osd: add last_user_version to pg_info_t

We add a corresponding user_version to pg_log_entry_t, and the logic
to assign from one to the other and to recover last_user_version from
a master's log. We aren't yet setting it to anything, though.

Signed-off-by: Greg Farnum <greg@inktank.com>

diff --git a/src/osd/PG.cc b/src/osd/PG.cc
index 49ea61a603ac9556c73fc3ebcf99b80a22d3044f..d8ada534c302bfb5e443ab0ece6867485250a63a 100644 (file)
--- a/src/osd/PG.cc
+++ b/src/osd/PG.cc
@@ -1751,6 +1751,8 @@ void PG::split_into(pg_t child_pgid, PG *child, unsigned split_bits)
   info.last_update = pg_log.get_head();
   child->info.last_update = child->pg_log.get_head();
 
+  child->info.last_user_version = info.last_user_version;
+
   info.log_tail = pg_log.get_tail();
   child->info.log_tail = child->pg_log.get_tail();
 
@@ -2304,6 +2306,7 @@ void PG::add_log_entry(pg_log_entry_t& e, bufferlist& log_bl)
   // raise last_update.
   assert(e.version > info.last_update);
   info.last_update = e.version;
+  info.last_user_version = e.user_version;
 
   // log mutation
   pg_log.add(e);

diff --git a/src/osd/PGLog.cc b/src/osd/PGLog.cc
index dac1f33fd91cde1e4b7e949dd43574bd3289e52f..6fcca1b41bd5203c2916fb6840e5f0a90c3f4dfd 100644 (file)
--- a/src/osd/PGLog.cc
+++ b/src/osd/PGLog.cc
@@ -514,6 +514,7 @@ void PGLog::merge_log(ObjectStore::Transaction& t,
     log.index();   
 
     info.last_update = log.head = olog.head;
+    info.last_user_version = oinfo.last_user_version;
     info.purged_snaps = oinfo.purged_snaps;
 
     // process divergent items

diff --git a/src/osd/osd_types.cc b/src/osd/osd_types.cc
index 9594cc6057938ac371c581d18dc36af0836a63b9..3f6045568d8a1be53e04de2eaee8322872cc7c4d 100644 (file)
--- a/src/osd/osd_types.cc
+++ b/src/osd/osd_types.cc
@@ -1553,7 +1553,7 @@ void pg_history_t::generate_test_instances(list<pg_history_t*>& o)
 
 void pg_info_t::encode(bufferlist &bl) const
 {
-  ENCODE_START(27, 26, bl);
+  ENCODE_START(28, 26, bl);
   ::encode(pgid, bl);
   ::encode(last_update, bl);
   ::encode(last_complete, bl);
@@ -1563,12 +1563,13 @@ void pg_info_t::encode(bufferlist &bl) const
   history.encode(bl);
   ::encode(purged_snaps, bl);
   ::encode(last_epoch_started, bl);
+  ::encode(last_user_version, bl);
   ENCODE_FINISH(bl);
 }
 
 void pg_info_t::decode(bufferlist::iterator &bl)
 {
-  DECODE_START_LEGACY_COMPAT_LEN(27, 26, 26, bl);
+  DECODE_START_LEGACY_COMPAT_LEN(28, 26, 26, bl);
   if (struct_v < 23) {
     old_pg_t opgid;
     ::decode(opgid, bl);
@@ -1598,6 +1599,10 @@ void pg_info_t::decode(bufferlist::iterator &bl)
   } else {
     ::decode(last_epoch_started, bl);
   }
+  if (struct_v >= 28)
+    ::decode(last_user_version, bl);
+  else
+    last_user_version = last_update.version;
   DECODE_FINISH(bl);
 }
 
@@ -1609,6 +1614,7 @@ void pg_info_t::dump(Formatter *f) const
   f->dump_stream("last_update") << last_update;
   f->dump_stream("last_complete") << last_complete;
   f->dump_stream("log_tail") << log_tail;
+  f->dump_int("last_user_version", last_user_version);
   f->dump_stream("last_backfill") << last_backfill;
   f->dump_stream("purged_snaps") << purged_snaps;
   f->open_object_section("history");
@@ -1634,6 +1640,7 @@ void pg_info_t::generate_test_instances(list<pg_info_t*>& o)
   o.back()->pgid = pg_t(1, 2, -1);
   o.back()->last_update = eversion_t(3, 4);
   o.back()->last_complete = eversion_t(5, 6);
+  o.back()->last_user_version = 2;
   o.back()->log_tail = eversion_t(7, 8);
   o.back()->last_backfill = hobject_t(object_t("objname"), "key", 123, 456, -1, "");
   list<pg_stat_t*> s;
@@ -1912,7 +1919,7 @@ void pg_log_entry_t::decode_with_checksum(bufferlist::iterator& p)
 
 void pg_log_entry_t::encode(bufferlist &bl) const
 {
-  ENCODE_START(7, 4, bl);
+  ENCODE_START(8, 4, bl);
   ::encode(op, bl);
   ::encode(soid, bl);
   ::encode(version, bl);
@@ -1934,12 +1941,13 @@ void pg_log_entry_t::encode(bufferlist &bl) const
   if (op == LOST_REVERT)
     ::encode(prior_version, bl);
   ::encode(snaps, bl);
+  ::encode(user_version, bl);
   ENCODE_FINISH(bl);
 }
 
 void pg_log_entry_t::decode(bufferlist::iterator &bl)
 {
-  DECODE_START_LEGACY_COMPAT_LEN(7, 4, 4, bl);
+  DECODE_START_LEGACY_COMPAT_LEN(8, 4, 4, bl);
   ::decode(op, bl);
   if (struct_v < 2) {
     sobject_t old_soid;
@@ -1976,6 +1984,11 @@ void pg_log_entry_t::decode(bufferlist::iterator &bl)
     ::decode(snaps, bl);
   }
 
+  if (struct_v >= 8)
+    ::decode(user_version, bl);
+  else
+    user_version = version.version;
+
   DECODE_FINISH(bl);
 }
 

diff --git a/src/osd/osd_types.h b/src/osd/osd_types.h
index 4c34e81871f6df44255bc8bb5ed427b5efcb55c8..ea399573eb0a8564ce3d8853e3a90f686f27d00b 100644 (file)
--- a/src/osd/osd_types.h
+++ b/src/osd/osd_types.h
@@ -1267,6 +1267,8 @@ struct pg_info_t {
   eversion_t last_complete;  // last version pg was complete through.
   epoch_t last_epoch_started;// last epoch at which this pg started on this osd
   
+  version_t last_user_version; // last user object version applied to store
+
   eversion_t log_tail;     // oldest log entry.
 
   hobject_t last_backfill;   // objects >= this and < last_complete may be missing
@@ -1278,11 +1280,13 @@ struct pg_info_t {
   pg_history_t history;
 
   pg_info_t()
-    : last_epoch_started(0), last_backfill(hobject_t::get_max())
+    : last_epoch_started(0), last_user_version(0),
+      last_backfill(hobject_t::get_max())
   { }
   pg_info_t(pg_t p)
     : pgid(p),
-      last_epoch_started(0), last_backfill(hobject_t::get_max())
+      last_epoch_started(0), last_user_version(0),
+      last_backfill(hobject_t::get_max())
   { }
   
   bool is_empty() const { return last_update.version == 0; }
@@ -1481,6 +1485,7 @@ struct pg_log_entry_t {
   __s32      op;
   hobject_t  soid;
   eversion_t version, prior_version, reverting_to;
+  version_t user_version; // the user version for this entry
   osd_reqid_t reqid;  // caller+tid to uniquely identify request
   utime_t     mtime;  // this is the _user_ mtime, mind you
   bufferlist snaps;   // only for clone entries
@@ -1490,12 +1495,13 @@ struct pg_log_entry_t {
   uint64_t offset;   // [soft state] my offset on disk
       
   pg_log_entry_t()
-    : op(0), invalid_hash(false), invalid_pool(false), offset(0) {}
+    : op(0), user_version(0),
+      invalid_hash(false), invalid_pool(false), offset(0) {}
   pg_log_entry_t(int _op, const hobject_t& _soid, 
                 const eversion_t& v, const eversion_t& pv,
                 const osd_reqid_t& rid, const utime_t& mt)
     : op(_op), soid(_soid), version(v),
-      prior_version(pv),
+      prior_version(pv), user_version(0),
       reqid(rid), mtime(mt), invalid_hash(false), invalid_pool(false),
       offset(0) {}