Consolidate garbage collection information into txn_gc_info. Explicitly

separate the difference between the oldest referenced xid we can use for
implicit promotion vs simple garbage collection.

ft/ft-flusher.cc

@@ -1544,11 +1544,7 @@ ft_merge_child(
     }
 }
 
-static void ft_flush_some_child(
-    FT ft,
-    FTNODE parent,
-    struct flusher_advice *fa
-    )
+void toku_ft_flush_some_child(FT ft, FTNODE parent, struct flusher_advice *fa)
 // Effect: This function does the following:
 //   - Pick a child of parent (the heaviest child),
 //   - flush from parent to child,
@@ -1562,7 +1558,7 @@ static void ft_flush_some_child(
     NONLEAF_CHILDINFO bnc = NULL;
     paranoid_invariant(parent->height>0);
     toku_assert_entire_node_in_memory(parent);
-    TXNID oldest_referenced_xid = parent->oldest_referenced_xid_known;
+    TXNID parent_oldest_referenced_xid_known = parent->oldest_referenced_xid_known;
 
     // pick the child we want to flush to
     int childnum = fa->pick_child(ft, parent, fa->extra);
@@ -1655,7 +1651,7 @@ static void ft_flush_some_child(
             ft,
             bnc,
             child,
-            oldest_referenced_xid
+            parent_oldest_referenced_xid_known
             );
         destroy_nonleaf_childinfo(bnc);
     }
@@ -1679,10 +1675,10 @@ static void ft_flush_some_child(
             parent = NULL;
         }
         //
-        // it is the responsibility of ft_flush_some_child to unpin child
+        // it is the responsibility of toku_ft_flush_some_child to unpin child
         //
         if (child->height > 0 && fa->should_recursively_flush(child, fa->extra)) {
-            ft_flush_some_child(ft, child, fa);
+            toku_ft_flush_some_child(ft, child, fa);
         }
         else {
             toku_unpin_ftnode_off_client_thread(ft, child);
@@ -1709,13 +1705,6 @@ static void ft_flush_some_child(
     }
 }
 
-void toku_ft_flush_some_child(FT ft, FTNODE parent, struct flusher_advice *fa) {
-    // Vanilla flush_some_child flushes from parent to child without
-    // providing a meaningful oldest_referenced_xid. No simple garbage
-    // collection is performed.
-    return ft_flush_some_child(ft, parent, fa);
-}
-
 static void
 update_cleaner_status(
     FTNODE node,
@@ -1857,7 +1846,7 @@ struct flusher_extra {
     FT h;
     FTNODE node;
     NONLEAF_CHILDINFO bnc;
-    TXNID oldest_referenced_xid;
+    TXNID parent_oldest_referenced_xid_known;
 };
 
 //
@@ -1896,16 +1885,16 @@ static void flush_node_fun(void *fe_v)
             fe->h,
             fe->bnc,
             fe->node,
-            fe->oldest_referenced_xid
+            fe->parent_oldest_referenced_xid_known
             );
         destroy_nonleaf_childinfo(fe->bnc);
 
         // after the flush has completed, now check to see if the node needs flushing
-        // If so, call ft_flush_some_child on the node (because this flush intends to
+        // If so, call toku_ft_flush_some_child on the node (because this flush intends to
         // pass a meaningful oldest referenced xid for simple garbage collection), and it is the
         // responsibility of the flush to unlock the node. otherwise, we unlock it here.
         if (fe->node->height > 0 && toku_ft_nonleaf_is_gorged(fe->node, fe->h->h->nodesize)) {
-            ft_flush_some_child(fe->h, fe->node, &fa);
+            toku_ft_flush_some_child(fe->h, fe->node, &fa);
         }
         else {
             toku_unpin_ftnode_off_client_thread(fe->h,fe->node);
@@ -1916,7 +1905,7 @@ static void flush_node_fun(void *fe_v)
         // bnc, which means we are tasked with flushing some
         // buffer in the node.
         // It is the responsibility of flush some child to unlock the node
-        ft_flush_some_child(fe->h, fe->node, &fa);
+        toku_ft_flush_some_child(fe->h, fe->node, &fa);
     }
     remove_background_job_from_cf(fe->h->cf);
     toku_free(fe);
@@ -1927,13 +1916,13 @@ place_node_and_bnc_on_background_thread(
     FT h,
     FTNODE node,
     NONLEAF_CHILDINFO bnc,
-    TXNID oldest_referenced_xid)
+    TXNID parent_oldest_referenced_xid_known)
 {
     struct flusher_extra *XMALLOC(fe);
     fe->h = h;
     fe->node = node;
     fe->bnc = bnc;
-    fe->oldest_referenced_xid = oldest_referenced_xid;
+    fe->parent_oldest_referenced_xid_known = parent_oldest_referenced_xid_known;
     cachefile_kibbutz_enq(h->cf, flush_node_fun, fe);
 }
 
@@ -1953,7 +1942,7 @@ place_node_and_bnc_on_background_thread(
 void toku_ft_flush_node_on_background_thread(FT h, FTNODE parent)
 {
     toku::context flush_ctx(CTX_FLUSH);
-    TXNID oldest_referenced_xid_known = parent->oldest_referenced_xid_known;
+    TXNID parent_oldest_referenced_xid_known = parent->oldest_referenced_xid_known;
     //
     // first let's see if we can detach buffer on client thread
     // and pick the child we want to flush to
@@ -1970,7 +1959,7 @@ void toku_ft_flush_node_on_background_thread(FT h, FTNODE parent)
         // In this case, we could not lock the child, so just place the parent on the background thread
         // In the callback, we will use toku_ft_flush_some_child, which checks to
         // see if we should blow away the old basement nodes.
-        place_node_and_bnc_on_background_thread(h, parent, NULL, oldest_referenced_xid_known);
+        place_node_and_bnc_on_background_thread(h, parent, NULL, parent_oldest_referenced_xid_known);
     }
     else {
         //
@@ -1999,7 +1988,7 @@ void toku_ft_flush_node_on_background_thread(FT h, FTNODE parent)
             // so, because we know for sure the child is not
             // reactive, we can unpin the parent
             //
-            place_node_and_bnc_on_background_thread(h, child, bnc, oldest_referenced_xid_known);
+            place_node_and_bnc_on_background_thread(h, child, bnc, parent_oldest_referenced_xid_known);
             toku_unpin_ftnode(h, parent);
         }
         else {
@@ -2009,7 +1998,7 @@ void toku_ft_flush_node_on_background_thread(FT h, FTNODE parent)
             toku_unpin_ftnode(h, child);
             // Again, we'll have the parent on the background thread, so
             // we don't need to destroy the basement nodes yet.
-            place_node_and_bnc_on_background_thread(h, parent, NULL, oldest_referenced_xid_known);
+            place_node_and_bnc_on_background_thread(h, parent, NULL, parent_oldest_referenced_xid_known);
         }
     }
 }

ft/ft-internal.h

@@ -229,7 +229,7 @@ long toku_bnc_memory_size(NONLEAF_CHILDINFO bnc);
 long toku_bnc_memory_used(NONLEAF_CHILDINFO bnc);
 void toku_bnc_insert_msg(NONLEAF_CHILDINFO bnc, const void *key, ITEMLEN keylen, const void *data, ITEMLEN datalen, enum ft_msg_type type, MSN msn, XIDS xids, bool is_fresh, DESCRIPTOR desc, ft_compare_func cmp);
 void toku_bnc_empty(NONLEAF_CHILDINFO bnc);
-void toku_bnc_flush_to_child(FT h, NONLEAF_CHILDINFO bnc, FTNODE child, TXNID oldest_referenced_xid);
+void toku_bnc_flush_to_child(FT h, NONLEAF_CHILDINFO bnc, FTNODE child, TXNID parent_oldest_referenced_xid_known);
 bool toku_bnc_should_promote(FT ft, NONLEAF_CHILDINFO bnc) __attribute__((const, nonnull));
 bool toku_ft_nonleaf_is_gorged(FTNODE node, uint32_t nodesize);
 
@@ -1027,7 +1027,7 @@ int toku_testsetup_insert_to_nonleaf (FT_HANDLE brt, BLOCKNUM, enum ft_msg_type,
 void toku_pin_node_with_min_bfe(FTNODE* node, BLOCKNUM b, FT_HANDLE t);
 
 // toku_ft_root_put_cmd() accepts non-constant cmd because this is where we set the msn
-void toku_ft_root_put_cmd(FT h, FT_MSG_S * cmd, TXNID oldest_referenced_xid, GC_INFO gc_info);
+void toku_ft_root_put_cmd(FT h, FT_MSG_S * cmd, txn_gc_info *gc_info);
 
 void
 toku_get_node_for_verify(
@@ -1198,9 +1198,7 @@ toku_ft_bn_apply_cmd_once (
     const FT_MSG cmd,
     uint32_t idx,
     LEAFENTRY le,
-    TXNID oldest_referenced_xid,
-    GC_INFO gc_info,
-    txn_manager_state *txn_state_for_gc,
+    txn_gc_info *gc_info,
     uint64_t *workdonep,
     STAT64INFO stats_to_update
     );
@@ -1212,9 +1210,7 @@ toku_ft_bn_apply_cmd (
     DESCRIPTOR desc,
     BASEMENTNODE bn,
     FT_MSG cmd,
-    TXNID oldest_referenced_xid,
-    GC_INFO gc_info,
-    txn_manager_state *txn_state_for_gc,
+    txn_gc_info *gc_info,
     uint64_t *workdone,
     STAT64INFO stats_to_update
     );
@@ -1227,8 +1223,7 @@ toku_ft_leaf_apply_cmd (
     FTNODE node,
     int target_childnum,
     FT_MSG cmd,
-    GC_INFO gc_info,
-    txn_manager_state *txn_state_for_gc,
+    txn_gc_info *gc_info,
     uint64_t *workdone,
     STAT64INFO stats_to_update
     );
@@ -1242,8 +1237,7 @@ toku_ft_node_put_cmd (
     int target_childnum,
     FT_MSG cmd,
     bool is_fresh,
-    GC_INFO gc_info,
-    txn_manager_state *txn_state_for_gc,
+    txn_gc_info *gc_info,
     size_t flow_deltas[],
     STAT64INFO stats_to_update
     );

ft/ft-ops.h

@@ -243,9 +243,11 @@ void toku_ft_delete (FT_HANDLE brt, DBT *k, TOKUTXN txn);
 // Effect: Delete a key from a brt if the oplsn is newer than the brt lsn.  This function is called during recovery.
 void toku_ft_maybe_delete (FT_HANDLE brt, DBT *k, TOKUTXN txn, bool oplsn_valid, LSN oplsn, bool do_logging);
 
-void toku_ft_send_insert(FT_HANDLE brt, DBT *key, DBT *val, XIDS xids, enum ft_msg_type type, TXNID oldest_referenced_xid, GC_INFO gc_info);
-void toku_ft_send_delete(FT_HANDLE brt, DBT *key, XIDS xids, TXNID oldest_referenced_xid, GC_INFO gc_info);
-void toku_ft_send_commit_any(FT_HANDLE brt, DBT *key, XIDS xids, TXNID oldest_referenced_xids, GC_INFO gc_info);
+TXNID toku_ft_get_oldest_referenced_xid_estimate(FT_HANDLE ft_h);
+
+void toku_ft_send_insert(FT_HANDLE brt, DBT *key, DBT *val, XIDS xids, enum ft_msg_type type, txn_gc_info *gc_info);
+void toku_ft_send_delete(FT_HANDLE brt, DBT *key, XIDS xids, txn_gc_info *gc_info);
+void toku_ft_send_commit_any(FT_HANDLE brt, DBT *key, XIDS xids, txn_gc_info *gc_info);
 
 int toku_close_ft_handle_nolsn (FT_HANDLE, char **error_string)  __attribute__ ((warn_unused_result));
 

ft/ft-test-helpers.cc

@@ -221,6 +221,7 @@ int toku_testsetup_insert_to_leaf (FT_HANDLE brt, BLOCKNUM blocknum, const char
                                     toku_fill_dbt(&valdbt, val, vallen) } } };
 
     static size_t zero_flow_deltas[] = { 0, 0 };
+    txn_gc_info gc_info(nullptr, TXNID_NONE, TXNID_NONE, true);
     toku_ft_node_put_cmd (
         brt->ft->compare_fun,
         brt->ft->update_fun,
@@ -229,8 +230,7 @@ int toku_testsetup_insert_to_leaf (FT_HANDLE brt, BLOCKNUM blocknum, const char
         -1,
         &cmd,
         true,
-        make_gc_info(true),
-        nullptr,
+        &gc_info,
         zero_flow_deltas,
         NULL
         );

ft/ftloader.cc

@@ -2925,7 +2925,8 @@ static void add_pair_to_leafnode (struct leaf_buf *lbuf, unsigned char *key, int
                      .xids = lbuf->xids,
                      .u = { .id = { &thekey, &theval } } };
     uint64_t workdone=0;
-    toku_ft_bn_apply_cmd_once(BLB(leafnode,0), &cmd, idx, NULL, TXNID_NONE, make_gc_info(true), nullptr, &workdone, stats_to_update);
+    txn_gc_info gc_info(nullptr, TXNID_NONE, TXNID_NONE, true);
+    toku_ft_bn_apply_cmd_once(BLB(leafnode,0), &cmd, idx, NULL, &gc_info, &workdone, stats_to_update);
 }
 
 static int write_literal(struct dbout *out, void*data,  size_t len) {

ft/fttypes.h

@@ -143,10 +143,6 @@ typedef TOKU_XA_XID *XIDP; // this is the type that's passed to the logger code
 
 static inline BLOCKNUM make_blocknum(int64_t b) { BLOCKNUM result={b}; return result; }
 
-typedef struct gc_info_s { bool mvcc_needed; } GC_INFO;
-
-static inline GC_INFO make_gc_info(bool mvcc_needed) { GC_INFO result = {mvcc_needed}; return result; }
-
 // This struct hold information about values stored in the cachetable.
 // As one can tell from the names, we are probably violating an
 // abstraction layer by placing names.

ft/leafentry.h

@@ -247,13 +247,11 @@ toku_le_apply_msg(FT_MSG   msg,
                   LEAFENTRY old_leafentry, // NULL if there was no stored data.
                   bn_data* data_buffer, // bn_data storing leafentry, if NULL, means there is no bn_data
                   uint32_t idx, // index in data_buffer where leafentry is stored (and should be replaced
-                  TXNID oldest_referenced_xid,
-                  GC_INFO gc_info,
-                  txn_manager_state *txn_state_for_gc,
+                  txn_gc_info *gc_info,
                   LEAFENTRY *new_leafentry_p,
                   int64_t * numbytes_delta_p);
 
-bool toku_le_worth_running_garbage_collection(LEAFENTRY le, TXNID oldest_referenced_xid_known);
+bool toku_le_worth_running_garbage_collection(LEAFENTRY le, txn_gc_info *gc_info);
 
 void
 toku_le_garbage_collect(LEAFENTRY old_leaf_entry,
@@ -261,11 +259,8 @@ toku_le_garbage_collect(LEAFENTRY old_leaf_entry,
                         uint32_t idx,
                         void* keyp,
                         uint32_t keylen,
+                        txn_gc_info *gc_info,
                         LEAFENTRY *new_leaf_entry,
-                        const xid_omt_t &snapshot_xids,
-                        const rx_omt_t &referenced_xids,
-                        const xid_omt_t &live_root_txns,
-                        TXNID oldest_referenced_xid_known,
                         int64_t * numbytes_delta_p);
 
 #endif /* TOKU_LEAFENTRY_H */

ft/log-internal.h

@@ -247,7 +247,6 @@ struct tokutxn {
     DB_TXN *container_db_txn; // reference to DB_TXN that contains this tokutxn
     xid_omt_t *live_root_txn_list; // the root txns live when the root ancestor (self if a root) started.
     XIDS xids; // Represents the xid list
-    TXNID oldest_referenced_xid;
 
     TOKUTXN snapshot_next;
     TOKUTXN snapshot_prev;

ft/roll.cc

@@ -96,6 +96,7 @@ PATENT RIGHTS GRANT:
 #include "ft.h"
 #include "ft-ops.h"
 #include "log-internal.h"
+//#include "txn_manager.h"
 #include "xids.h"
 #include "rollback-apply.h"
 
@@ -265,7 +266,14 @@ static int do_insertion (enum ft_msg_type type, FILENUM filenum, BYTESTRING key,
                                           ? toku_fill_dbt(&data_dbt, data->data, data->len)
                                           : toku_init_dbt(&data_dbt) } } };
 
-        toku_ft_root_put_cmd(h, &ftcmd, txn->oldest_referenced_xid, make_gc_info(!txn->for_recovery));
+        TXN_MANAGER txn_manager = toku_logger_get_txn_manager(txn->logger);
+        TXNID oldest_referenced_xid_estimate = toku_txn_manager_get_oldest_referenced_xid_estimate(txn_manager);
+        txn_gc_info gc_info(nullptr,
+                            oldest_referenced_xid_estimate,
+                            // no messages above us, we can implicitly promote uxrs based on this xid
+                            oldest_referenced_xid_estimate,
+                            !txn->for_recovery);
+        toku_ft_root_put_cmd(h, &ftcmd, &gc_info);
         if (reset_root_xid_that_created) {
             TXNID new_root_xid_that_created = xids_get_outermost_xid(xids);
             toku_reset_root_xid_that_created(h, new_root_xid_that_created);

ft/tests/make-tree.cc

@@ -124,8 +124,9 @@ append_leaf(FTNODE leafnode, void *key, size_t keylen, void *val, size_t vallen)
     MSN msn = next_dummymsn();
 
     // apply an insert to the leaf node
+    txn_gc_info gc_info(nullptr, TXNID_NONE, TXNID_NONE, false);
     FT_MSG_S cmd = { FT_INSERT, msn, xids_get_root_xids(), .u = {.id = { &thekey, &theval }} };
-    toku_ft_bn_apply_cmd_once(BLB(leafnode,0), &cmd, idx, NULL, TXNID_NONE, make_gc_info(false), NULL, NULL);
+    toku_ft_bn_apply_cmd_once(BLB(leafnode,0), &cmd, idx, NULL, &gc_info, NULL, NULL);
 
     leafnode->max_msn_applied_to_node_on_disk = msn;
 

ft/tests/msnfilter.cc

@@ -132,8 +132,9 @@ append_leaf(FT_HANDLE brt, FTNODE leafnode, void *key, uint32_t keylen, void *va
     MSN msn = next_dummymsn();
     brt->ft->h->max_msn_in_ft = msn;
     FT_MSG_S cmd = { FT_INSERT, msn, xids_get_root_xids(), .u={.id = { &thekey, &theval }} };
+    txn_gc_info gc_info(nullptr, TXNID_NONE, TXNID_NONE, false);
 
-    toku_ft_leaf_apply_cmd(brt->ft->compare_fun, brt->ft->update_fun, &brt->ft->cmp_descriptor, leafnode, -1, &cmd, make_gc_info(false), nullptr, nullptr);
+    toku_ft_leaf_apply_cmd(brt->ft->compare_fun, brt->ft->update_fun, &brt->ft->cmp_descriptor, leafnode, -1, &cmd, &gc_info, nullptr, nullptr);
     {
 	int r = toku_ft_lookup(brt, &thekey, lookup_checkf, &pair);
 	assert(r==0);
@@ -141,7 +142,7 @@ append_leaf(FT_HANDLE brt, FTNODE leafnode, void *key, uint32_t keylen, void *va
     }
 
     FT_MSG_S badcmd = { FT_INSERT, msn, xids_get_root_xids(), .u={.id = { &thekey, &badval }} };
-    toku_ft_leaf_apply_cmd(brt->ft->compare_fun, brt->ft->update_fun, &brt->ft->cmp_descriptor, leafnode, -1, &badcmd, make_gc_info(false), nullptr, nullptr);
+    toku_ft_leaf_apply_cmd(brt->ft->compare_fun, brt->ft->update_fun, &brt->ft->cmp_descriptor, leafnode, -1, &badcmd, &gc_info, nullptr, nullptr);
 
     // message should be rejected for duplicate msn, row should still have original val
     {
@@ -154,7 +155,7 @@ append_leaf(FT_HANDLE brt, FTNODE leafnode, void *key, uint32_t keylen, void *va
     msn = next_dummymsn();
     brt->ft->h->max_msn_in_ft = msn;
     FT_MSG_S cmd2 = { FT_INSERT, msn, xids_get_root_xids(), .u={.id = { &thekey, &val2 }} };
-    toku_ft_leaf_apply_cmd(brt->ft->compare_fun, brt->ft->update_fun, &brt->ft->cmp_descriptor, leafnode, -1, &cmd2, make_gc_info(false), nullptr, nullptr);
+    toku_ft_leaf_apply_cmd(brt->ft->compare_fun, brt->ft->update_fun, &brt->ft->cmp_descriptor, leafnode, -1, &cmd2, &gc_info, nullptr, nullptr);
 
     // message should be accepted, val should have new value
     {
@@ -166,7 +167,7 @@ append_leaf(FT_HANDLE brt, FTNODE leafnode, void *key, uint32_t keylen, void *va
     // now verify that message with lesser (older) msn is rejected
     msn.msn = msn.msn - 10;
     FT_MSG_S cmd3 = { FT_INSERT, msn, xids_get_root_xids(), .u={.id = { &thekey, &badval } }};
-    toku_ft_leaf_apply_cmd(brt->ft->compare_fun, brt->ft->update_fun, &brt->ft->cmp_descriptor, leafnode, -1, &cmd3, make_gc_info(false), nullptr, nullptr);
+    toku_ft_leaf_apply_cmd(brt->ft->compare_fun, brt->ft->update_fun, &brt->ft->cmp_descriptor, leafnode, -1, &cmd3, &gc_info, nullptr, nullptr);
 
     // message should be rejected, val should still have value in pair2
     {

ft/tests/orthopush-flush.cc

@@ -96,6 +96,7 @@ PATENT RIGHTS GRANT:
 static TOKUTXN const null_txn = 0;
 static DB * const null_db = 0;
 static const char *fname = TOKU_TEST_FILENAME;
+static txn_gc_info non_mvcc_gc_info(nullptr, TXNID_NONE, TXNID_NONE, false);
 
 static int dummy_cmp(DB *db __attribute__((unused)),
                      const DBT *a, const DBT *b) {
@@ -217,8 +218,8 @@ insert_random_message_to_bn(
     *keylenp = keydbt->size;
     *keyp = toku_xmemdup(keydbt->data, keydbt->size);
     int64_t numbytes;
-    toku_le_apply_msg(&msg, NULL, NULL, 0, TXNID_NONE, make_gc_info(false), save, &numbytes);
-    toku_ft_bn_apply_cmd(t->ft->compare_fun, t->ft->update_fun, NULL, blb, &msg, TXNID_NONE, make_gc_info(false), NULL, NULL);
+    toku_le_apply_msg(&msg, NULL, NULL, 0, &non_mvcc_gc_info, save, &numbytes);
+    toku_ft_bn_apply_cmd(t->ft->compare_fun, t->ft->update_fun, NULL, blb, &msg, &non_mvcc_gc_info, NULL, NULL);
     if (msn.msn > blb->max_msn_applied.msn) {
         blb->max_msn_applied = msn;
     }
@@ -267,12 +268,12 @@ insert_same_message_to_bns(
     *keylenp = keydbt->size;
     *keyp = toku_xmemdup(keydbt->data, keydbt->size);
     int64_t numbytes;
-    toku_le_apply_msg(&msg, NULL, NULL, 0, TXNID_NONE, make_gc_info(false), save, &numbytes);
-    toku_ft_bn_apply_cmd(t->ft->compare_fun, t->ft->update_fun, NULL, blb1, &msg, TXNID_NONE, make_gc_info(false), NULL, NULL);
+    toku_le_apply_msg(&msg, NULL, NULL, 0, &non_mvcc_gc_info, save, &numbytes);
+    toku_ft_bn_apply_cmd(t->ft->compare_fun, t->ft->update_fun, NULL, blb1, &msg, &non_mvcc_gc_info, NULL, NULL);
     if (msn.msn > blb1->max_msn_applied.msn) {
         blb1->max_msn_applied = msn;
     }
-    toku_ft_bn_apply_cmd(t->ft->compare_fun, t->ft->update_fun, NULL, blb2, &msg, TXNID_NONE, make_gc_info(false), NULL, NULL);
+    toku_ft_bn_apply_cmd(t->ft->compare_fun, t->ft->update_fun, NULL, blb2, &msg, &non_mvcc_gc_info, NULL, NULL);
     if (msn.msn > blb2->max_msn_applied.msn) {
         blb2->max_msn_applied = msn;
     }
@@ -684,7 +685,7 @@ flush_to_leaf(FT_HANDLE t, bool make_leaf_up_to_date, bool use_flush) {
     if (make_leaf_up_to_date) {
         for (i = 0; i < num_parent_messages; ++i) {
             if (!parent_messages_is_fresh[i]) {
-                toku_ft_leaf_apply_cmd(t->ft->compare_fun, t->ft->update_fun, &t->ft->descriptor, child, -1, parent_messages[i], make_gc_info(false), NULL, NULL);
+                toku_ft_leaf_apply_cmd(t->ft->compare_fun, t->ft->update_fun, &t->ft->descriptor, child, -1, parent_messages[i], &non_mvcc_gc_info, NULL, NULL);
             }
         }
         for (i = 0; i < 8; ++i) {
@@ -908,7 +909,7 @@ flush_to_leaf_with_keyrange(FT_HANDLE t, bool make_leaf_up_to_date) {
         for (i = 0; i < num_parent_messages; ++i) {
             if (dummy_cmp(NULL, parent_messages[i]->u.id.key, &childkeys[7]) <= 0 &&
                 !parent_messages_is_fresh[i]) {
-                toku_ft_leaf_apply_cmd(t->ft->compare_fun, t->ft->update_fun, &t->ft->descriptor, child, -1, parent_messages[i], make_gc_info(false), NULL, NULL);
+                toku_ft_leaf_apply_cmd(t->ft->compare_fun, t->ft->update_fun, &t->ft->descriptor, child, -1, parent_messages[i], &non_mvcc_gc_info, NULL, NULL);
             }
         }
         for (i = 0; i < 8; ++i) {
@@ -1104,8 +1105,8 @@ compare_apply_and_flush(FT_HANDLE t, bool make_leaf_up_to_date) {
     if (make_leaf_up_to_date) {
         for (i = 0; i < num_parent_messages; ++i) {
             if (!parent_messages_is_fresh[i]) {
-                toku_ft_leaf_apply_cmd(t->ft->compare_fun, t->ft->update_fun, &t->ft->descriptor, child1, -1, parent_messages[i], make_gc_info(false), NULL, NULL);
-                toku_ft_leaf_apply_cmd(t->ft->compare_fun, t->ft->update_fun, &t->ft->descriptor, child2, -1, parent_messages[i], make_gc_info(false), NULL, NULL);
+                toku_ft_leaf_apply_cmd(t->ft->compare_fun, t->ft->update_fun, &t->ft->descriptor, child1, -1, parent_messages[i], &non_mvcc_gc_info, NULL, NULL);
+                toku_ft_leaf_apply_cmd(t->ft->compare_fun, t->ft->update_fun, &t->ft->descriptor, child2, -1, parent_messages[i], &non_mvcc_gc_info, NULL, NULL);
             }
         }
         for (i = 0; i < 8; ++i) {

ft/tests/test-leafentry-nested.cc

@@ -453,12 +453,12 @@ test_le_apply(ULE ule_initial, FT_MSG msg, ULE ule_expected) {
 
     size_t result_memsize = 0;
     int64_t ignoreme;
+    txn_gc_info gc_info(nullptr, TXNID_NONE, TXNID_NONE, true);
     toku_le_apply_msg(msg,
                       le_initial,
                       nullptr,
                       0,
-                      TXNID_NONE,
-                      make_gc_info(true),
+                      &gc_info,
                       &le_result,
                       &ignoreme);
     if (le_result) {
@@ -751,7 +751,8 @@ static bool ule_worth_running_garbage_collection(ULE ule, TXNID oldest_reference
     LEAFENTRY le;
     int r = le_pack(ule, nullptr, 0, nullptr, 0, 0, &le); CKERR(r);
     invariant_notnull(le);
-    bool worth_running = toku_le_worth_running_garbage_collection(le, oldest_referenced_xid_known);
+    txn_gc_info gc_info(nullptr, oldest_referenced_xid_known, oldest_referenced_xid_known, true);
+    bool worth_running = toku_le_worth_running_garbage_collection(le, &gc_info);
     toku_free(le);
     return worth_running;
 }

ft/tests/verify-bad-msn.cc

@@ -128,7 +128,8 @@ append_leaf(FTNODE leafnode, void *key, size_t keylen, void *val, size_t vallen)
 
     // apply an insert to the leaf node
     FT_MSG_S cmd = { FT_INSERT, msn, xids_get_root_xids(), .u={.id = { &thekey, &theval }} };
-    toku_ft_bn_apply_cmd_once(BLB(leafnode, 0), &cmd, idx, NULL, TXNID_NONE, make_gc_info(false), NULL, NULL);
+    txn_gc_info gc_info(nullptr, TXNID_NONE, TXNID_NONE, false);
+    toku_ft_bn_apply_cmd_once(BLB(leafnode, 0), &cmd, idx, NULL, &gc_info, NULL, NULL);
 
     // Create bad tree (don't do following):
     // leafnode->max_msn_applied_to_node = msn;

ft/tests/verify-bad-pivots.cc

@@ -116,7 +116,8 @@ append_leaf(FTNODE leafnode, void *key, size_t keylen, void *val, size_t vallen)
     // apply an insert to the leaf node
     MSN msn = next_dummymsn();
     FT_MSG_S cmd = { FT_INSERT, msn, xids_get_root_xids(), .u={.id = { &thekey, &theval }} };
-    toku_ft_bn_apply_cmd_once(BLB(leafnode, 0), &cmd, idx, NULL, TXNID_NONE, make_gc_info(false), NULL, NULL);
+    txn_gc_info gc_info(nullptr, TXNID_NONE, TXNID_NONE, false);
+    toku_ft_bn_apply_cmd_once(BLB(leafnode, 0), &cmd, idx, NULL, &gc_info, NULL, NULL);
 
     // dont forget to dirty the node
     leafnode->dirty = 1;

ft/tests/verify-dup-in-leaf.cc

@@ -117,7 +117,8 @@ append_leaf(FTNODE leafnode, void *key, size_t keylen, void *val, size_t vallen)
     // apply an insert to the leaf node
     MSN msn = next_dummymsn();
     FT_MSG_S cmd = { FT_INSERT, msn, xids_get_root_xids(), .u={.id = { &thekey, &theval }} };
-    toku_ft_bn_apply_cmd_once(BLB(leafnode, 0), &cmd, idx, NULL, TXNID_NONE, make_gc_info(false), NULL, NULL);
+    txn_gc_info gc_info(nullptr, TXNID_NONE, TXNID_NONE, false);
+    toku_ft_bn_apply_cmd_once(BLB(leafnode, 0), &cmd, idx, NULL, &gc_info, NULL, NULL);
 
     // dont forget to dirty the node
     leafnode->dirty = 1;

ft/tests/verify-dup-pivots.cc

@@ -116,7 +116,8 @@ append_leaf(FTNODE leafnode, void *key, size_t keylen, void *val, size_t vallen)
     // apply an insert to the leaf node
     MSN msn = next_dummymsn();
     FT_MSG_S cmd = { FT_INSERT, msn, xids_get_root_xids(), .u={.id = { &thekey, &theval }} };
-    toku_ft_bn_apply_cmd_once(BLB(leafnode, 0), &cmd, idx, NULL, TXNID_NONE, make_gc_info(false), NULL, NULL);
+    txn_gc_info gc_info(nullptr, TXNID_NONE, TXNID_NONE, false);
+    toku_ft_bn_apply_cmd_once(BLB(leafnode, 0), &cmd, idx, NULL, &gc_info, NULL, NULL);
 
     // dont forget to dirty the node
     leafnode->dirty = 1;

ft/tests/verify-misrouted-msgs.cc

@@ -117,7 +117,8 @@ append_leaf(FTNODE leafnode, void *key, size_t keylen, void *val, size_t vallen)
     // apply an insert to the leaf node
     MSN msn = next_dummymsn();
     FT_MSG_S cmd = { FT_INSERT, msn, xids_get_root_xids(), .u={.id = { &thekey, &theval }} };
-    toku_ft_bn_apply_cmd_once(BLB(leafnode,0), &cmd, idx, NULL, TXNID_NONE, make_gc_info(false), NULL, NULL);
+    txn_gc_info gc_info(nullptr, TXNID_NONE, TXNID_NONE, false);
+    toku_ft_bn_apply_cmd_once(BLB(leafnode,0), &cmd, idx, NULL, &gc_info, NULL, NULL);
 
     // dont forget to dirty the node
     leafnode->dirty = 1;

ft/tests/verify-unsorted-leaf.cc

@@ -119,7 +119,8 @@ append_leaf(FTNODE leafnode, void *key, size_t keylen, void *val, size_t vallen)
     // apply an insert to the leaf node
     MSN msn = next_dummymsn();
     FT_MSG_S cmd = { FT_INSERT, msn, xids_get_root_xids(), .u={.id = { &thekey, &theval }} };
-    toku_ft_bn_apply_cmd_once(BLB(leafnode, 0), &cmd, idx, NULL, TXNID_NONE, make_gc_info(false), NULL, NULL);
+    txn_gc_info gc_info(nullptr, TXNID_NONE, TXNID_NONE, false);
+    toku_ft_bn_apply_cmd_once(BLB(leafnode, 0), &cmd, idx, NULL, &gc_info, NULL, NULL);
 
     // dont forget to dirty the node
     leafnode->dirty = 1;

ft/tests/verify-unsorted-pivots.cc

@@ -116,7 +116,8 @@ append_leaf(FTNODE leafnode, void *key, size_t keylen, void *val, size_t vallen)
     // apply an insert to the leaf node
     MSN msn = next_dummymsn();
     FT_MSG_S cmd = { FT_INSERT, msn, xids_get_root_xids(), .u={.id = { &thekey, &theval }} };
-    toku_ft_bn_apply_cmd_once(BLB(leafnode, 0), &cmd, idx, NULL, TXNID_NONE, make_gc_info(false), NULL, NULL);
+    txn_gc_info gc_info(nullptr, TXNID_NONE, TXNID_NONE, false);
+    toku_ft_bn_apply_cmd_once(BLB(leafnode, 0), &cmd, idx, NULL, &gc_info, NULL, NULL);
 
     // dont forget to dirty the node
     leafnode->dirty = 1;

ft/txn.cc

@@ -239,7 +239,6 @@ toku_txn_begin_with_xid (
         }
         else {
             parent->child_manager->start_child_txn_for_recovery(txn, parent, xid);
-            txn->oldest_referenced_xid = parent->oldest_referenced_xid;
         }
     }
     else {
@@ -255,7 +254,6 @@ toku_txn_begin_with_xid (
         }
         else {
             parent->child_manager->start_child_txn(txn, parent);
-            txn->oldest_referenced_xid = parent->oldest_referenced_xid;
             toku_txn_manager_handle_snapshot_create_for_child_txn(
                 txn, 
                 logger->txn_manager, 
@@ -327,7 +325,6 @@ static txn_child_manager tcm;
         .container_db_txn = container_db_txn,
         .live_root_txn_list = nullptr,
         .xids = NULL,
-        .oldest_referenced_xid = TXNID_NONE,
         .snapshot_next = NULL,
         .snapshot_prev = NULL,
         .begin_was_logged = false,

ft/txn_manager.cc

@@ -615,7 +615,6 @@ void toku_txn_manager_start_txn_for_recovery(
     // using xid that is passed in
     txn_manager->last_xid = max_xid(txn_manager->last_xid, xid);
     toku_txn_update_xids_in_txn(txn, xid);
-    txn->oldest_referenced_xid = TXNID_NONE;
 
     uint32_t idx;
     int r = txn_manager->live_root_txns.find_zero<TOKUTXN, find_xid>(txn, nullptr, &idx);
@@ -838,7 +837,6 @@ void txn_manager_state::init(TXN_MANAGER txn_manager) {
         &referenced_xids,
         &live_root_txns
         );
-    oldest_referenced_xid_for_simple_gc = txn_manager->last_calculated_oldest_referenced_xid;
 }
 
 void txn_manager_state::destroy() {

ft/txn_manager.h

@@ -130,13 +130,6 @@ struct txn_manager_state {
     rx_omt_t referenced_xids;
     xid_omt_t live_root_txns;
 
-    // the oldest xid in any live list
-    //
-    // suitible for simple garbage collection that cleans up multiple committed
-    // transaction records into one. not suitible for implicit promotions, which
-    // must be correct in the face of abort messages - see ftnode->oldest_referenced_xid
-    TXNID oldest_referenced_xid_for_simple_gc;
-
     txn_manager_state() { }
     void init(TXN_MANAGER txn_manager);
     void destroy();
@@ -145,6 +138,32 @@ private:
     txn_manager_state(txn_manager_state &rhs); // shouldn't need to copy construct
 };
 
+// represents all of the information needed to run garbage collection
+struct txn_gc_info {
+    txn_gc_info(txn_manager_state *st, TXNID xid_sgc, TXNID xid_ip, bool mvcc)
+        : txn_state_for_gc(st),
+          oldest_referenced_xid_for_simple_gc(xid_sgc),
+          oldest_referenced_xid_for_implicit_promotion(xid_ip),
+          mvcc_needed(mvcc) {
+    }
+
+    // a snapshot of the transcation system. may be null.
+    txn_manager_state *const txn_state_for_gc;
+
+    // the oldest xid in any live list
+    //
+    // suitible for simple garbage collection that cleans up multiple committed
+    // transaction records into one. not suitible for implicit promotions, which
+    // must be correct in the face of abort messages - see ftnode->oldest_referenced_xid
+    const TXNID oldest_referenced_xid_for_simple_gc;
+
+    // lower bound on the oldest xid in any live when the messages to be cleaned
+    // had no messages above them. suitable for implicitly promoting a provisonal uxr.
+    const TXNID oldest_referenced_xid_for_implicit_promotion;
+
+    // whether or not mvcc is actually needed - false during recovery and non-transactional systems
+    const bool mvcc_needed;
+};
 
 void toku_txn_manager_init(TXN_MANAGER* txn_manager);
 void toku_txn_manager_destroy(TXN_MANAGER txn_manager);

ft/ule.cc

@@ -321,18 +321,18 @@ xid_reads_committed_xid(TXNID tl1, TXNID xc, const xid_omt_t &snapshot_txnids, c
 // so we get rid of them.
 //
 static void
-ule_simple_garbage_collection(ULE ule, TXNID oldest_referenced_xid, GC_INFO gc_info) {
+ule_simple_garbage_collection(ULE ule, txn_gc_info *gc_info) {
     uint32_t curr_index = 0;
     uint32_t num_entries;
     if (ule->num_cuxrs == 1) {
         goto done;
     }
-    if (gc_info.mvcc_needed) {
+    if (gc_info->mvcc_needed) {
         // starting at the top of the committed stack, find the first
         // uxr with a txnid that is less than oldest_referenced_xid
         for (uint32_t i = 0; i < ule->num_cuxrs; i++) {
             curr_index = ule->num_cuxrs - i - 1;
-            if (ule->uxrs[curr_index].xid < oldest_referenced_xid) {
+            if (ule->uxrs[curr_index].xid < gc_info->oldest_referenced_xid_for_simple_gc) {
                 break;
             }
         }
@@ -484,11 +484,10 @@ toku_le_apply_msg(FT_MSG   msg,
                   LEAFENTRY old_leafentry, // NULL if there was no stored data.
                   bn_data* data_buffer, // bn_data storing leafentry, if NULL, means there is no bn_data
                   uint32_t idx, // index in data_buffer where leafentry is stored (and should be replaced
-                  TXNID oldest_referenced_xid,
-                  GC_INFO gc_info,
-                  txn_manager_state *txn_state_for_gc,
+                  txn_gc_info *gc_info,
                   LEAFENTRY *new_leafentry_p,
                   int64_t * numbytes_delta_p) {  // change in total size of key and val, not including any overhead
+    invariant_notnull(gc_info);
     paranoid_invariant_notnull(new_leafentry_p);
     ULE_S ule;
     int64_t oldnumbytes = 0;
@@ -514,18 +513,14 @@ toku_le_apply_msg(FT_MSG   msg,
 
     // - we may be able to immediately promote the newly-apllied outermost provisonal uxr
     // - either way, run simple gc first, and then full gc if there are still some committed uxrs.
-    ule_try_promote_provisional_outermost(&ule, oldest_referenced_xid);
-    ule_simple_garbage_collection(&ule,
-                                  txn_state_for_gc != nullptr ?
-                                      txn_state_for_gc->oldest_referenced_xid_for_simple_gc :
-                                      oldest_referenced_xid,
-                                  gc_info);
-    if (ule.num_cuxrs > 1 && txn_state_for_gc != nullptr) {
+    ule_try_promote_provisional_outermost(&ule, gc_info->oldest_referenced_xid_for_implicit_promotion);
+    ule_simple_garbage_collection(&ule, gc_info);
+    if (ule.num_cuxrs > 1 && gc_info->txn_state_for_gc != nullptr) {
         size_t size_before_gc = ule_packed_memsize(&ule);
         ule_garbage_collect(&ule,
-                            txn_state_for_gc->snapshot_xids,
-                            txn_state_for_gc->referenced_xids,
-                            txn_state_for_gc->live_root_txns
+                            gc_info->txn_state_for_gc->snapshot_xids,
+                            gc_info->txn_state_for_gc->referenced_xids,
+                            gc_info->txn_state_for_gc->live_root_txns
                             );
         size_t size_after_gc = ule_packed_memsize(&ule);
 
@@ -549,7 +544,7 @@ toku_le_apply_msg(FT_MSG   msg,
     ule_cleanup(&ule);
 }
 
-bool toku_le_worth_running_garbage_collection(LEAFENTRY le, TXNID oldest_referenced_xid_known) {
+bool toku_le_worth_running_garbage_collection(LEAFENTRY le, txn_gc_info *gc_info) {
 // Effect: Quickly determines if it's worth trying to run garbage collection on a leafentry
 // Return: True if it makes sense to try garbage collection, false otherwise.
 // Rationale: Garbage collection is likely to clean up under two circumstances:
@@ -565,7 +560,8 @@ bool toku_le_worth_running_garbage_collection(LEAFENTRY le, TXNID oldest_referen
     } else {
         paranoid_invariant(le->u.mvcc.num_cxrs == 1);
     }
-    return le->u.mvcc.num_pxrs > 0 && le_outermost_uncommitted_xid(le) < oldest_referenced_xid_known;
+    return le->u.mvcc.num_pxrs > 0 &&
+           le_outermost_uncommitted_xid(le) < gc_info->oldest_referenced_xid_for_implicit_promotion;
 }
 
 // Garbage collect one leaf entry, using the given OMT's.
@@ -592,12 +588,12 @@ toku_le_garbage_collect(LEAFENTRY old_leaf_entry,
                         uint32_t idx,
                         void* keyp,
                         uint32_t keylen,
+                        txn_gc_info *gc_info,
                         LEAFENTRY *new_leaf_entry,
-                        const xid_omt_t &snapshot_xids,
-                        const rx_omt_t &referenced_xids,
-                        const xid_omt_t &live_root_txns,
-                        TXNID oldest_referenced_xid_known,
                         int64_t * numbytes_delta_p) {
+    // We shouldn't want to run gc without having provided a snapshot of the txn system.
+    invariant_notnull(gc_info);
+    invariant_notnull(gc_info->txn_state_for_gc);
     paranoid_invariant_notnull(new_leaf_entry);
     ULE_S ule;
     int64_t oldnumbytes = 0;
@@ -621,15 +617,14 @@ toku_le_garbage_collect(LEAFENTRY old_leaf_entry,
     // The oldest known refeferenced xid is a lower bound on the oldest possible
     // live xid, so we use that. It's usually close enough to get rid of most
     // garbage in leafentries.
-    TXNID oldest_possible_live_xid = oldest_referenced_xid_known;
-    ule_try_promote_provisional_outermost(&ule, oldest_possible_live_xid);
+    ule_try_promote_provisional_outermost(&ule, gc_info->oldest_referenced_xid_for_implicit_promotion);
     // No need to run simple gc here if we're going straight for full gc.
     if (ule.num_cuxrs > 1) {
         size_t size_before_gc = ule_packed_memsize(&ule);
         ule_garbage_collect(&ule,
-                            snapshot_xids,
-                            referenced_xids,
-                            live_root_txns);
+                            gc_info->txn_state_for_gc->snapshot_xids,
+                            gc_info->txn_state_for_gc->referenced_xids,
+                            gc_info->txn_state_for_gc->live_root_txns);
         size_t size_after_gc = ule_packed_memsize(&ule);
 
         STATUS_INC(LE_APPLY_GC_BYTES_IN, size_before_gc);

src/indexer-undo-do.cc

@@ -615,7 +615,13 @@ indexer_ft_delete_committed(DB_INDEXER *indexer, DB *hotdb, DBT *hotkey, XIDS xi
     } else {
         result = toku_ydb_check_avail_fs_space(indexer->i->env);
         if (result == 0) {
-            toku_ft_send_delete(db_struct_i(hotdb)->ft_handle, hotkey, xids, TXNID_NONE, make_gc_info(true));
+            TXNID oldest_referenced_xid_estimate =
+                toku_ft_get_oldest_referenced_xid_estimate(db_struct_i(hotdb)->ft_handle);
+            txn_gc_info gc_info(nullptr,
+                                oldest_referenced_xid_estimate,
+                                oldest_referenced_xid_estimate,
+                                true);
+            toku_ft_send_delete(db_struct_i(hotdb)->ft_handle, hotkey, xids, &gc_info);
         }
     }
     return result;
@@ -651,7 +657,13 @@ indexer_ft_insert_committed(DB_INDEXER *indexer, DB *hotdb, DBT *hotkey, DBT *ho
     } else {
         result = toku_ydb_check_avail_fs_space(indexer->i->env);
         if (result == 0) {
-            toku_ft_send_insert(db_struct_i(hotdb)->ft_handle, hotkey, hotval, xids, FT_INSERT, TXNID_NONE, make_gc_info(true));
+            TXNID oldest_referenced_xid_estimate =
+                toku_ft_get_oldest_referenced_xid_estimate(db_struct_i(hotdb)->ft_handle);
+            txn_gc_info gc_info(nullptr,
+                                oldest_referenced_xid_estimate,
+                                oldest_referenced_xid_estimate,
+                                true);
+            toku_ft_send_insert(db_struct_i(hotdb)->ft_handle, hotkey, hotval, xids, FT_INSERT, &gc_info);
         }
     }
     return result;
@@ -670,8 +682,15 @@ indexer_ft_commit(DB_INDEXER *indexer, DB *hotdb, DBT *hotkey, XIDS xids) {
             result = indexer->i->test_commit_any(indexer, hotdb, hotkey, xids);
         } else {
             result = toku_ydb_check_avail_fs_space(indexer->i->env);
-            if (result == 0)
-                toku_ft_send_commit_any(db_struct_i(hotdb)->ft_handle, hotkey, xids, TXNID_NONE, make_gc_info(true));
+            if (result == 0) {
+                TXNID oldest_referenced_xid_estimate =
+                    toku_ft_get_oldest_referenced_xid_estimate(db_struct_i(hotdb)->ft_handle);
+                txn_gc_info gc_info(nullptr,
+                                    oldest_referenced_xid_estimate,
+                                    oldest_referenced_xid_estimate,
+                                    true);
+                toku_ft_send_commit_any(db_struct_i(hotdb)->ft_handle, hotkey, xids, &gc_info);
+            }
         }
     }
     return result;