replay(MDSRank *mds, LogSegment *logseg, MDSlaveUpdate *slaveup)
 	dout(10) << " creating new rename import bound " << *dir << dendl;
 	dir->state_clear(CDir::STATE_AUTH);
 	mds->mdcache->adjust_subtree_auth(dir, CDIR_AUTH_UNDEF);
+
+        if (!(++count % 1000))
+          mds->heartbeat_reset();
       }
     }
 
@@ -1419,6 +1435,9 @@ void EMetaBlob::replay(MDSRank *mds, LogSegment *logseg, MDSlaveUpdate *slaveup)
 	continue;
       ceph_assert(p->first->is_dir());
       mds->mdcache->adjust_subtree_after_rename(p->first, p->second, false);
+
+      if (!(++count % 1000))
+        mds->heartbeat_reset();
     }
   }
 
@@ -1434,6 +1453,9 @@ void EMetaBlob::replay(MDSRank *mds, LogSegment *logseg, MDSlaveUpdate *slaveup)
 	  in->snaprealm->adjust_parent();
       } else
 	mds->mdcache->remove_inode_recursive(in);
+
+      if (!(++count % 1000))
+        mds->heartbeat_reset();
     }
   }
 
@@ -1444,6 +1466,9 @@ void EMetaBlob::replay(MDSRank *mds, LogSegment *logseg, MDSlaveUpdate *slaveup)
     MDSTableClient *client = mds->get_table_client(p.first);
     if (client)
       client->got_journaled_agree(p.second, logseg);
+
+    if (!(++count % 1000))
+      mds->heartbeat_reset();
   }
 
   // opened ino?
@@ -1534,6 +1559,9 @@ void EMetaBlob::replay(MDSRank *mds, LogSegment *logseg, MDSlaveUpdate *slaveup)
     CInode *in = mds->mdcache->get_inode(ino);
     ceph_assert(in);
     mds->mdcache->add_recovered_truncate(in, logseg);
+
+    if (!(++count % 1000))
+      mds->heartbeat_reset();
   }
   for (const auto& p : truncate_finish) {
     LogSegment *ls = mds->mdlog->get_segment(p.second);
@@ -1542,6 +1570,9 @@ void EMetaBlob::replay(MDSRank *mds, LogSegment *logseg, MDSlaveUpdate *slaveup)
       ceph_assert(in);
       mds->mdcache->remove_recovered_truncate(in, ls);
     }
+
+    if (!(++count % 1000))
+      mds->heartbeat_reset();
   }
 
   // destroyed inodes
@@ -1561,6 +1592,9 @@ void EMetaBlob::replay(MDSRank *mds, LogSegment *logseg, MDSlaveUpdate *slaveup)
       } else {
 	dout(10) << "EMetaBlob.replay destroyed " << *p << ", not in cache" << dendl;
       }
+
+      if (!(++count % 1000))
+        mds->heartbeat_reset();
     }
     mds->mdcache->open_file_table.note_destroyed_inos(logseg->seq, destroyed_inodes);
   }
@@ -1580,6 +1614,9 @@ void EMetaBlob::replay(MDSRank *mds, LogSegment *logseg, MDSlaveUpdate *slaveup)
 	  session->trim_completed_requests(p.second);
       }
     }
+
+    if (!(++count % 1000))
+      mds->heartbeat_reset();
   }
 
   // client flushes
@@ -1593,6 +1630,9 @@ void EMetaBlob::replay(MDSRank *mds, LogSegment *logseg, MDSlaveUpdate *slaveup)
 	  session->trim_completed_flushes(p.second);
       }
     }
+
+    if (!(++count % 1000))
+      mds->heartbeat_reset();
   }
 
   // update segment
-- 
2.39.5