mds: journal oldest client tid

Journal the client's safe tid with new requests.  This keeps the client
completed_requests list trimmed, so that we don't build up a ginormous
list of all requests over the entire journal.

Forward rolling mds format change.

diff --git a/src/TODO b/src/TODO
index c1e4f480eab28e615814670421b5100399c2aa27..9969094c67723899c91e675abe86402ae967430e 100644 (file)
--- a/src/TODO
+++ b/src/TODO
@@ -390,7 +390,7 @@ uclient
 - hadoop: clean up assert usage
 
 mds
-- should we occasionally journal session completed requests??
+- put inode dirty fields into dirty_bits_t to reduce per-inode memory footprint
 - don't sync log on every clientreplay request?
 - pass issued, wanted into eval(lock) when eval() already has it?  (and otherwise optimize eval paths..)
 - add an up:shadow mode?

diff --git a/src/mds/Server.cc b/src/mds/Server.cc
index 8d05c7abc85e7f2ac5d84966180cbe1b656eee43..fd36335b09d7ebc0f2bdb1fc2618ff8c58345961 100644 (file)
--- a/src/mds/Server.cc
+++ b/src/mds/Server.cc
@@ -2276,7 +2276,7 @@ void Server::handle_client_openc(MDRequest *mdr)
   mdr->ls = mdlog->get_current_segment();
   EUpdate *le = new EUpdate(mdlog, "openc");
   mdlog->start_entry(le);
-  le->metablob.add_client_req(req->get_reqid());
+  le->metablob.add_client_req(req->get_reqid(), req->get_oldest_client_tid());
   journal_allocated_inos(mdr, &le->metablob);
   mdcache->predirty_journal_parents(mdr, &le->metablob, in, dn->get_dir(), PREDIRTY_PRIMARY|PREDIRTY_DIR, 1);
   le->metablob.add_primary_dentry(dn, true, in);
@@ -2608,7 +2608,7 @@ void Server::handle_client_setattr(MDRequest *mdr)
   pi->ctime = now;
 
   // log + wait
-  le->metablob.add_client_req(req->get_reqid());
+  le->metablob.add_client_req(req->get_reqid(), req->get_oldest_client_tid());
   mdcache->predirty_journal_parents(mdr, &le->metablob, cur, 0, PREDIRTY_PRIMARY, false);
   mdcache->journal_dirty_inode(mdr, &le->metablob, cur);
   
@@ -2649,7 +2649,7 @@ void Server::handle_client_opent(MDRequest *mdr, int cmode)
   EUpdate *le = new EUpdate(mdlog, "open_truncate");
   mdlog->start_entry(le);
   le->metablob.add_truncate_start(in->ino());
-  le->metablob.add_client_req(mdr->reqid);
+  le->metablob.add_client_req(mdr->reqid, mdr->client_request->get_oldest_client_tid());
 
   mdcache->predirty_journal_parents(mdr, &le->metablob, in, 0, PREDIRTY_PRIMARY, false);
   mdcache->journal_dirty_inode(mdr, &le->metablob, in);
@@ -2720,7 +2720,7 @@ void Server::handle_client_setlayout(MDRequest *mdr)
   mdr->ls = mdlog->get_current_segment();
   EUpdate *le = new EUpdate(mdlog, "setlayout");
   mdlog->start_entry(le);
-  le->metablob.add_client_req(req->get_reqid());
+  le->metablob.add_client_req(req->get_reqid(), req->get_oldest_client_tid());
   mdcache->predirty_journal_parents(mdr, &le->metablob, cur, 0, PREDIRTY_PRIMARY, false);
   mdcache->journal_dirty_inode(mdr, &le->metablob, cur);
   
@@ -2802,7 +2802,7 @@ void Server::handle_client_setxattr(MDRequest *mdr)
   mdr->ls = mdlog->get_current_segment();
   EUpdate *le = new EUpdate(mdlog, "setxattr");
   mdlog->start_entry(le);
-  le->metablob.add_client_req(req->get_reqid());
+  le->metablob.add_client_req(req->get_reqid(), req->get_oldest_client_tid());
   mdcache->predirty_journal_parents(mdr, &le->metablob, cur, 0, PREDIRTY_PRIMARY, false);
   mdcache->journal_cow_inode(mdr, &le->metablob, cur);
   le->metablob.add_primary_dentry(cur->get_projected_parent_dn(), true, cur, 0, 0, px);
@@ -2847,7 +2847,7 @@ void Server::handle_client_removexattr(MDRequest *mdr)
   mdr->ls = mdlog->get_current_segment();
   EUpdate *le = new EUpdate(mdlog, "removexattr");
   mdlog->start_entry(le);
-  le->metablob.add_client_req(req->get_reqid());
+  le->metablob.add_client_req(req->get_reqid(), req->get_oldest_client_tid());
   mdcache->predirty_journal_parents(mdr, &le->metablob, cur, 0, PREDIRTY_PRIMARY, false);
   mdcache->journal_cow_inode(mdr, &le->metablob, cur);
   le->metablob.add_primary_dentry(cur->get_projected_parent_dn(), true, cur, 0, 0, px);
@@ -2944,7 +2944,7 @@ void Server::handle_client_mknod(MDRequest *mdr)
   mdr->ls = mdlog->get_current_segment();
   EUpdate *le = new EUpdate(mdlog, "mknod");
   mdlog->start_entry(le);
-  le->metablob.add_client_req(req->get_reqid());
+  le->metablob.add_client_req(req->get_reqid(), req->get_oldest_client_tid());
   journal_allocated_inos(mdr, &le->metablob);
   
   mdcache->predirty_journal_parents(mdr, &le->metablob, newi, dn->get_dir(),
@@ -2997,7 +2997,7 @@ void Server::handle_client_mkdir(MDRequest *mdr)
   mdr->ls = mdlog->get_current_segment();
   EUpdate *le = new EUpdate(mdlog, "mkdir");
   mdlog->start_entry(le);
-  le->metablob.add_client_req(req->get_reqid());
+  le->metablob.add_client_req(req->get_reqid(), req->get_oldest_client_tid());
   journal_allocated_inos(mdr, &le->metablob);
   mdcache->predirty_journal_parents(mdr, &le->metablob, newi, dn->get_dir(), PREDIRTY_PRIMARY|PREDIRTY_DIR, 1);
   le->metablob.add_primary_dentry(dn, true, newi);
@@ -3061,7 +3061,7 @@ void Server::handle_client_symlink(MDRequest *mdr)
   mdr->ls = mdlog->get_current_segment();
   EUpdate *le = new EUpdate(mdlog, "symlink");
   mdlog->start_entry(le);
-  le->metablob.add_client_req(req->get_reqid());
+  le->metablob.add_client_req(req->get_reqid(), req->get_oldest_client_tid());
   journal_allocated_inos(mdr, &le->metablob);
   mdcache->predirty_journal_parents(mdr, &le->metablob, newi, dn->get_dir(), PREDIRTY_PRIMARY|PREDIRTY_DIR, 1);
   le->metablob.add_primary_dentry(dn, true, newi);
@@ -3172,7 +3172,7 @@ void Server::_link_local(MDRequest *mdr, CDentry *dn, CInode *targeti)
   // log + wait
   EUpdate *le = new EUpdate(mdlog, "link_local");
   mdlog->start_entry(le);
-  le->metablob.add_client_req(mdr->reqid);
+  le->metablob.add_client_req(mdr->reqid, mdr->client_request->get_oldest_client_tid());
   mdcache->predirty_journal_parents(mdr, &le->metablob, targeti, dn->get_dir(), PREDIRTY_DIR, 1);      // new dn
   mdcache->predirty_journal_parents(mdr, &le->metablob, targeti, 0, PREDIRTY_PRIMARY);           // targeti
   le->metablob.add_remote_dentry(dn, true, targeti->ino(), targeti->d_type());  // new remote
@@ -3262,7 +3262,7 @@ void Server::_link_remote(MDRequest *mdr, bool inc, CDentry *dn, CInode *targeti
   mdr->ls = mdlog->get_current_segment();
   EUpdate *le = new EUpdate(mdlog, inc ? "link_remote":"unlink_remote");
   mdlog->start_entry(le);
-  le->metablob.add_client_req(mdr->reqid);
+  le->metablob.add_client_req(mdr->reqid, mdr->client_request->get_oldest_client_tid());
   if (!mdr->more()->slaves.empty()) {
     dout(20) << " noting uncommitted_slaves " << mdr->more()->slaves << dendl;
     le->reqid = mdr->reqid;
@@ -3765,7 +3765,7 @@ void Server::_unlink_local(MDRequest *mdr, CDentry *dn, CDentry *straydn)
   // prepare log entry
   EUpdate *le = new EUpdate(mdlog, "unlink_local");
   mdlog->start_entry(le);
-  le->metablob.add_client_req(mdr->reqid);
+  le->metablob.add_client_req(mdr->reqid, mdr->client_request->get_oldest_client_tid());
 
   if (straydn) {
     assert(dnl->is_primary());
@@ -4253,7 +4253,7 @@ void Server::handle_client_rename(MDRequest *mdr)
   mdr->ls = mdlog->get_current_segment();
   EUpdate *le = new EUpdate(mdlog, "rename");
   mdlog->start_entry(le);
-  le->metablob.add_client_req(mdr->reqid);
+  le->metablob.add_client_req(mdr->reqid, mdr->client_request->get_oldest_client_tid());
   if (!mdr->more()->slaves.empty()) {
     dout(20) << " noting uncommitted_slaves " << mdr->more()->slaves << dendl;
     
@@ -5438,7 +5438,7 @@ void Server::handle_client_mksnap(MDRequest *mdr)
   EUpdate *le = new EUpdate(mdlog, "mksnap");
   mdlog->start_entry(le);
 
-  le->metablob.add_client_req(req->get_reqid());
+  le->metablob.add_client_req(req->get_reqid(), req->get_oldest_client_tid());
   le->metablob.add_table_transaction(TABLE_SNAP, stid);
   mdcache->predirty_journal_parents(mdr, &le->metablob, diri, 0, PREDIRTY_PRIMARY, false);
   mdcache->journal_cow_inode(mdr, &le->metablob, diri);
@@ -5584,7 +5584,7 @@ void Server::handle_client_rmsnap(MDRequest *mdr)
   EUpdate *le = new EUpdate(mdlog, "rmsnap");
   mdlog->start_entry(le);
 
-  le->metablob.add_client_req(req->get_reqid());
+  le->metablob.add_client_req(req->get_reqid(), req->get_oldest_client_tid());
   le->metablob.add_table_transaction(TABLE_SNAP, stid);
   mdcache->predirty_journal_parents(mdr, &le->metablob, diri, 0, PREDIRTY_PRIMARY, false);
   mdcache->journal_cow_inode(mdr, &le->metablob, diri);

diff --git a/src/mds/SessionMap.h b/src/mds/SessionMap.h
index 515050316344fa434cb4a4d04b7077fa5ab5d103..05f4b69a194aa021e2ed770a4e8f5dc64e1c8f8e 100644 (file)
--- a/src/mds/SessionMap.h
+++ b/src/mds/SessionMap.h
@@ -330,10 +330,12 @@ public:
     Session *session = get_session(rid.name);
     return session && session->have_completed_request(rid.tid);
   }
-  void add_completed_request(metareqid_t rid) {
+  void add_completed_request(metareqid_t rid, tid_t tid=0) {
     Session *session = get_session(rid.name);
     assert(session);
     session->add_completed_request(rid.tid);
+    if (tid)
+      session->trim_completed_requests(tid);
   }
   void trim_completed_requests(entity_name_t c, tid_t tid) {
     Session *session = get_session(c);

diff --git a/src/mds/events/EMetaBlob.h b/src/mds/events/EMetaBlob.h
index 7dbedab4f2aa10ef6ecc1125cb43c1692dca71f2..24a8d7b87d13f49d728464fc85ea4991140cb969 100644 (file)
--- a/src/mds/events/EMetaBlob.h
+++ b/src/mds/events/EMetaBlob.h
@@ -356,11 +356,11 @@ private:
   vector<inodeno_t> destroyed_inodes;
 
   // idempotent op(s)
-  list<metareqid_t> client_reqs;
+  list<pair<metareqid_t,__u64> > client_reqs;
 
  public:
   void encode(bufferlist& bl) const {
-    __u8 struct_v = 1;
+    __u8 struct_v = 2;
     ::encode(struct_v, bl);
     ::encode(lump_order, bl);
     ::encode(lump_map, bl);
@@ -403,7 +403,16 @@ private:
     ::decode(truncate_start, bl);
     ::decode(truncate_finish, bl);
     ::decode(destroyed_inodes, bl);
-    ::decode(client_reqs, bl);
+    if (struct_v >= 2) 
+      ::decode(client_reqs, bl);
+    else {
+      list<metareqid_t> r;
+      ::decode(r, bl);
+      while (!r.empty()) {
+       client_reqs.push_back(pair<metareqid_t,__u64>(r.front(), 0));
+       r.pop_front();
+      }
+    }
   }
 
 
@@ -427,8 +436,8 @@ private:
     }
   }
 
-  void add_client_req(metareqid_t r) {
-    client_reqs.push_back(r);
+  void add_client_req(metareqid_t r, __u64 tid=0) {
+    client_reqs.push_back(pair<metareqid_t,__u64>(r, tid));
   }
 
   void add_table_transaction(int table, version_t tid) {

diff --git a/src/mds/journal.cc b/src/mds/journal.cc
index 9c2bb4c7adb27363767ca7b6cd65e1ccc1cd6d05..6e6cc26689f4bb4fd22ec4c7c1d6ed9895ae8981 100644 (file)
--- a/src/mds/journal.cc
+++ b/src/mds/journal.cc
@@ -673,13 +673,13 @@ void EMetaBlob::replay(MDS *mds, LogSegment *logseg)
   }
 
   // client requests
-  for (list<metareqid_t>::iterator p = client_reqs.begin();
+  for (list<pair<metareqid_t, __u64> >::iterator p = client_reqs.begin();
        p != client_reqs.end();
        ++p)
-    if (p->name.is_client()) {
-      dout(10) << "EMetaBlob.replay request " << *p << dendl;
-      if (mds->sessionmap.have_session(p->name))
-       mds->sessionmap.add_completed_request(*p);
+    if (p->first.name.is_client()) {
+      dout(10) << "EMetaBlob.replay request " << p->first << " " << p->second << dendl;
+      if (mds->sessionmap.have_session(p->first.name))
+       mds->sessionmap.add_completed_request(p->first, p->second);
     }