#27 bigtxn with checkpoints passes the bigtxn27 test

ft/checkpoint.cc

@@ -183,6 +183,7 @@ static LSN last_completed_checkpoint_lsn;
 
 static toku_pthread_rwlock_t checkpoint_safe_lock;
 static toku_pthread_rwlock_t multi_operation_lock;
+static toku_pthread_rwlock_t big_multi_operation_lock;
 
 static bool initialized = false;     // sanity check
 static volatile bool locked_mo = false;       // true when the multi_operation write lock is held (by checkpoint)
@@ -203,7 +204,8 @@ multi_operation_lock_init(void) {
     // TODO: need to figure out how to make writer-preferential rwlocks
     // happen on osx
 #endif
-    toku_pthread_rwlock_init(&multi_operation_lock, &attr); 
+    toku_pthread_rwlock_init(&multi_operation_lock, &attr);
+    toku_pthread_rwlock_init(&big_multi_operation_lock, &attr);
     pthread_rwlockattr_destroy(&attr);
     locked_mo = false;
 }
@@ -211,10 +213,12 @@ multi_operation_lock_init(void) {
 static void
 multi_operation_lock_destroy(void) {
     toku_pthread_rwlock_destroy(&multi_operation_lock);
+    toku_pthread_rwlock_destroy(&big_multi_operation_lock);
 }
 
 static void 
 multi_operation_checkpoint_lock(void) {
+    toku_pthread_rwlock_wrlock(&big_multi_operation_lock);
     toku_pthread_rwlock_wrlock(&multi_operation_lock);   
     locked_mo = true;
 }
@@ -222,7 +226,8 @@ multi_operation_checkpoint_lock(void) {
 static void 
 multi_operation_checkpoint_unlock(void) {
     locked_mo = false;
-    toku_pthread_rwlock_wrunlock(&multi_operation_lock); 
+    toku_pthread_rwlock_wrunlock(&multi_operation_lock);
+    toku_pthread_rwlock_wrunlock(&big_multi_operation_lock);    
 }
 
 static void
@@ -264,6 +269,14 @@ toku_multi_operation_client_unlock(void) {
     toku_pthread_rwlock_rdunlock(&multi_operation_lock); 
 }
 
+void toku_big_multi_operation_client_lock(void) {
+    toku_pthread_rwlock_rdlock(&big_multi_operation_lock);
+}
+
+void toku_big_multi_operation_client_unlock(void) {
+    toku_pthread_rwlock_rdunlock(&big_multi_operation_lock);
+}
+
 void 
 toku_checkpoint_safe_client_lock(void) {
     if (locked_cs)

ft/checkpoint.h

@@ -135,8 +135,10 @@ void toku_checkpoint_safe_client_unlock(void);
  *****/
 
 void toku_multi_operation_client_lock(void);
+void toku_big_multi_operation_client_lock(void);
 
 void toku_multi_operation_client_unlock(void);
+void toku_big_multi_operation_client_unlock(void);
 
 
 // Initialize the checkpoint mechanism, must be called before any client operations.

ft/txn.cc

@@ -786,7 +786,9 @@ void toku_txn_unpin_live_txn(TOKUTXN txn) {
     toku_txn_unlock_state(txn);
 }
 
-
+bool toku_txn_has_spilled_rollback(TOKUTXN txn) {
+    return txn_has_spilled_rollback_logs(txn);
+}
 
 #include <toku_race_tools.h>
 void __attribute__((__constructor__)) toku_txn_status_helgrind_ignore(void);

ft/txn.h

@@ -216,4 +216,6 @@ void toku_txn_unlock_state(TOKUTXN txn);
 void toku_txn_pin_live_txn_unlocked(TOKUTXN txn);
 void toku_txn_unpin_live_txn(TOKUTXN txn);
 
+bool toku_txn_has_spilled_rollback(TOKUTXN txn);
+
 #endif //TOKUTXN_H

src/tests/bigtxn27.cc

@@ -92,8 +92,8 @@ PATENT RIGHTS GRANT:
 #include <pthread.h>
 
 // verify that a commit of a big txn does not block the commits of other txn's
-// commit writer happens before bigtxn commit happens before checkpoint
-static int state = 0;
+// commit writer (0) happens before bigtxn commit (1) happens before checkpoint (2)
+static int test_state = 0;
 
 static void *checkpoint_thread(void *arg) {
     sleep(1);
@@ -102,7 +102,8 @@ static void *checkpoint_thread(void *arg) {
     int r = env->txn_checkpoint(env, 0, 0, 0);
     assert(r == 0);
     printf("%s done\n", __FUNCTION__);
-    assert(toku_sync_fetch_and_add(&state, 1) == 2);
+    int old_state = toku_sync_fetch_and_add(&test_state, 1);
+    assert(old_state == 2);
     return arg;
 }
 
@@ -132,18 +133,25 @@ static void *w_thread(void *arg) {
     r = txn->commit(txn, 0);
     assert(r == 0);
     printf("%s done\n", __FUNCTION__);
-    assert(toku_sync_fetch_and_add(&state, 1) == 0);
+    int old_state = toku_sync_fetch_and_add(&test_state, 1);
+    assert(old_state == 0);
     return arg;
 }
 
 static void bigtxn_progress(TOKU_TXN_PROGRESS progress, void *extra) {
     printf("%s %lu %lu %p\n", __FUNCTION__, progress->entries_processed, progress->entries_total, extra);
-    sleep(10);
+    sleep(1);
 }
 
-int test_main (int argc __attribute__((__unused__)), char *const argv[] __attribute__((__unused__))) {
+int test_main (int argc, char *const argv[]) {
     int r;
-    int N = 10000;
+    int N = 25000;
+    for (int i = 1; i < argc; i++) {
+        if (strcmp(argv[i], "--N") == 0 && i+1 < argc) {
+            N = atoi(argv[++i]);
+            continue;
+        }
+    }
 
     toku_os_recursive_delete(TOKU_TEST_FILENAME);
     r = toku_os_mkdir(TOKU_TEST_FILENAME, S_IRWXU+S_IRWXG+S_IRWXO);
@@ -153,6 +161,10 @@ int test_main (int argc __attribute__((__unused__)), char *const argv[] __attrib
     r = db_env_create(&env, 0);
     assert(r == 0);
 
+    // avoid locktree escalation by picking a big enough lock tree
+    r = env->set_lk_max_memory(env, 128*1024*1024);
+    assert(r == 0);
+
     r = env->open(env, TOKU_TEST_FILENAME, DB_INIT_MPOOL|DB_CREATE|DB_THREAD |DB_INIT_LOCK|DB_INIT_LOG|DB_INIT_TXN|DB_PRIVATE, S_IRWXU+S_IRWXG+S_IRWXO);
     assert(r == 0);
 
@@ -167,27 +179,34 @@ int test_main (int argc __attribute__((__unused__)), char *const argv[] __attrib
     r = env->txn_begin(env, NULL, &bigtxn, 0); 
     assert(r == 0);
 
+    // use a big key so that the rollback log spills
+    char k[1024]; memset(k, 0, sizeof k);
+    char v[8]; memset(v, 0, sizeof v);
+
     for (int i = 0; i < N; i++) {
-        DBT key = { .data = &i, .size = sizeof i };
-        DBT val = { .data = &i, .size = sizeof i };
+        memcpy(k, &i, sizeof i);
+        memcpy(v, &i, sizeof i);
+        DBT key = { .data = k, .size = sizeof k };
+        DBT val = { .data = v, .size = sizeof v };
         r = db->put(db, bigtxn, &key, &val, 0);
         assert(r == 0);
         if ((i % 10000) == 0)
             printf("put %d\n", i);
     }
 
-    pthread_t checkpoint_tid;
+    pthread_t checkpoint_tid = 0;
     r = pthread_create(&checkpoint_tid, NULL, checkpoint_thread, env);
     assert(r == 0);
 
-    pthread_t w_tid;
+    pthread_t w_tid = 0;
     struct writer_arg w_arg = { env, db, N };
     r = pthread_create(&w_tid, NULL, w_thread, &w_arg);
     assert(r == 0);
 
     r = bigtxn->commit_with_progress(bigtxn, 0, bigtxn_progress, NULL);
     assert(r == 0);
-    assert(toku_sync_fetch_and_add(&state, 1) == 1);
+    int old_state = toku_sync_fetch_and_add(&test_state, 1);
+    assert(old_state == 1);
 
     void *ret;
     r = pthread_join(w_tid, &ret);

src/ydb_txn.cc

@@ -136,12 +136,12 @@ toku_txn_destroy(DB_TXN *txn) {
 static int
 toku_txn_commit(DB_TXN * txn, uint32_t flags,
                 TXN_PROGRESS_POLL_FUNCTION poll, void *poll_extra,
-                bool release_mo_lock) {
+                bool release_mo_lock, bool big_txn) {
     HANDLE_PANICKED_ENV(txn->mgrp);
     //Recursively kill off children
     if (db_txn_struct_i(txn)->child) {
         //commit of child sets the child pointer to NULL
-        int r_child = toku_txn_commit(db_txn_struct_i(txn)->child, flags, NULL, NULL, false);
+        int r_child = toku_txn_commit(db_txn_struct_i(txn)->child, flags, NULL, NULL, false, false);
         if (r_child !=0 && !toku_env_is_panicked(txn->mgrp)) {
             env_panic(txn->mgrp, r_child, "Recursive child commit failed during parent commit.\n");
         }
@@ -192,7 +192,11 @@ toku_txn_commit(DB_TXN * txn, uint32_t flags,
     // begin checkpoint logs these associations, so we must be protect
     // the changing of these associations with checkpointing
     if (release_mo_lock) {
-        toku_multi_operation_client_unlock();
+        if (big_txn) {
+            toku_big_multi_operation_client_unlock();
+        } else {
+            toku_multi_operation_client_unlock();
+        }
     }
     toku_txn_maybe_fsync_log(logger, do_fsync_lsn, do_fsync);
     if (flags!=0) {
@@ -218,7 +222,7 @@ toku_txn_abort(DB_TXN * txn,
     //Recursively kill off children (abort or commit are both correct, commit is cheaper)
     if (db_txn_struct_i(txn)->child) {
         //commit of child sets the child pointer to NULL
-        int r_child = toku_txn_commit(db_txn_struct_i(txn)->child, DB_TXN_NOSYNC, NULL, NULL, false);
+        int r_child = toku_txn_commit(db_txn_struct_i(txn)->child, DB_TXN_NOSYNC, NULL, NULL, false, false);
         if (r_child !=0 && !toku_env_is_panicked(txn->mgrp)) {
             env_panic(txn->mgrp, r_child, "Recursive child commit failed during parent abort.\n");
         }
@@ -270,7 +274,7 @@ toku_txn_xa_prepare (DB_TXN *txn, TOKU_XA_XID *xid) {
         //commit of child sets the child pointer to NULL
 
         // toku_txn_commit will take the mo_lock if not held and a non-readonly txn is found.
-        int r_child = toku_txn_commit(db_txn_struct_i(txn)->child, 0, NULL, NULL, false);
+        int r_child = toku_txn_commit(db_txn_struct_i(txn)->child, 0, NULL, NULL, false, false);
         if (r_child !=0 && !toku_env_is_panicked(txn->mgrp)) {
             env_panic(txn->mgrp, r_child, "Recursive child commit failed during parent commit.\n");
         }
@@ -324,18 +328,24 @@ static int
 locked_txn_commit_with_progress(DB_TXN *txn, uint32_t flags,
                                 TXN_PROGRESS_POLL_FUNCTION poll, void* poll_extra) {
     bool holds_mo_lock = false;
-    if (!toku_txn_is_read_only(db_txn_struct_i(txn)->tokutxn)) {
-        // A readonly transaction does no logging, and therefore does not
-        // need the MO lock.
-        toku_multi_operation_client_lock();
+    bool big_txn = false;
+    TOKUTXN tokutxn = db_txn_struct_i(txn)->tokutxn;
+    if (!toku_txn_is_read_only(tokutxn)) {
+        // A readonly transaction does no logging, and therefore does not need the MO lock.
         holds_mo_lock = true;
+        if (toku_txn_has_spilled_rollback(tokutxn)) {
+            big_txn = true;
+            toku_big_multi_operation_client_lock();
+        } else {
+            toku_multi_operation_client_lock();
+        }
     }
     // cannot begin a checkpoint.
     // the multi operation lock is taken the first time we
     // see a non-readonly txn in the recursive commit.
     // But released in the first-level toku_txn_commit (if taken),
     // this way, we don't hold it while we fsync the log.
-    int r = toku_txn_commit(txn, flags, poll, poll_extra, holds_mo_lock);
+    int r = toku_txn_commit(txn, flags, poll, poll_extra, holds_mo_lock, big_txn);
     return r;
 }
 
@@ -347,15 +357,25 @@ locked_txn_abort_with_progress(DB_TXN *txn,
     // see a non-readonly txn in the abort (or recursive commit).
     // But released here so we don't have to hold additional state.
     bool holds_mo_lock = false;
-    if (!toku_txn_is_read_only(db_txn_struct_i(txn)->tokutxn)) {
-        // A readonly transaction does no logging, and therefore does not
-        // need the MO lock.
-        toku_multi_operation_client_lock();
+    bool big_txn = false;
+    TOKUTXN tokutxn = db_txn_struct_i(txn)->tokutxn;
+    if (!toku_txn_is_read_only(tokutxn)) {
+        // A readonly transaction does no logging, and therefore does not need the MO lock.
         holds_mo_lock = true;
+        if (toku_txn_has_spilled_rollback(tokutxn)) {
+            big_txn = true;
+            toku_big_multi_operation_client_lock();
+        } else {
+            toku_multi_operation_client_lock();
+        }
     }
     int r = toku_txn_abort(txn, poll, poll_extra);
     if (holds_mo_lock) {
-        toku_multi_operation_client_unlock();
+        if (big_txn) {
+            toku_big_multi_operation_client_unlock();
+        } else {
+            toku_multi_operation_client_unlock();
+        }
     }
     return r;
 }