closes[t:2440] Merge #2440 to main.

closes[t:2512] merge #2512 to main
Delete dev branch

git-svn-id: file:///svn/toku/tokudb@19439 c7de825b-a66e-492c-adef-691d508d4ae1

newbrt/backwards_10.c

@@ -435,6 +435,7 @@ deserialize_brtheader_10 (int fd, struct rbuf *rb, struct brt_header **brth) {
     h->panic_string = 0;
     toku_list_init(&h->live_brts);
     toku_list_init(&h->zombie_brts);
+    toku_list_init(&h->checkpoint_before_commit_link);
     //version MUST be in network order on disk regardless of disk order
     h->layout_version = rbuf_network_int(&rc);
     assert(h->layout_version==BRT_LAYOUT_VERSION_10);

newbrt/brt-internal.h

@@ -180,8 +180,10 @@ struct brt_header {
     // If a transaction locked the BRT when it was empty, which transaction?  (Only the latest one matters)
     // 0 if no such transaction
     TXNID txnid_that_created_or_locked_when_empty;
+    TXNID txnid_that_suppressed_recovery_logs;
     struct toku_list live_brts;
     struct toku_list zombie_brts;
+    struct toku_list checkpoint_before_commit_link;
 };
 
 struct brt {

newbrt/brt-serialize.c

@@ -1372,6 +1372,7 @@ deserialize_brtheader (int fd, struct rbuf *rb, struct brt_header **brth) {
     h->panic_string = 0;
     toku_list_init(&h->live_brts);
     toku_list_init(&h->zombie_brts);
+    toku_list_init(&h->checkpoint_before_commit_link);
     //version MUST be in network order on disk regardless of disk order
     h->layout_version = rbuf_network_int(&rc);
     //TODO: #1924

newbrt/brt.c

@@ -2686,17 +2686,23 @@ toku_brt_log_put_multiple (TOKUTXN txn, BRT src_brt, BRT *brts, int num_brts, co
     TOKULOGGER logger = toku_txn_logger(txn);
     if (logger) {
         FILENUM  fnums[num_brts];
-        FILENUMS filenums = {.num = num_brts, .filenums = fnums};
         int i;
+        int num_unsuppressed_brts = 0;
         for (i = 0; i < num_brts; i++) {
-            fnums[i] = toku_cachefile_filenum(brts[i]->cf);
+            if (brts[i]->h->txnid_that_suppressed_recovery_logs == TXNID_NONE) {
+                //Logging not suppressed for this brt.
+                fnums[num_unsuppressed_brts++] = toku_cachefile_filenum(brts[i]->cf);
             }
+        }
+        if (num_unsuppressed_brts) {
+            FILENUMS filenums = {.num = num_unsuppressed_brts, .filenums = fnums};
             BYTESTRING keybs = {.len=key->size, .data=key->data};
             BYTESTRING valbs = {.len=val->size, .data=val->data};
             TXNID xid = toku_txn_get_txnid(txn);
             FILENUM src_filenum = src_brt ? toku_cachefile_filenum(src_brt->cf) : FILENUM_NONE;
             r = toku_log_enq_insert_multiple(logger, (LSN*)0, 0, src_filenum, filenums, xid, keybs, valbs);
         }
+    }
     return r;
 }
 
@@ -2725,7 +2731,8 @@ int toku_brt_maybe_insert (BRT brt, DBT *key, DBT *val, TOKUTXN txn, BOOL oplsn_
         message_xids = xids_get_root_xids();
     }
     TOKULOGGER logger = toku_txn_logger(txn);
-    if (do_logging && logger) {
+    if (do_logging && logger &&
+        brt->h->txnid_that_suppressed_recovery_logs == TXNID_NONE) {
         BYTESTRING keybs = {.len=key->size, .data=key->data};
         BYTESTRING valbs = {.len=val->size, .data=val->data};
         if (type == BRT_INSERT) {
@@ -2759,17 +2766,23 @@ toku_brt_log_del_multiple (TOKUTXN txn, BRT src_brt, BRT *brts, int num_brts, co
     TOKULOGGER logger = toku_txn_logger(txn);
     if (logger) {
         FILENUM  fnums[num_brts];
-        FILENUMS filenums = {.num = num_brts, .filenums = fnums};
         int i;
+        int num_unsuppressed_brts = 0;
         for (i = 0; i < num_brts; i++) {
-            fnums[i] = toku_cachefile_filenum(brts[i]->cf);
+            if (brts[i]->h->txnid_that_suppressed_recovery_logs == TXNID_NONE) {
+                //Logging not suppressed for this brt.
+                fnums[num_unsuppressed_brts++] = toku_cachefile_filenum(brts[i]->cf);
+            }
         }
+        if (num_unsuppressed_brts) {
+            FILENUMS filenums = {.num = num_unsuppressed_brts, .filenums = fnums};
             BYTESTRING keybs = {.len=key->size, .data=key->data};
             BYTESTRING valbs = {.len=val->size, .data=val->data};
             TXNID xid = toku_txn_get_txnid(txn);
             FILENUM src_filenum = src_brt ? toku_cachefile_filenum(src_brt->cf) : FILENUM_NONE;
             r = toku_log_enq_delete_multiple(logger, (LSN*)0, 0, src_filenum, filenums, xid, keybs, valbs);
         }
+    }
     return r;
 }
 
@@ -2791,7 +2804,8 @@ int toku_brt_maybe_delete(BRT brt, DBT *key, TOKUTXN txn, BOOL oplsn_valid, LSN
         message_xids = xids_get_root_xids();
     }
     TOKULOGGER logger = toku_txn_logger(txn);
-    if (do_logging && logger) {
+    if (do_logging && logger &&
+        brt->h->txnid_that_suppressed_recovery_logs == TXNID_NONE) {
         BYTESTRING keybs = {.len=key->size, .data=key->data};
         r = toku_log_enq_delete_any(logger, (LSN*)0, 0, toku_cachefile_filenum(brt->cf), xid, keybs);
         if (r!=0) return r;
@@ -3022,6 +3036,7 @@ brt_init_header (BRT t) {
 
     toku_list_init(&t->h->live_brts);
     toku_list_init(&t->h->zombie_brts);
+    toku_list_init(&t->h->checkpoint_before_commit_link);
     int r = brt_init_header_partial(t);
     if (r==0) toku_block_verify_no_free_blocknums(t->h->blocktable);
     return r;
@@ -3782,7 +3797,7 @@ brtheader_note_unpin_by_checkpoint (CACHEFILE UU(cachefile), void *header_v)
 // Write checkpoint-in-progress versions of header and translation to disk (really to OS internal buffer).
 // Must have access to fd (protected)
 int
-toku_brtheader_checkpoint (CACHEFILE UU(cachefile), int fd, void *header_v)
+toku_brtheader_checkpoint (CACHEFILE cf, int fd, void *header_v)
 {
     struct brt_header *h = header_v;
     struct brt_header *ch = h->checkpoint_header;
@@ -3794,6 +3809,11 @@ toku_brtheader_checkpoint (CACHEFILE UU(cachefile), int fd, void *header_v)
     if (ch->panic!=0) goto handle_error;
     assert(ch->type == BRTHEADER_CHECKPOINT_INPROGRESS);
     if (ch->dirty) {	// this is only place this bit is tested (in checkpoint_header)
+        TOKULOGGER logger = toku_cachefile_logger(cf);
+        if (logger) {
+            r = toku_logger_fsync_if_lsn_not_fsynced(logger, ch->checkpoint_lsn);
+            if (r!=0) goto handle_error;
+        }
 	{
 	    ch->checkpoint_count++;
 	    // write translation and header to disk (or at least to OS internal buffer)
@@ -5338,7 +5358,7 @@ int toku_brt_maybe_delete_both(BRT brt, DBT *key, DBT *val, TOKUTXN txn, BOOL op
         message_xids = xids_get_root_xids();
     }
     TOKULOGGER logger = toku_txn_logger(txn);
-    if (logger) {
+    if (logger && brt->h->txnid_that_suppressed_recovery_logs == TXNID_NONE) {
         BYTESTRING keybs = {.len=key->size, .data=key->data};
         BYTESTRING valbs = {.len=val->size, .data=val->data};
         r = toku_log_enq_delete_both(logger, (LSN*)0, 0, toku_cachefile_filenum(brt->cf), xid, keybs, valbs);
@@ -5693,25 +5713,18 @@ toku_brt_is_empty (BRT brt) {
     return is_empty;
 }
 
-int
-toku_brt_note_table_lock (BRT brt, TOKUTXN txn, BOOL ignore_not_empty) {
-    int r = 0;
-    if (brt->h->txnid_that_created_or_locked_when_empty != toku_txn_get_txnid(txn) &&
-        (ignore_not_empty || toku_brt_is_empty(brt)) &&
-        brt->h->txnid_that_created_or_locked_when_empty == TXNID_NONE)
-    {
-        brt->h->txnid_that_created_or_locked_when_empty = toku_txn_get_txnid(txn);
-        r = toku_txn_note_brt(txn, brt);
-        assert(r==0);
-        r = toku_logger_save_rollback_tablelock_on_empty_table(txn, toku_cachefile_filenum(brt->cf));
-        if (r==0) {
-            TOKULOGGER logger = toku_txn_logger(txn);
-            TXNID xid = toku_txn_get_txnid(txn);
-            r = toku_log_tablelock_on_empty_table(logger, (LSN*)NULL,
-                                                  0, toku_cachefile_filenum(brt->cf), xid);
-        }
-    }
-    return r;
+//Suppress both rollback and recovery logs.
+void
+toku_brt_suppress_recovery_logs (BRT brt, TOKUTXN txn) {
+    assert(brt->h->txnid_that_created_or_locked_when_empty == toku_txn_get_txnid(txn));
+    assert(brt->h->txnid_that_suppressed_recovery_logs     == TXNID_NONE);
+    brt->h->txnid_that_suppressed_recovery_logs            = toku_txn_get_txnid(txn);
+    toku_list_push(&txn->checkpoint_before_commit, &brt->h->checkpoint_before_commit_link);
+}
+
+BOOL
+toku_brt_is_recovery_logging_suppressed (BRT brt) {
+    return brt->h->txnid_that_suppressed_recovery_logs != TXNID_NONE;
 }
 
 LSN toku_brt_checkpoint_lsn(BRT brt) {

newbrt/brt.h

@@ -206,8 +206,11 @@ void toku_maybe_truncate_cachefile (CACHEFILE cf, int fd, u_int64_t size_used);
 int maybe_preallocate_in_file (int fd, u_int64_t size);
 // Effect: If file size is less than SIZE, make it bigger by either doubling it or growing by 16MB whichever is less.
 
-int toku_brt_note_table_lock (BRT brt, TOKUTXN txn, BOOL ignore_not_empty);
-// Effect: Record the fact that the BRT has a table lock (and thus no other txn will modify it until this txn completes.  As a result, we can limit the amount of information in the rollback data structure.
+void toku_brt_suppress_recovery_logs (BRT brt, TOKUTXN txn);
+// Effect: suppresses recovery logs
+// Requires: this is a (target) redirected brt
+//           implies: txnid_that_created_or_locked_when_empty matches txn 
+//           implies: toku_txn_note_brt(brt, txn) has been called
 
 int toku_brt_zombie_needed (BRT brt);
 
@@ -217,6 +220,7 @@ BOOL toku_brt_is_empty (BRT brt);
 
 double get_tdiff(void) __attribute__((__visibility__("default")));
 
+BOOL toku_brt_is_recovery_logging_suppressed (BRT);
 //TODO: #1485 once we have multiple main threads, restore this code, analyze performance.
 #ifndef TOKU_MULTIPLE_MAIN_THREADS
 #define TOKU_MULTIPLE_MAIN_THREADS 0

newbrt/cachetable.h

@@ -322,4 +322,5 @@ void toku_cachetable_set_env_dir(CACHETABLE ct, char *env_dir);
 char * toku_construct_full_name(int count, ...);
 char * toku_cachetable_get_fname_in_cwd(CACHETABLE ct, const char * fname_in_env);
 
+int toku_cachetable_local_checkpoint_for_commit(CACHETABLE ct, TOKUTXN txn, uint32_t n, CACHEFILE cachefiles[n]);
 #endif

newbrt/log-internal.h

@@ -135,6 +135,7 @@ struct tokutxn {
     uint32_t   current_rollback_hash;
     BOOL       recovered_from_checkpoint;
     ROLLBACK_LOG_NODE pinned_inprogress_rollback_log;
+    struct toku_list checkpoint_before_commit;
 };
 
 struct txninfo {

newbrt/log.h

@@ -13,8 +13,8 @@
 #include "memory.h"
 #include "x1764.h"
 
-typedef void(*voidfp)(void);
-typedef void(*YIELDF)(voidfp, void*);
+typedef void(*voidfp)(void *thunk);
+typedef void(*YIELDF)(voidfp, void *fpthunk, void *yieldthunk);
 struct roll_entry;
 
 #include "logger.h"

newbrt/logformat.c

@@ -77,8 +77,6 @@ const struct logtype rollbacks[] = {
                             {"BLOCKNUM",  "spilled_tail", 0},
                             {"u_int32_t", "spilled_tail_hash", 0},
                             NULLFIELD}},
-    {"tablelock_on_empty_table", 'L', FA{{"FILENUM", "filenum", 0},
-					 NULLFIELD}},
     {"load", 'l', FA{{"BYTESTRING", "old_iname", 0},
                      {"BYTESTRING", "new_iname", 0},
                      NULLFIELD}},
@@ -90,6 +88,7 @@ const struct logtype rollbacks[] = {
 
 const struct logtype logtypes[] = {
     // Records produced by checkpoints
+    {"local_txn_checkpoint", 'c', FA{{"TXNID",      "xid", 0}, NULLFIELD}},
     {"begin_checkpoint", 'x', FA{{"u_int64_t", "timestamp", 0}, NULLFIELD}},
     {"end_checkpoint",   'X', FA{{"TXNID", "xid", 0},           // xid is LSN of begin_checkpoint
 				 {"u_int64_t", "timestamp", 0},
@@ -143,9 +142,6 @@ const struct logtype logtypes[] = {
     {"fdelete", 'U', FA{{"TXNID",      "xid", 0},
 			{"BYTESTRING", "iname", 0},
 			NULLFIELD}},
-    {"tablelock_on_empty_table", 'L', FA{{"FILENUM",    "filenum", 0},
-                           {"TXNID",      "xid", 0},
-                           NULLFIELD}},
     {"enq_insert", 'I', FA{{"FILENUM",    "filenum", 0},
                            {"TXNID",      "xid", 0},
                            {"BYTESTRING", "key", 0},
@@ -432,6 +428,7 @@ generate_log_reader (void) {
     fprintf2(cf, hf, "int toku_log_fread_backward (FILE *infile, struct log_entry *le)");
     fprintf(hf, ";\n");
     fprintf(cf, "{\n");
+    fprintf(cf, "  memset(le, 0, sizeof(*le));\n");
     fprintf(cf, "  {\n    long pos = ftell(infile);\n    if (pos<=12) return -1;\n  }\n");
     fprintf(cf, "  int r = fseek(infile, -4, SEEK_CUR); \n");//              assert(r==0);\n");
     fprintf(cf, "  if (r!=0) return errno;\n");

newbrt/logger.c

@@ -458,6 +458,24 @@ int toku_logger_fsync (TOKULOGGER logger)
     return r;
 }
 
+int
+toku_logger_fsync_if_lsn_not_fsynced (TOKULOGGER logger, LSN lsn) {
+    int r = 0;
+    if (logger->is_panicked) r = EINVAL;
+    else if (logger->write_log_files && logger->fsynced_lsn.lsn < lsn.lsn) {
+        r = ml_lock(&logger->input_lock);        assert(r==0);
+        logger->input_lock_ctr++;
+        r = toku_logger_maybe_fsync(logger, lsn, TRUE);
+        if (r!=0) {
+            toku_logger_panic(logger, r);
+        }
+        else {
+            assert(logger->fsynced_lsn.lsn >= lsn.lsn);
+        }
+    }
+    return r;
+}
+
 void toku_logger_panic (TOKULOGGER logger, int err) {
     logger->panic_errno=err;
     logger->is_panicked=TRUE;

newbrt/logger.h

@@ -21,6 +21,7 @@ int toku_logger_open_rollback(TOKULOGGER logger, CACHETABLE cachetable, BOOL cre
 int toku_logger_close_rollback(TOKULOGGER logger, BOOL recovery_failed);
 
 int toku_logger_fsync (TOKULOGGER logger);
+int toku_logger_fsync_if_lsn_not_fsynced(TOKULOGGER logger, LSN lsn);
 void toku_logger_panic (TOKULOGGER logger, int err);
 int toku_logger_panicked(TOKULOGGER logger);
 int toku_logger_is_open(TOKULOGGER logger);

newbrt/recover.c

@@ -231,8 +231,8 @@ static const char *recover_state(RECOVER_ENV renv) {
 
 // function supplied to transaction commit and abort
 // No yielding is necessary, but it must call the f function if provided.
-static void recover_yield(voidfp f, void *UU(extra)) {
-    if (f) f();
+static void recover_yield(voidfp f, void *fpthunk, void *UU(yieldthunk)) {
+    if (f) f(fpthunk);
 }
 
 static int
@@ -290,6 +290,30 @@ static int internal_recover_fopen_or_fcreate (RECOVER_ENV renv, BOOL must_create
     return 0;
 }
 
+static int toku_recover_local_txn_checkpoint (struct logtype_local_txn_checkpoint *l, RECOVER_ENV UU(renv)) {
+    int r;
+    switch (renv->ss.ss) {
+    case FORWARD_BETWEEN_CHECKPOINT_BEGIN_END:
+    case FORWARD_NEWER_CHECKPOINT_END: {
+        // assert that the transaction exists
+        TOKUTXN txn = NULL;
+        r = toku_txnid2txn(renv->logger, l->xid, &txn);
+        assert(r == 0 && txn != NULL);
+        r = 0;
+        break;
+    }
+    default:
+        assert(0);
+        return 0;
+    }
+    return r;
+}
+
+static int toku_recover_backward_local_txn_checkpoint (struct logtype_local_txn_checkpoint *UU(l), RECOVER_ENV UU(renv)) {
+    // nothing
+    return 0;
+}
+
 static int toku_recover_begin_checkpoint (struct logtype_begin_checkpoint *l, RECOVER_ENV renv) {
     int r;
     switch (renv->ss.ss) {
@@ -549,9 +573,7 @@ static int toku_recover_xcommit (struct logtype_xcommit *l, RECOVER_ENV renv) {
 
     // commit the transaction
     r = toku_txn_commit_with_lsn(txn, TRUE, recover_yield, NULL, l->lsn,
-				 NULL, NULL,
-				 // No need to release locks during recovery.
-				 NULL, NULL, NULL);
+				 NULL, NULL);
     assert(r == 0);
 
     // close the transaction
@@ -709,26 +731,6 @@ static int toku_recover_backward_fdelete (struct logtype_fdelete *UU(l), RECOVER
     return 0;
 }
 
-static int toku_recover_tablelock_on_empty_table(struct logtype_tablelock_on_empty_table *l, RECOVER_ENV renv) {
-    struct file_map_tuple *tuple = NULL;
-    int r = file_map_find(&renv->fmap, l->filenum, &tuple);
-    if (r==0) {
-        //Our work is only if it is open
-        TOKUTXN txn = NULL;
-        r = toku_txnid2txn(renv->logger, l->xid, &txn);
-        assert(r == 0);
-        assert(txn != NULL);
-        r = toku_brt_note_table_lock(tuple->brt, txn, TRUE);
-        assert(r == 0);
-    }
-    return 0;
-}
-
-static int toku_recover_backward_tablelock_on_empty_table(struct logtype_tablelock_on_empty_table *UU(l), RECOVER_ENV UU(renv)) {
-    // nothing
-    return 0;
-}
-
 static int toku_recover_enq_insert (struct logtype_enq_insert *l, RECOVER_ENV renv) {
     int r;
     TOKUTXN txn = NULL;

newbrt/roll.c

@@ -315,7 +315,7 @@ toku_apply_rollinclude (TXNID      xid,
             r = func(txn, item, yield, yieldv, oplsn);
             if (r!=0) return r;
             count++;
-            if (count%2 == 0) yield(NULL, yieldv);
+            if (count%2 == 0) yield(NULL, NULL, yieldv);
         }
         if (next_log.b == spilled_head.b) {
             assert(!found_head);
@@ -382,55 +382,6 @@ toku_rollback_rollinclude (TXNID      xid,
     return r;
 }
 
-int
-toku_rollback_tablelock_on_empty_table (FILENUM filenum,
-                                        TOKUTXN txn,
-                                        YIELDF  yield,
-                                        void*   yield_v,
-                                        LSN     oplsn)
-{
-    //TODO: Replace truncate function with something that doesn't need to mess with checkpoints.
-    // on rollback we have to make the file be empty, since we locked an empty table, and then may have done things to it.
-
-    CACHEFILE cf;
-    //printf("%s:%d committing insert %s %s\n", __FILE__, __LINE__, key.data, data.data);
-    int r = toku_cachefile_of_filenum(txn->logger->ct, filenum, &cf);
-    if (r==ENOENT) { //Missing file on recovered transaction is not an error
-        assert(txn->recovered_from_checkpoint);
-        r = 0;
-        goto done;
-    }
-    assert(r==0);
-
-    OMTVALUE brtv=NULL;
-    r = toku_omt_find_zero(txn->open_brts, find_brt_from_filenum, &filenum, &brtv, NULL, NULL);
-    assert(r==0);
-    BRT brt = brtv;
-    {   //Do NOT truncate the file if
-        //the file already survived the truncate and was checkpointed.
-        LSN treelsn = toku_brt_checkpoint_lsn(brt);
-        if (oplsn.lsn != 0 && oplsn.lsn <= treelsn.lsn) {
-            r = 0;
-            goto done;
-        }
-    }
-    toku_poll_txn_progress_function(txn, FALSE, TRUE);
-    yield(toku_checkpoint_safe_client_lock, yield_v);
-    toku_poll_txn_progress_function(txn, FALSE, FALSE);
-    r = toku_brt_truncate(brt);
-    assert(r==0);
-    toku_checkpoint_safe_client_unlock();
-
-done:
-    return r; 
-}
-
-int
-toku_commit_tablelock_on_empty_table (FILENUM filenum, TOKUTXN txn, YIELDF UU(yield), void* UU(yield_v), LSN UU(oplsn))
-{
-    return do_nothing_with_filenum(txn, filenum);
-}
-
 int
 toku_commit_load (BYTESTRING old_iname,
                   BYTESTRING UU(new_iname),

newbrt/rollback.c

@@ -104,7 +104,7 @@ toku_apply_txn (TOKUTXN txn, YIELDF yield, void*yieldv, LSN lsn,
                 r = func(txn, item, yield, yieldv, lsn);
                 if (r!=0) return r;
                 count++;
-                if (count%2 == 0) yield(NULL, yieldv);
+                if (count%2 == 0) yield(NULL, NULL, yieldv);
             }
         }
         if (next_log.b == txn->spilled_rollback_head.b) {
@@ -206,6 +206,11 @@ static int note_brt_used_in_txns_parent(OMTVALUE brtv, u_int32_t UU(index), void
         //Pass magic "no rollback needed" flag to parent.
         brt->h->txnid_that_created_or_locked_when_empty = toku_txn_get_txnid(parent);
     }
+    if (r==0 &&
+        brt->h->txnid_that_suppressed_recovery_logs == toku_txn_get_txnid(child)) {
+        //Pass magic "no recovery needed" flag to parent.
+        brt->h->txnid_that_suppressed_recovery_logs = toku_txn_get_txnid(parent);
+    }
     return r;
 }
 
@@ -279,6 +284,12 @@ int toku_rollback_commit(TOKUTXN txn, YIELDF yield, void*yieldv, LSN lsn) {
         r = toku_omt_iterate(txn->open_brts, note_brt_used_in_txns_parent, txn);
         assert(r==0);
 
+        // Merge the list of headers that must be checkpointed before commit
+        while (!toku_list_empty(&txn->checkpoint_before_commit)) {
+            struct toku_list *list = toku_list_pop(&txn->checkpoint_before_commit);
+            toku_list_push(&txn->parent->checkpoint_before_commit, list);
+        }
+
         //If this transaction needs an fsync (if it commits)
         //save that in the parent.  Since the commit really happens in the root txn.
         txn->parent->force_fsync_on_commit |= txn->force_fsync_on_commit;
@@ -293,6 +304,11 @@ int toku_rollback_commit(TOKUTXN txn, YIELDF yield, void*yieldv, LSN lsn) {
 
 int toku_rollback_abort(TOKUTXN txn, YIELDF yield, void*yieldv, LSN lsn) {
     int r;
+    //Empty the list
+    while (!toku_list_empty(&txn->checkpoint_before_commit)) {
+        toku_list_pop(&txn->checkpoint_before_commit);
+    }
+
     r = toku_apply_txn(txn, yield, yieldv, lsn, toku_abort_rollback_item);
     assert(r==0);
     return r;
@@ -557,7 +573,10 @@ static int remove_txn (OMTVALUE brtv, u_int32_t UU(idx), void *txnv) {
     r = toku_omt_delete_at(brt->txns, index);
     assert(r==0);
     if (txn->txnid64==brt->h->txnid_that_created_or_locked_when_empty) {
-        brt->h->txnid_that_created_or_locked_when_empty = 0;
+        brt->h->txnid_that_created_or_locked_when_empty = TXNID_NONE;
+    }
+    if (txn->txnid64==brt->h->txnid_that_suppressed_recovery_logs) {
+        brt->h->txnid_that_suppressed_recovery_logs = TXNID_NONE;
     }
     if (!toku_brt_zombie_needed(brt) && brt->was_closed) {
         //Close immediately.

newbrt/rwlock.h

@@ -147,7 +147,13 @@ static inline int rwlock_readers(RWLOCK rwlock) {
     return rwlock->reader;
 }
 
-// returns: the number of writers
+// returns: the number of readers who are waiting for the lock
+
+static inline int rwlock_blocked_readers(RWLOCK rwlock) {
+    return rwlock->want_read;
+}
+
+// returns: the number of writers who are waiting for the lock
 
 static inline int rwlock_blocked_writers(RWLOCK rwlock) {
     return rwlock->want_write;

newbrt/txn.c

@@ -5,6 +5,7 @@
 
 #include "includes.h"
 #include "txn.h"
+#include "checkpoint.h"
 
 int toku_txn_begin_txn (TOKUTXN parent_tokutxn, TOKUTXN *tokutxn, TOKULOGGER logger) {
     return toku_txn_begin_with_xid(parent_tokutxn, tokutxn, logger, 0);
@@ -70,6 +71,7 @@ int toku_txn_begin_with_xid (TOKUTXN parent_tokutxn, TOKUTXN *tokutxn, TOKULOGGE
     result->rollentry_raw_count = 0;
     result->force_fsync_on_commit = FALSE;
     result->recovered_from_checkpoint = FALSE;
+    toku_list_init(&result->checkpoint_before_commit);
     *tokutxn = result;
     return 0;
 
@@ -112,16 +114,52 @@ toku_txn_load_txninfo (TOKUTXN txn, TXNINFO info) {
 
 // Doesn't close the txn, just performs the commit operations.
 int toku_txn_commit_txn(TOKUTXN txn, int nosync, YIELDF yield, void *yieldv,
-                        TXN_PROGRESS_POLL_FUNCTION poll, void *poll_extra,
-			void (*release_locks)(void*), void(*reacquire_locks)(void*), void *locks_thunk) {
+                        TXN_PROGRESS_POLL_FUNCTION poll, void *poll_extra) {
     return toku_txn_commit_with_lsn(txn, nosync, yield, yieldv, ZERO_LSN,
-				    poll, poll_extra,
-				    release_locks, reacquire_locks, locks_thunk);
+				    poll, poll_extra);
+}
+
+struct xcommit_info {
+    int r;
+    TOKUTXN txn;
+    int do_fsync;
+};
+
+//Called during a yield (ydb lock NOT held).
+static void
+local_checkpoints_and_log_xcommit(void *thunk) {
+    struct xcommit_info *info = thunk;
+    TOKUTXN txn = info->txn;
+
+    if (!txn->parent && !toku_list_empty(&txn->checkpoint_before_commit)) {
+        //Do local checkpoints that must happen BEFORE logging xcommit
+        uint32_t num_cachefiles = 0;
+        uint32_t list_size = 16;
+        CACHEFILE *cachefiles= NULL;
+        XMALLOC_N(list_size, cachefiles);
+        while (!toku_list_empty(&txn->checkpoint_before_commit)) {
+            struct toku_list *list = toku_list_pop(&txn->checkpoint_before_commit);
+            struct brt_header *h = toku_list_struct(list,
+                                                    struct brt_header,
+                                                    checkpoint_before_commit_link);
+            cachefiles[num_cachefiles++] = h->cf;
+            if (num_cachefiles == list_size) {
+                list_size *= 2;
+                XREALLOC_N(list_size, cachefiles);
+            }
+        }
+        assert(num_cachefiles);
+        CACHETABLE ct = toku_cachefile_get_cachetable(cachefiles[0]);
+
+        int r = toku_cachetable_local_checkpoint_for_commit(ct, txn, num_cachefiles, cachefiles);
+        assert(r==0);
+    }
+
+    info->r = toku_log_xcommit(txn->logger, (LSN*)0, info->do_fsync, txn->txnid64); // exits holding neither of the tokulogger locks.
 }
 
 int toku_txn_commit_with_lsn(TOKUTXN txn, int nosync, YIELDF yield, void *yieldv, LSN oplsn,
-                             TXN_PROGRESS_POLL_FUNCTION poll, void *poll_extra,
-			     void (*release_locks)(void*), void(*reacquire_locks)(void*), void *locks_thunk) {
+                             TXN_PROGRESS_POLL_FUNCTION poll, void *poll_extra) {
     int r;
     // panic handled in log_commit
 
@@ -131,9 +169,15 @@ int toku_txn_commit_with_lsn(TOKUTXN txn, int nosync, YIELDF yield, void *yieldv
     txn->progress_poll_fun = poll;
     txn->progress_poll_fun_extra = poll_extra;
 
-    if (release_locks) release_locks(locks_thunk);
-    r = toku_log_xcommit(txn->logger, (LSN*)0, do_fsync, txn->txnid64); // exits holding neither of the tokulogger locks.
-    if (reacquire_locks) reacquire_locks(locks_thunk);
+    {
+        struct xcommit_info info = {
+            .r = 0,
+            .txn = txn,
+            .do_fsync = do_fsync
+        };
+        yield(local_checkpoints_and_log_xcommit, &info, yieldv);
+        r = info.r;
+    }
     if (r!=0)
         return r;
     r = toku_rollback_commit(txn, yield, yieldv, oplsn);
@@ -152,6 +196,7 @@ int toku_txn_abort_with_lsn(TOKUTXN txn, YIELDF yield, void *yieldv, LSN oplsn,
     // Must undo everything.  Must undo it all in reverse order.
     // Build the reverse list
     //printf("%s:%d abort\n", __FILE__, __LINE__);
+
     txn->progress_poll_fun = poll;
     txn->progress_poll_fun_extra = poll_extra;
     int r=0;

newbrt/txn.h

@@ -10,11 +10,9 @@ int toku_txn_begin_with_xid (TOKUTXN parent_tokutxn, TOKUTXN *tokutxn, TOKULOGGE
 int toku_txn_load_txninfo (TOKUTXN txn, TXNINFO info);
 
 int toku_txn_commit_txn (TOKUTXN txn, int nosync, YIELDF yield, void *yieldv,
-			 TXN_PROGRESS_POLL_FUNCTION poll, void *poll_extra,
-			 void (*release_locks)(void*), void(*reacquire_locks)(void*), void *locks_thunk);
+			 TXN_PROGRESS_POLL_FUNCTION poll, void *poll_extra);
 int toku_txn_commit_with_lsn(TOKUTXN txn, int nosync, YIELDF yield, void *yieldv, LSN oplsn,
-			     TXN_PROGRESS_POLL_FUNCTION poll, void *poll_extra,
-			     void (*release_locks)(void*), void(*reacquire_locks)(void*), void *locks_thunk);
+			     TXN_PROGRESS_POLL_FUNCTION poll, void *poll_extra);
 
 int toku_txn_abort_txn(TOKUTXN txn, YIELDF yield, void *yieldv,
                        TXN_PROGRESS_POLL_FUNCTION poll, void *poll_extra);

src/loader.c

@@ -71,14 +71,15 @@ static void free_loader_resources(DB_LOADER *loader)
         if (loader->i->ekeys)              toku_free(loader->i->ekeys);
         if (loader->i->evals)              toku_free(loader->i->evals);
 
-        for (int i=0; i<loader->i->N; i++) {
-            if (loader->i->inames_in_env[i]) toku_free(loader->i->inames_in_env[i]);
-        }
-
         if (loader->i->err_key.data)       toku_free(loader->i->err_key.data);
         if (loader->i->err_val.data)       toku_free(loader->i->err_val.data);
 
-        if (loader->i->inames_in_env)      toku_free(loader->i->inames_in_env);
+        if (loader->i->inames_in_env) {
+            for (int i=0; i<loader->i->N; i++) {
+                if (loader->i->inames_in_env[i]) toku_free(loader->i->inames_in_env[i]);
+            }
+            toku_free(loader->i->inames_in_env);
+        }
         if (loader->i->temp_file_template) toku_free(loader->i->temp_file_template);
         if (loader->i->brt_loader)         toku_free(loader->i->brt_loader);
 
@@ -162,7 +163,7 @@ int toku_loader_create_loader(DB_ENV *env,
     int r;
     // lock tables and check empty
     for(int i=0;i<N;i++) {
-        r = toku_db_pre_acquire_table_lock(dbs[i], txn);
+        r = toku_db_pre_acquire_table_lock(dbs[i], txn, TRUE);
         if ( r!=0 ) {
             free_loader(loader);
             return -1;
@@ -200,9 +201,10 @@ int toku_loader_create_loader(DB_ENV *env,
         LSN load_lsn;
         r = locked_ydb_load_inames (env, txn, N, dbs, new_inames_in_env, &load_lsn);
         if ( r!=0 ) {
+            toku_free(new_inames_in_env);
             toku_free(descriptors);
             free_loader(loader);
-            return -1;
+            return r;
         }
         toku_brt_loader_open(&loader->i->brt_loader,
                              loader->i->env->i->cachetable,

src/tests/progress.c

@@ -66,6 +66,7 @@
 static DB_ENV *env = NULL;
 static DB_TXN *txn_parent = NULL;
 static DB_TXN *txn_child  = NULL;
+static DB_TXN *txn_hold_dname_lock  = NULL;
 static DB     *db;
 static char *dname = DICT_0;
 static DBT key;
@@ -108,6 +109,37 @@ end_env(void) {
     env = NULL;
 }
 
+static void
+start_txn_prevent_dname_lock(void) {
+    assert(env!=NULL);
+    assert(txn_hold_dname_lock==NULL);
+    int r;
+    r=env->txn_begin(env, 0, &txn_hold_dname_lock, 0);
+    CKERR(r);
+    DB *db2;
+
+    r = db_create(&db2, env, 0);
+    CKERR(r);
+
+    r=db2->open(db2, txn_hold_dname_lock, dname, 0, DB_BTREE, DB_CREATE, S_IRWXU+S_IRWXG+S_IRWXO);
+    CKERR(r);
+    r = db2->close(db2, 0);
+}
+
+static void nopoll(TOKU_TXN_PROGRESS UU(progress), void *UU(extra)) {
+    assert(FALSE);
+}
+
+static void
+commit_txn_prevent_dname_lock(void) {
+    assert(env!=NULL);
+    assert(txn_hold_dname_lock!=NULL);
+    int r;
+    r = txn_hold_dname_lock->commit_with_progress(txn_hold_dname_lock, 0, nopoll, NULL);
+    CKERR(r);
+    txn_hold_dname_lock = NULL;
+}
+
 static void
 start_txn(void) {
     assert(env!=NULL);
@@ -301,43 +333,29 @@ progress_test_1(int n, int commit) {
     end_env();
 }
 
-struct progress_stall_expect {
-    int  num_calls;
-    BOOL has_been_stalled;
-};
+static void
+abort_txn_stall_checkpoint(void) {
+    //We have disabled the norollback log fallback optimization.
+    //Checkpoint will not stall
+    assert(env!=NULL);
+    assert(txn_parent);
+    assert(!txn_child);
     
-static void stall_poll(TOKU_TXN_PROGRESS progress, void *extra) {
-    struct progress_stall_expect *info = extra;
-    info->num_calls++;
-    assert(info->num_calls <= 2);
-    assert(progress->is_commit == FALSE);
-    if (!info->has_been_stalled) {
-        assert(info->num_calls==1);
-        assert(progress->stalled_on_checkpoint); 
-        info->has_been_stalled = TRUE;
-    }
-    else {
-        assert(info->num_calls==2);
-        assert(!progress->stalled_on_checkpoint); 
-    }
+    int r;
+    r=txn_parent->abort_with_progress(txn_parent, nopoll, NULL);
+    CKERR(r);
+    txn_parent = NULL;
 }
 
-
 static void
-abort_txn_stall_checkpoint(void) {
+abort_txn_nostall_checkpoint(void) {
     assert(env!=NULL);
     assert(txn_parent);
     assert(!txn_child);
     
-    struct progress_stall_expect extra = {
-        .num_calls = 0,
-        .has_been_stalled = FALSE
-    };
-
     int r;
-    r=txn_parent->abort_with_progress(txn_parent, stall_poll, &extra);
+    r=txn_parent->abort_with_progress(txn_parent, nopoll, NULL);
     CKERR(r);
-    assert(extra.num_calls == 2);
     txn_parent = NULL;
 }
 
@@ -358,12 +376,25 @@ progress_test_2(void) {
     start_env();
     open_db();
     start_txn();
+    start_txn_prevent_dname_lock();
     lock();
+    commit_txn_prevent_dname_lock();
     abort_txn_stall_checkpoint();
     close_db();
     end_env();
 }
 
+static void
+progress_test_3(void) {
+    start_env();
+    open_db();
+    start_txn();
+    lock();
+    abort_txn_nostall_checkpoint();
+    close_db();
+    end_env();
+}
+
 int
 test_main (int argc, char * const argv[])
 {
@@ -373,5 +404,6 @@ test_main (int argc, char * const argv[])
         progress_test_1(4, commit);
     }
     progress_test_2();
+    progress_test_3();
     return 0;
 }

src/tests/recover-2483.c

@@ -41,7 +41,7 @@ do_x1_shutdown (void) {
         DB_LOADER *loader;
         DB *dbs[1] = {db};
         uint32_t db_flags[1]  = {DB_NOOVERWRITE};
-        uint32_t dbt_flags[1];
+        uint32_t dbt_flags[1] = {0};
         uint32_t loader_flags = 0;
 
         r = env->create_loader(env, tid, &loader, NULL, 1, dbs, db_flags, dbt_flags, loader_flags);

src/ydb-internal.h

@@ -197,6 +197,6 @@ struct __toku_dbc_internal {
     struct simple_dbt *skey,*sval;
 };
 
-int toku_db_pre_acquire_table_lock(DB *db, DB_TXN *txn);
+int toku_db_pre_acquire_table_lock(DB *db, DB_TXN *txn, BOOL just_lock);
 
 #endif

src/ydb.c

@@ -1830,18 +1830,9 @@ static int toku_txn_release_locks(DB_TXN* txn) {
 
 // Yield the lock so someone else can work, and then reacquire the lock.
 // Useful while processing commit or rollback logs, to allow others to access the system.
-static void ydb_yield (voidfp f, void *UU(v)) {
+static void ydb_yield (voidfp f, void *fv, void *UU(v)) {
     toku_ydb_unlock(); 
-    if (f) f();
-    toku_ydb_lock();
-}
-
-static void release_ydb_lock_callback (void *ignore __attribute__((__unused__))) {
-    //printf("%8.6fs Thread %ld release\n", get_tdiff(), pthread_self());
-    toku_ydb_unlock(); 
-}
-static void reacquire_ydb_lock_callback (void *ignore __attribute__((__unused__))) {
-    //printf("%8.6fs Thread %ld reacquire\n", get_tdiff(), pthread_self());
+    if (f) f(fv);
     toku_ydb_lock();
 }
 
@@ -1886,8 +1877,7 @@ static int toku_txn_commit(DB_TXN * txn, u_int32_t flags,
 	// Calls ydb_yield(NULL) occasionally
         //r = toku_logger_commit(db_txn_struct_i(txn)->tokutxn, nosync, ydb_yield, NULL);
         r = toku_txn_commit_txn(db_txn_struct_i(txn)->tokutxn, nosync, ydb_yield, NULL,
-				poll, poll_extra,
-				release_ydb_lock_callback, reacquire_ydb_lock_callback, NULL);
+				poll, poll_extra);
 
     if (r!=0 && !toku_env_is_panicked(txn->mgrp)) {
         txn->mgrp->i->is_panicked = r;
@@ -4729,7 +4719,7 @@ toku_env_dbremove(DB_ENV * env, DB_TXN *txn, const char *fname, const char *dbna
                 if (r==0) {
                     DB* zombie = env_get_zombie_db_with_dname(env, dname);
                     if (zombie)
-                        r = toku_db_pre_acquire_table_lock(zombie, child);
+                        r = toku_db_pre_acquire_table_lock(zombie, child, TRUE);
                     if (r!=0)
                         toku_ydb_do_error(env, r, "Cannot remove dictionary.\n");
                 }
@@ -4846,7 +4836,7 @@ toku_env_dbrename(DB_ENV *env, DB_TXN *txn, const char *fname, const char *dbnam
             if (r==0) {
                 zombie = env_get_zombie_db_with_dname(env, dname);
                 if (zombie)
-                    r = toku_db_pre_acquire_table_lock(zombie, child);
+                    r = toku_db_pre_acquire_table_lock(zombie, child, TRUE);
                 if (r!=0)
                     toku_ydb_do_error(env, r, "Cannot rename dictionary.\n");
             }
@@ -5041,7 +5031,7 @@ static int toku_db_pre_acquire_read_lock(DB *db, DB_TXN *txn, const DBT *key_lef
 
 //static int toku_db_pre_acquire_table_lock(DB *db, DB_TXN *txn) {
 // needed by loader.c
-int toku_db_pre_acquire_table_lock(DB *db, DB_TXN *txn) {
+int toku_db_pre_acquire_table_lock(DB *db, DB_TXN *txn, BOOL just_lock) {
     HANDLE_PANICKED_DB(db);
     if (!db->i->lt || !txn) return EINVAL;
 
@@ -5053,8 +5043,40 @@ int toku_db_pre_acquire_table_lock(DB *db, DB_TXN *txn) {
     r = toku_lt_acquire_range_write_lock(db->i->lt, db, id_anc,
                                          toku_lt_neg_infinity, toku_lt_neg_infinity,
                                          toku_lt_infinity,     toku_lt_infinity);
-    if (r==0) {
-	r = toku_brt_note_table_lock(db->i->brt, db_txn_struct_i(txn)->tokutxn, FALSE); // tell the BRT layer that the table is locked (so that it can reduce the amount of rollback data.
+    if (r==0 && !just_lock &&
+        !toku_brt_is_recovery_logging_suppressed(db->i->brt) &&
+        toku_brt_is_empty(db->i->brt) &&
+        db_is_nodup(db) //TODO: Remove this check once we kill dupsort support.
+    ) {
+        //Try to suppress both rollback and recovery logs
+        DB_LOADER *loader;
+        DB *dbs[1] = {db};
+        uint32_t db_flags[1]  = {DB_NOOVERWRITE};
+        uint32_t dbt_flags[1] = {0};
+        uint32_t loader_flags = 0;
+        DB_ENV *env = db->dbenv;
+
+        toku_ydb_unlock(); //Cannot hold ydb lock when creating loader
+        int r_loader;
+        r_loader = env->create_loader(env, txn, &loader, NULL, 1, dbs, db_flags, dbt_flags, loader_flags);
+        if (r_loader==0) {
+            int r2;
+            r2 = loader->set_error_callback(loader, NULL, NULL);
+            assert(r2==0);
+            r2 = loader->set_poll_function(loader, NULL, NULL);
+            assert(r2==0);
+            // close the loader
+            r2 = loader->close(loader);
+            assert(r2==0);
+            toku_brt_suppress_recovery_logs(db->i->brt, db_txn_struct_i(txn)->tokutxn);
+        }
+        else if (r_loader != DB_LOCK_NOTGRANTED) {
+            //Lock not granted is not an error.
+            //It just means we cannot use the loader optimization.
+            assert(r==0);
+            r = r_loader;
+        }
+        toku_ydb_lock(); //Reaquire ydb lock.
     }
 
     return r;
@@ -5174,7 +5196,7 @@ static int locked_db_pre_acquire_read_lock(DB *db, DB_TXN *txn, const DBT *key_l
 
 static int locked_db_pre_acquire_table_lock(DB *db, DB_TXN *txn) {
     toku_ydb_lock();
-    int r = toku_db_pre_acquire_table_lock(db, txn);
+    int r = toku_db_pre_acquire_table_lock(db, txn, FALSE);
     toku_ydb_unlock();
     return r;
 }
@@ -5202,7 +5224,7 @@ static int toku_db_truncate(DB *db, DB_TXN *txn, u_int32_t *row_count, u_int32_t
 
     // acquire a table lock
     if (txn) {
-        r = toku_db_pre_acquire_table_lock(db, txn);
+        r = toku_db_pre_acquire_table_lock(db, txn, TRUE);
         if (r != 0)
             return r;
     }
@@ -5601,10 +5623,10 @@ ydb_load_inames(DB_ENV * env, DB_TXN * txn, int N, DB * dbs[N], char * new_iname
 	char hint[strlen(dname) + 1];
 	create_iname_hint(dname, hint);
 	char * new_iname = create_iname(env, xid, hint, i);               // allocates memory for iname_in_env
+	new_inames_in_env[i] = new_iname;
         toku_fill_dbt(&iname_dbt, new_iname, strlen(new_iname) + 1);      // iname_in_env goes in directory
         rval = toku_db_put(env->i->directory, child, &dname_dbt, &iname_dbt, DB_YESOVERWRITE);  // DB_YESOVERWRITE necessary
 	if (rval) break;
-	new_inames_in_env[i] = new_iname;
     }
 
     // Generate load log entries.

toku_include/Makefile.include

@@ -360,17 +360,23 @@ else
 endif
 
 .PHONY: tags
-tags: cscope.out TAGS
+tags: $(TOKUROOT)cscope.out $(TOKUROOT)TAGS
 
+ifneq ($(TOKUROOT),./)
 TAGS: $(TOKUROOT)*/*.[ch] $(TOKUROOT)*/*/*.[ch] $(TOKUROOT)*/*/*/*.[ch]
-	rm -f `find $(TOKUROOT) -type f -name TAGS` #Delete all other tag files
 	etags $(TOKUROOT)*/*.[ch] $(TOKUROOT)*/*/*.[ch] $(TOKUROOT)*/*/*/*.[ch]
+endif
+
+$(TOKUROOT)TAGS: $(TOKUROOT)*/*.[ch] $(TOKUROOT)*/*/*.[ch] $(TOKUROOT)*/*/*/*.[ch]
+	cd $(TOKUROOT) && etags */*.[ch] */*/*.[ch] */*/*/*.[ch]
+
+$(TOKUROOT)cscope.files: $(TOKUROOT)*/*.[ch] $(TOKUROOT)*/*/*.[ch] $(TOKUROOT)*/*/*/*.[ch]
+	cd $(TOKUROOT) && (echo */*.[ch] */*/*.[ch] */*/*/*.[ch] | tr " " "\n") > $(notdir $@)
 
-cscope.files: $(TOKUROOT)*/*.[ch] $(TOKUROOT)*/*/*.[ch] $(TOKUROOT)*/*/*/*.[ch]
-	(echo $(TOKUROOT)*/*.[ch] $(TOKUROOT)*/*/*.[ch] $(TOKUROOT)*/*/*/*.[ch] | tr " " "\n") > $@
+$(TOKUROOT)cscope.in.out $(TOKUROOT)cscope.po.out: $(TOKUROOT)cscope.out;
 
-cscope.out: cscope.files $(TOKUROOT)*/*.[ch] $(TOKUROOT)*/*/*.[ch] $(TOKUROOT)*/*/*/*.[ch]
-	$(CSCOPE) -qb
+$(TOKUROOT)cscope.out: $(TOKUROOT)cscope.files $(TOKUROOT)*/*.[ch] $(TOKUROOT)*/*/*.[ch] $(TOKUROOT)*/*/*/*.[ch]
+	cd $(TOKUROOT) && $(CSCOPE) -qb
 
 .PHONY: clean clean-default %.dir.clean
 clean: clean-default