path: root/storage/tokudb/PerconaFT/src/ydb_write.cc

/* -*- mode: C++; c-basic-offset: 4; indent-tabs-mode: nil -*- */
// vim: ft=cpp:expandtab:ts=8:sw=4:softtabstop=4:
#ident "$Id$"
/*======
This file is part of PerconaFT.


Copyright (c) 2006, 2015, Percona and/or its affiliates. All rights reserved.

    PerconaFT is free software: you can redistribute it and/or modify
    it under the terms of the GNU General Public License, version 2,
    as published by the Free Software Foundation.

    PerconaFT is distributed in the hope that it will be useful,
    but WITHOUT ANY WARRANTY; without even the implied warranty of
    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
    GNU General Public License for more details.

    You should have received a copy of the GNU General Public License
    along with PerconaFT.  If not, see <http://www.gnu.org/licenses/>.

----------------------------------------

    PerconaFT is free software: you can redistribute it and/or modify
    it under the terms of the GNU Affero General Public License, version 3,
    as published by the Free Software Foundation.

    PerconaFT is distributed in the hope that it will be useful,
    but WITHOUT ANY WARRANTY; without even the implied warranty of
    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
    GNU Affero General Public License for more details.

    You should have received a copy of the GNU Affero General Public License
    along with PerconaFT.  If not, see <http://www.gnu.org/licenses/>.
======= */

#ident "Copyright (c) 2006, 2015, Percona and/or its affiliates. All rights reserved."

#include <db.h>
#include "ydb-internal.h"
#include "indexer.h"
#include <ft/log_header.h>
#include <ft/cachetable/checkpoint.h>
#include "ydb_row_lock.h"
#include "ydb_write.h"
#include "ydb_db.h"
#include <portability/toku_atomic.h>
#include <util/status.h>

static YDB_WRITE_LAYER_STATUS_S ydb_write_layer_status;
#ifdef STATUS_VALUE
#undef STATUS_VALUE
#endif
#define STATUS_VALUE(x) ydb_write_layer_status.status[x].value.num

#define STATUS_INIT(k,c,t,l,inc) TOKUFT_STATUS_INIT(ydb_write_layer_status, k, c, t, l, inc)

static void
ydb_write_layer_status_init (void) {
    // Note, this function initializes the keyname, type, and legend fields.
    // Value fields are initialized to zero by compiler.
    STATUS_INIT(YDB_LAYER_NUM_INSERTS,                nullptr, UINT64,   "dictionary inserts", TOKU_ENGINE_STATUS);
    STATUS_INIT(YDB_LAYER_NUM_INSERTS_FAIL,           nullptr, UINT64,   "dictionary inserts fail", TOKU_ENGINE_STATUS);
    STATUS_INIT(YDB_LAYER_NUM_DELETES,                nullptr, UINT64,   "dictionary deletes", TOKU_ENGINE_STATUS);
    STATUS_INIT(YDB_LAYER_NUM_DELETES_FAIL,           nullptr, UINT64,   "dictionary deletes fail", TOKU_ENGINE_STATUS);
    STATUS_INIT(YDB_LAYER_NUM_UPDATES,                nullptr, UINT64,   "dictionary updates", TOKU_ENGINE_STATUS);
    STATUS_INIT(YDB_LAYER_NUM_UPDATES_FAIL,           nullptr, UINT64,   "dictionary updates fail", TOKU_ENGINE_STATUS);
    STATUS_INIT(YDB_LAYER_NUM_UPDATES_BROADCAST,      nullptr, UINT64,   "dictionary broadcast updates", TOKU_ENGINE_STATUS);
    STATUS_INIT(YDB_LAYER_NUM_UPDATES_BROADCAST_FAIL, nullptr, UINT64,   "dictionary broadcast updates fail", TOKU_ENGINE_STATUS);
    STATUS_INIT(YDB_LAYER_NUM_MULTI_INSERTS,          nullptr, UINT64,   "dictionary multi inserts", TOKU_ENGINE_STATUS);
    STATUS_INIT(YDB_LAYER_NUM_MULTI_INSERTS_FAIL,     nullptr, UINT64,   "dictionary multi inserts fail", TOKU_ENGINE_STATUS);
    STATUS_INIT(YDB_LAYER_NUM_MULTI_DELETES,          nullptr, UINT64,   "dictionary multi deletes", TOKU_ENGINE_STATUS);
    STATUS_INIT(YDB_LAYER_NUM_MULTI_DELETES_FAIL,     nullptr, UINT64,   "dictionary multi deletes fail", TOKU_ENGINE_STATUS);
    STATUS_INIT(YDB_LAYER_NUM_MULTI_UPDATES,          nullptr, UINT64,   "dictionary updates multi", TOKU_ENGINE_STATUS);
    STATUS_INIT(YDB_LAYER_NUM_MULTI_UPDATES_FAIL,     nullptr, UINT64,   "dictionary updates multi fail", TOKU_ENGINE_STATUS);
    ydb_write_layer_status.initialized = true;
}
#undef STATUS_INIT

void
ydb_write_layer_get_status(YDB_WRITE_LAYER_STATUS statp) {
    if (!ydb_write_layer_status.initialized)
        ydb_write_layer_status_init();
    *statp = ydb_write_layer_status;
}


static inline uint32_t 
get_prelocked_flags(uint32_t flags) {
    uint32_t lock_flags = flags & (DB_PRELOCKED | DB_PRELOCKED_WRITE);
    return lock_flags;
}

// these next two static functions are defined
// both here and ydb.c. We should find a good
// place for them.
static int
ydb_getf_do_nothing(DBT const* UU(key), DBT const* UU(val), void* UU(extra)) {
    return 0;
}

// Check if the available file system space is less than the reserve
// Returns ENOSPC if not enough space, othersize 0
static inline int 
env_check_avail_fs_space(DB_ENV *env) {
    int r = env->i->fs_state == FS_RED ? ENOSPC : 0; 
    if (r) {
        env->i->enospc_redzone_ctr++;
    }
    return r;
}

// Return 0 if proposed pair do not violate size constraints of DB
// (insertion is legal)
// Return non zero otherwise.
static int
db_put_check_size_constraints(DB *db, const DBT *key, const DBT *val) {
    int r = 0;
    unsigned int klimit, vlimit;

    toku_ft_get_maximum_advised_key_value_lengths(&klimit, &vlimit);
    if (key->size > klimit) {
        r = toku_ydb_do_error(db->dbenv, EINVAL, 
                "The largest key allowed is %u bytes", klimit);
    } else if (val->size > vlimit) {
        r = toku_ydb_do_error(db->dbenv, EINVAL, 
                "The largest value allowed is %u bytes", vlimit);
    }
    return r;
}

//Return 0 if insert is legal
static int
db_put_check_overwrite_constraint(DB *db, DB_TXN *txn, DBT *key,
                                  uint32_t lock_flags, uint32_t overwrite_flag) {
    int r;

    if (overwrite_flag == 0) { // 0 (yesoverwrite) does not impose constraints.
        r = 0;
    } else if (overwrite_flag == DB_NOOVERWRITE) {
        // Check if (key,anything) exists in dictionary.
        // If exists, fail.  Otherwise, do insert.
        // The DB_RMW flag causes the cursor to grab a write lock instead of a read lock on the key if it exists.
        r = db_getf_set(db, txn, lock_flags|DB_SERIALIZABLE|DB_RMW, key, ydb_getf_do_nothing, NULL);
        if (r == DB_NOTFOUND) 
            r = 0;
        else if (r == 0)      
            r = DB_KEYEXIST;
        //Any other error is passed through.
    } else if (overwrite_flag == DB_NOOVERWRITE_NO_ERROR) {
        r = 0;
    } else {
        //Other flags are not (yet) supported.
        r = EINVAL;
    }
    return r;
}


int
toku_db_del(DB *db, DB_TXN *txn, DBT *key, uint32_t flags, bool holds_mo_lock) {
    HANDLE_PANICKED_DB(db);
    HANDLE_DB_ILLEGAL_WORKING_PARENT_TXN(db, txn);
    HANDLE_READ_ONLY_TXN(txn);

    uint32_t unchecked_flags = flags;
    //DB_DELETE_ANY means delete regardless of whether it exists in the db.
    bool error_if_missing = (bool)(!(flags&DB_DELETE_ANY));
    unchecked_flags &= ~DB_DELETE_ANY;
    uint32_t lock_flags = get_prelocked_flags(flags);
    unchecked_flags &= ~lock_flags;
    bool do_locking = (bool)(db->i->lt && !(lock_flags&DB_PRELOCKED_WRITE));

    int r = 0;
    if (unchecked_flags!=0) {
        r = EINVAL;
    }

    if (r == 0 && error_if_missing) {
        //Check if the key exists in the db.
        r = db_getf_set(db, txn, lock_flags|DB_SERIALIZABLE|DB_RMW, key, ydb_getf_do_nothing, NULL);
    }
    if (r == 0 && do_locking) {
        //Do locking if necessary.
        r = toku_db_get_point_write_lock(db, txn, key);
    }
    if (r == 0) {
        //Do the actual deleting.
        if (!holds_mo_lock) toku_multi_operation_client_lock();
        toku_ft_delete(db->i->ft_handle, key, txn ? db_txn_struct_i(txn)->tokutxn : 0);
        if (!holds_mo_lock) toku_multi_operation_client_unlock();
    }

    if (r == 0) {
        STATUS_VALUE(YDB_LAYER_NUM_DELETES)++;  // accountability 
    }
    else {
        STATUS_VALUE(YDB_LAYER_NUM_DELETES_FAIL)++;  // accountability 
    }
    return r;
}

static int
db_put(DB *db, DB_TXN *txn, DBT *key, DBT *val, int flags, bool do_log) {
    int r = 0;
    bool unique = false;
    enum ft_msg_type type = FT_INSERT;
    if (flags == DB_NOOVERWRITE) {
        unique = true;
    } else if (flags == DB_NOOVERWRITE_NO_ERROR) {
        type = FT_INSERT_NO_OVERWRITE;
    } else if (flags != 0) {
        // All other non-zero flags are unsupported
        r = EINVAL;
    }
    if (r == 0) {
        TOKUTXN ttxn = txn ? db_txn_struct_i(txn)->tokutxn : nullptr;
        if (unique) {
            r = toku_ft_insert_unique(db->i->ft_handle, key, val, ttxn, do_log);
        } else {
            toku_ft_maybe_insert(db->i->ft_handle, key, val, ttxn, false, ZERO_LSN, do_log, type);
        }
        invariant(r == DB_KEYEXIST || r == 0);
    }
    return r;
}

int
toku_db_put(DB *db, DB_TXN *txn, DBT *key, DBT *val, uint32_t flags, bool holds_mo_lock) {
    HANDLE_PANICKED_DB(db);
    HANDLE_DB_ILLEGAL_WORKING_PARENT_TXN(db, txn);
    HANDLE_READ_ONLY_TXN(txn);
    int r = 0;

    uint32_t lock_flags = get_prelocked_flags(flags);
    flags &= ~lock_flags;

    r = db_put_check_size_constraints(db, key, val);

    //Do locking if necessary.
    bool do_locking = (bool)(db->i->lt && !(lock_flags&DB_PRELOCKED_WRITE));
    if (r == 0 && do_locking) {
        r = toku_db_get_point_write_lock(db, txn, key);
    }
    if (r == 0) {
        //Insert into the ft.
        if (!holds_mo_lock) toku_multi_operation_client_lock();
        r = db_put(db, txn, key, val, flags, true);
        if (!holds_mo_lock) toku_multi_operation_client_unlock();
    }

    if (r == 0) {
        // helgrind flags a race on this status update.  we increment it atomically to satisfy helgrind.
        // STATUS_VALUE(YDB_LAYER_NUM_INSERTS)++;  // accountability 
        (void) toku_sync_fetch_and_add(&STATUS_VALUE(YDB_LAYER_NUM_INSERTS), 1);
    } else {
        // STATUS_VALUE(YDB_LAYER_NUM_INSERTS_FAIL)++;  // accountability 
        (void) toku_sync_fetch_and_add(&STATUS_VALUE(YDB_LAYER_NUM_INSERTS_FAIL), 1);
    }

    return r;
}

static int
toku_db_update(DB *db, DB_TXN *txn,
               const DBT *key,
               const DBT *update_function_extra,
               uint32_t flags) {
    HANDLE_PANICKED_DB(db);
    HANDLE_DB_ILLEGAL_WORKING_PARENT_TXN(db, txn);
    HANDLE_READ_ONLY_TXN(txn);
    int r = 0;

    uint32_t lock_flags = get_prelocked_flags(flags);
    flags &= ~lock_flags;

    r = db_put_check_size_constraints(db, key, update_function_extra);
    if (r != 0) { goto cleanup; }

    bool do_locking;
    do_locking = (db->i->lt && !(lock_flags & DB_PRELOCKED_WRITE));
    if (do_locking) {
        r = toku_db_get_point_write_lock(db, txn, key);
        if (r != 0) { goto cleanup; }
    }

    TOKUTXN ttxn;
    ttxn = txn ? db_txn_struct_i(txn)->tokutxn : NULL;
    toku_multi_operation_client_lock();
    toku_ft_maybe_update(db->i->ft_handle, key, update_function_extra, ttxn,
                              false, ZERO_LSN, true);
    toku_multi_operation_client_unlock();

cleanup:
    if (r == 0) 
        STATUS_VALUE(YDB_LAYER_NUM_UPDATES)++;  // accountability 
    else
        STATUS_VALUE(YDB_LAYER_NUM_UPDATES_FAIL)++;  // accountability 
    return r;
}


// DB_IS_RESETTING_OP is true if the dictionary should be considered as if created by this transaction.
// For example, it will be true if toku_db_update_broadcast() is used to implement a schema change (such
// as adding a column), and will be false if used simply to update all the rows of a table (such as 
// incrementing a field).
static int
toku_db_update_broadcast(DB *db, DB_TXN *txn,
                         const DBT *update_function_extra,
                         uint32_t flags) {
    HANDLE_PANICKED_DB(db);
    HANDLE_DB_ILLEGAL_WORKING_PARENT_TXN(db, txn);
    HANDLE_READ_ONLY_TXN(txn);
    int r = 0;

    uint32_t lock_flags = get_prelocked_flags(flags);
    flags &= ~lock_flags;
    uint32_t is_resetting_op_flag = flags & DB_IS_RESETTING_OP;
    flags &= is_resetting_op_flag;
    bool is_resetting_op = (is_resetting_op_flag != 0);
    

    if (is_resetting_op) {
        if (txn->parent != NULL) {
            r = EINVAL; // cannot have a parent if you are a resetting op
            goto cleanup;
        }
        r = toku_db_pre_acquire_fileops_lock(db, txn);
        if (r != 0) { goto cleanup; }
    }
    {
        DBT null_key;
        toku_init_dbt(&null_key);
        r = db_put_check_size_constraints(db, &null_key, update_function_extra);
        if (r != 0) { goto cleanup; }
    }

    bool do_locking;
    do_locking = (db->i->lt && !(lock_flags & DB_PRELOCKED_WRITE));
    if (do_locking) {
        r = toku_db_pre_acquire_table_lock(db, txn);
        if (r != 0) { goto cleanup; }
    }

    TOKUTXN ttxn;
    ttxn = txn ? db_txn_struct_i(txn)->tokutxn : NULL;
    toku_multi_operation_client_lock();
    toku_ft_maybe_update_broadcast(db->i->ft_handle, update_function_extra, ttxn,
                                        false, ZERO_LSN, true, is_resetting_op);
    toku_multi_operation_client_unlock();

cleanup:
    if (r == 0) 
        STATUS_VALUE(YDB_LAYER_NUM_UPDATES_BROADCAST)++;  // accountability 
    else
        STATUS_VALUE(YDB_LAYER_NUM_UPDATES_BROADCAST_FAIL)++;  // accountability 
    return r;
}

static void
log_del_single(DB_TXN *txn, FT_HANDLE ft_handle, const DBT *key) {
    TOKUTXN ttxn = db_txn_struct_i(txn)->tokutxn;
    toku_ft_log_del(ttxn, ft_handle, key);
}

static uint32_t
sum_size(uint32_t num_arrays, DBT_ARRAY keys[], uint32_t overhead) {
    uint32_t sum = 0;
    for (uint32_t i = 0; i < num_arrays; i++) {
        for (uint32_t j = 0; j < keys[i].size; j++) {
            sum += keys[i].dbts[j].size + overhead;
        }
    }
    return sum;
}

static void
log_del_multiple(DB_TXN *txn, DB *src_db, const DBT *key, const DBT *val, uint32_t num_dbs, FT_HANDLE fts[], DBT_ARRAY keys[]) {
    if (num_dbs > 0) {
        TOKUTXN ttxn = db_txn_struct_i(txn)->tokutxn;
        FT_HANDLE src_ft  = src_db ? src_db->i->ft_handle : NULL;
        uint32_t del_multiple_size = key->size + val->size + num_dbs*sizeof (uint32_t) + toku_log_enq_delete_multiple_overhead;
        uint32_t del_single_sizes = sum_size(num_dbs, keys, toku_log_enq_delete_any_overhead);
        if (del_single_sizes < del_multiple_size) {
            for (uint32_t i = 0; i < num_dbs; i++) {
                for (uint32_t j = 0; j < keys[i].size; j++) {
                    log_del_single(txn, fts[i], &keys[i].dbts[j]);
                }
            }
        } else {
            toku_ft_log_del_multiple(ttxn, src_ft, fts, num_dbs, key, val);
        }
    }
}

static uint32_t 
lookup_src_db(uint32_t num_dbs, DB *db_array[], DB *src_db) {
    uint32_t which_db;
    for (which_db = 0; which_db < num_dbs; which_db++) 
        if (db_array[which_db] == src_db)
            break;
    return which_db;
}

static int
do_del_multiple(DB_TXN *txn, uint32_t num_dbs, DB *db_array[], DBT_ARRAY keys[], DB *src_db, const DBT *src_key, bool indexer_shortcut) {
    int r = 0;
    TOKUTXN ttxn = db_txn_struct_i(txn)->tokutxn;
    for (uint32_t which_db = 0; r == 0 && which_db < num_dbs; which_db++) {
        DB *db = db_array[which_db];

        paranoid_invariant(keys[which_db].size <= keys[which_db].capacity);

        // if db is being indexed by an indexer, then insert a delete message into the db if the src key is to the left or equal to the 
        // indexers cursor.  we have to get the src_db from the indexer and find it in the db_array.
        int do_delete = true;
        DB_INDEXER *indexer = toku_db_get_indexer(db);
        if (indexer && !indexer_shortcut) { // if this db is the index under construction
            DB *indexer_src_db = toku_indexer_get_src_db(indexer);
            invariant(indexer_src_db != NULL);
            const DBT *indexer_src_key;
            if (src_db == indexer_src_db)
                indexer_src_key = src_key;
            else {
                uint32_t which_src_db = lookup_src_db(num_dbs, db_array, indexer_src_db);
                invariant(which_src_db < num_dbs);
                // The indexer src db must have exactly one item or we don't know how to continue.
                invariant(keys[which_src_db].size == 1);
                indexer_src_key = &keys[which_src_db].dbts[0];
            }
            do_delete = toku_indexer_should_insert_key(indexer, indexer_src_key);
            toku_indexer_update_estimate(indexer);
        }
        if (do_delete) {
            for (uint32_t i = 0; i < keys[which_db].size; i++) {
                toku_ft_maybe_delete(db->i->ft_handle, &keys[which_db].dbts[i], ttxn, false, ZERO_LSN, false);
            }
        }
    }
    return r;
}

//
// if a hot index is in progress, gets the indexer
// also verifies that there is at most one hot index
// in progress. If it finds more than one, then returns EINVAL
// 
static int
get_indexer_if_exists(
    uint32_t num_dbs, 
    DB **db_array, 
    DB *src_db,
    DB_INDEXER** indexerp,
    bool *src_db_is_indexer_src
    ) 
{
    int r = 0;
    DB_INDEXER* first_indexer = NULL;
    for (uint32_t i = 0; i < num_dbs; i++) {
        DB_INDEXER* indexer = toku_db_get_indexer(db_array[i]);
        if (indexer) {
            if (!first_indexer) {
                first_indexer = indexer;
            }
            else if (first_indexer != indexer) {
                r = EINVAL;
            }
        }
    }
    if (r == 0) {
        if (first_indexer) {
            DB* indexer_src_db = toku_indexer_get_src_db(first_indexer);
            // we should just make this an invariant
            if (src_db == indexer_src_db) {
                *src_db_is_indexer_src = true;
            }
        }
        *indexerp = first_indexer;
    }
    return r;
}

int
env_del_multiple(
    DB_ENV *env, 
    DB *src_db, 
    DB_TXN *txn, 
    const DBT *src_key, 
    const DBT *src_val, 
    uint32_t num_dbs, 
    DB **db_array, 
    DBT_ARRAY *keys,
    uint32_t *flags_array) 
{
    int r;
    DBT_ARRAY del_keys[num_dbs];
    DB_INDEXER* indexer = NULL;

    HANDLE_PANICKED_ENV(env);
    HANDLE_READ_ONLY_TXN(txn);

    uint32_t lock_flags[num_dbs];
    uint32_t remaining_flags[num_dbs];
    FT_HANDLE fts[num_dbs];
    bool indexer_lock_taken = false;
    bool src_same = false;
    bool indexer_shortcut = false;
    if (!txn) {
        r = EINVAL;
        goto cleanup;
    }
    if (!env->i->generate_row_for_del) {
        r = EINVAL;
        goto cleanup;
    }

    HANDLE_ILLEGAL_WORKING_PARENT_TXN(env, txn);
    r = get_indexer_if_exists(num_dbs, db_array, src_db, &indexer, &src_same);
    if (r) {
        goto cleanup;
    }

    for (uint32_t which_db = 0; which_db < num_dbs; which_db++) {
        DB *db = db_array[which_db];
        lock_flags[which_db] = get_prelocked_flags(flags_array[which_db]);
        remaining_flags[which_db] = flags_array[which_db] & ~lock_flags[which_db];

        if (db == src_db) {
            del_keys[which_db].size = 1;
            del_keys[which_db].capacity = 1;
            del_keys[which_db].dbts = const_cast<DBT*>(src_key);
        }
        else {
            //Generate the key
            r = env->i->generate_row_for_del(db, src_db, &keys[which_db], src_key, src_val);
            if (r != 0) goto cleanup;
            del_keys[which_db] = keys[which_db];
            paranoid_invariant(del_keys[which_db].size <= del_keys[which_db].capacity);
        }

        if (remaining_flags[which_db] & ~DB_DELETE_ANY) {
            r = EINVAL;
            goto cleanup;
        }
        bool error_if_missing = (bool)(!(remaining_flags[which_db]&DB_DELETE_ANY));
        for (uint32_t which_key = 0; which_key < del_keys[which_db].size; which_key++) {
            DBT *del_key = &del_keys[which_db].dbts[which_key];
            if (error_if_missing) {
                //Check if the key exists in the db.
                //Grabs a write lock
                r = db_getf_set(db, txn, lock_flags[which_db]|DB_SERIALIZABLE|DB_RMW, del_key, ydb_getf_do_nothing, NULL);
                if (r != 0) goto cleanup;
            } else if (db->i->lt && !(lock_flags[which_db] & DB_PRELOCKED_WRITE)) {  //Do locking if necessary.
                //Needs locking
                r = toku_db_get_point_write_lock(db, txn, del_key);
                if (r != 0) goto cleanup;
            }
        }
        fts[which_db] = db->i->ft_handle;
    }

    if (indexer) {
        // do a cheap check
        if (src_same) {
            bool may_insert = toku_indexer_may_insert(indexer, src_key);
            if (!may_insert) {
                toku_indexer_lock(indexer);
                indexer_lock_taken = true;
            }
            else {
                indexer_shortcut = true;
            }
        }
    }
    toku_multi_operation_client_lock();
    log_del_multiple(txn, src_db, src_key, src_val, num_dbs, fts, del_keys);
    r = do_del_multiple(txn, num_dbs, db_array, del_keys, src_db, src_key, indexer_shortcut);
    toku_multi_operation_client_unlock();
    if (indexer_lock_taken) {
        toku_indexer_unlock(indexer);
    }

cleanup:
    if (r == 0)
        STATUS_VALUE(YDB_LAYER_NUM_MULTI_DELETES) += num_dbs;  // accountability 
    else
        STATUS_VALUE(YDB_LAYER_NUM_MULTI_DELETES_FAIL) += num_dbs;  // accountability 
    return r;
}

static void
log_put_multiple(DB_TXN *txn, DB *src_db, const DBT *src_key, const DBT *src_val, uint32_t num_dbs, FT_HANDLE fts[]) {
    if (num_dbs > 0) {
        TOKUTXN ttxn = db_txn_struct_i(txn)->tokutxn;
        FT_HANDLE src_ft  = src_db ? src_db->i->ft_handle : NULL;
        toku_ft_log_put_multiple(ttxn, src_ft, fts, num_dbs, src_key, src_val);
    }
}

// Requires: If remaining_flags is non-null, this function performs any required uniqueness checks
//           Otherwise, the caller is responsible.
static int
do_put_multiple(DB_TXN *txn, uint32_t num_dbs, DB *db_array[], DBT_ARRAY keys[], DBT_ARRAY vals[], uint32_t *remaining_flags, DB *src_db, const DBT *src_key, bool indexer_shortcut) {
    int r = 0;
    for (uint32_t which_db = 0; which_db < num_dbs; which_db++) {
        DB *db = db_array[which_db];

        invariant(keys[which_db].size == vals[which_db].size);
        paranoid_invariant(keys[which_db].size <= keys[which_db].capacity);
        paranoid_invariant(vals[which_db].size <= vals[which_db].capacity);

        if (keys[which_db].size > 0) {
            bool do_put = true;
            DB_INDEXER *indexer = toku_db_get_indexer(db);
            if (indexer && !indexer_shortcut) { // if this db is the index under construction
                DB *indexer_src_db = toku_indexer_get_src_db(indexer);
                invariant(indexer_src_db != NULL);
                const DBT *indexer_src_key;
                if (src_db == indexer_src_db)
                    indexer_src_key = src_key;
                else {
                    uint32_t which_src_db = lookup_src_db(num_dbs, db_array, indexer_src_db);
                    invariant(which_src_db < num_dbs);
                    // The indexer src db must have exactly one item or we don't know how to continue.
                    invariant(keys[which_src_db].size == 1);
                    indexer_src_key = &keys[which_src_db].dbts[0];
                }
                do_put = toku_indexer_should_insert_key(indexer, indexer_src_key);
                toku_indexer_update_estimate(indexer);
            }
            if (do_put) {
                for (uint32_t i = 0; i < keys[which_db].size; i++) {
                    int flags = 0;
                    if (remaining_flags != nullptr) {
                        flags = remaining_flags[which_db];
                        invariant(!(flags & DB_NOOVERWRITE_NO_ERROR));
                    }
                    r = db_put(db, txn, &keys[which_db].dbts[i], &vals[which_db].dbts[i], flags, false);
                    if (r != 0) {
                        goto done;
                    }
                }
            }
        }
    }
done:
    return r;
}

static int
env_put_multiple_internal(
    DB_ENV *env, 
    DB *src_db, 
    DB_TXN *txn, 
    const DBT *src_key, 
    const DBT *src_val, 
    uint32_t num_dbs, 
    DB **db_array, 
    DBT_ARRAY *keys,
    DBT_ARRAY *vals,
    uint32_t *flags_array) 
{
    int r;
    DBT_ARRAY put_keys[num_dbs];
    DBT_ARRAY put_vals[num_dbs];
    DB_INDEXER* indexer = NULL;

    HANDLE_PANICKED_ENV(env);
    HANDLE_READ_ONLY_TXN(txn);

    uint32_t lock_flags[num_dbs];
    uint32_t remaining_flags[num_dbs];
    FT_HANDLE fts[num_dbs];
    bool indexer_shortcut = false;
    bool indexer_lock_taken = false;
    bool src_same = false;

    if (!txn || !num_dbs) {
        r = EINVAL;
        goto cleanup;
    }
    if (!env->i->generate_row_for_put) {
        r = EINVAL;
        goto cleanup;
    }

    HANDLE_ILLEGAL_WORKING_PARENT_TXN(env, txn);
    r = get_indexer_if_exists(num_dbs, db_array, src_db, &indexer, &src_same);
    if (r) {
        goto cleanup;
    }

    for (uint32_t which_db = 0; which_db < num_dbs; which_db++) {
        DB *db = db_array[which_db];

        lock_flags[which_db] = get_prelocked_flags(flags_array[which_db]);
        remaining_flags[which_db] = flags_array[which_db] & ~lock_flags[which_db];

        //Generate the row
        if (db == src_db) {
            put_keys[which_db].size = put_keys[which_db].capacity = 1;
            put_keys[which_db].dbts = const_cast<DBT*>(src_key);

            put_vals[which_db].size = put_vals[which_db].capacity = 1;
            put_vals[which_db].dbts = const_cast<DBT*>(src_val);
        }
        else {
            r = env->i->generate_row_for_put(db, src_db, &keys[which_db], &vals[which_db], src_key, src_val);
            if (r != 0) goto cleanup;

            paranoid_invariant(keys[which_db].size <= keys[which_db].capacity);
            paranoid_invariant(vals[which_db].size <= vals[which_db].capacity);
            paranoid_invariant(keys[which_db].size == vals[which_db].size);

            put_keys[which_db] = keys[which_db];
            put_vals[which_db] = vals[which_db];
        }
        for (uint32_t i = 0; i < put_keys[which_db].size; i++) {
            DBT &put_key = put_keys[which_db].dbts[i];
            DBT &put_val = put_vals[which_db].dbts[i];

            // check size constraints
            r = db_put_check_size_constraints(db, &put_key, &put_val);
            if (r != 0) goto cleanup;

            if (remaining_flags[which_db] == DB_NOOVERWRITE_NO_ERROR) {
                //put_multiple does not support delaying the no error, since we would
                //have to log the flag in the put_multiple.
                r = EINVAL; goto cleanup;
            }

            //Do locking if necessary.
            if (db->i->lt && !(lock_flags[which_db] & DB_PRELOCKED_WRITE)) {
                //Needs locking
                r = toku_db_get_point_write_lock(db, txn, &put_key);
                if (r != 0) goto cleanup;
            }
        }
        fts[which_db] = db->i->ft_handle;
    }
    
    if (indexer) {
        // do a cheap check
        if (src_same) {
            bool may_insert = toku_indexer_may_insert(indexer, src_key);
            if (!may_insert) {
                toku_indexer_lock(indexer);
                indexer_lock_taken = true;
            }
            else {
                indexer_shortcut = true;
            }
        }
    }
    toku_multi_operation_client_lock();
    r = do_put_multiple(txn, num_dbs, db_array, put_keys, put_vals, remaining_flags, src_db, src_key, indexer_shortcut);
    if (r == 0) {
        log_put_multiple(txn, src_db, src_key, src_val, num_dbs, fts);
    }
    toku_multi_operation_client_unlock();
    if (indexer_lock_taken) {
        toku_indexer_unlock(indexer);
    }

cleanup:
    if (r == 0)
        STATUS_VALUE(YDB_LAYER_NUM_MULTI_INSERTS) += num_dbs;  // accountability 
    else
        STATUS_VALUE(YDB_LAYER_NUM_MULTI_INSERTS_FAIL) += num_dbs;  // accountability 
    return r;
}

static void swap_dbts(DBT *a, DBT *b) {
    DBT c;
    c = *a;
    *a = *b;
    *b = c;
}

//TODO: 26 Add comment in API description about.. new val.size being generated as '0' REQUIRES old_val.size == 0
//
int
env_update_multiple(DB_ENV *env, DB *src_db, DB_TXN *txn,
                    DBT *old_src_key, DBT *old_src_data,
                    DBT *new_src_key, DBT *new_src_data,
                    uint32_t num_dbs, DB **db_array, uint32_t* flags_array,
                    uint32_t num_keys, DBT_ARRAY keys[],
                    uint32_t num_vals, DBT_ARRAY vals[]) {
    int r = 0;

    HANDLE_PANICKED_ENV(env);
    DB_INDEXER* indexer = NULL;
    bool indexer_shortcut = false;
    bool indexer_lock_taken = false;
    bool src_same = false;
    HANDLE_READ_ONLY_TXN(txn);
    DBT_ARRAY old_key_arrays[num_dbs];
    DBT_ARRAY new_key_arrays[num_dbs];
    DBT_ARRAY new_val_arrays[num_dbs];

    if (!txn) {
        r = EINVAL;
        goto cleanup;
    }
    if (!env->i->generate_row_for_put) {
        r = EINVAL;
        goto cleanup;
    }

    if (num_dbs + num_dbs > num_keys || num_dbs > num_vals) {
        r = ENOMEM; goto cleanup;
    }

    HANDLE_ILLEGAL_WORKING_PARENT_TXN(env, txn);
    r = get_indexer_if_exists(num_dbs, db_array, src_db, &indexer, &src_same);
    if (r) {
        goto cleanup;
    }

    {
        uint32_t n_del_dbs = 0;
        DB *del_dbs[num_dbs];
        FT_HANDLE del_fts[num_dbs];
        DBT_ARRAY del_key_arrays[num_dbs];

        uint32_t n_put_dbs = 0;
        DB *put_dbs[num_dbs];
        FT_HANDLE put_fts[num_dbs];
        DBT_ARRAY put_key_arrays[num_dbs];
        DBT_ARRAY put_val_arrays[num_dbs];

        uint32_t lock_flags[num_dbs];
        uint32_t remaining_flags[num_dbs];

        for (uint32_t which_db = 0; which_db < num_dbs; which_db++) {
            DB *db = db_array[which_db];

            lock_flags[which_db] = get_prelocked_flags(flags_array[which_db]);
            remaining_flags[which_db] = flags_array[which_db] & ~lock_flags[which_db];

            if (db == src_db) {
                // Copy the old keys
                old_key_arrays[which_db].size = old_key_arrays[which_db].capacity = 1;
                old_key_arrays[which_db].dbts = old_src_key;

                // Copy the new keys and vals
                new_key_arrays[which_db].size = new_key_arrays[which_db].capacity = 1;
                new_key_arrays[which_db].dbts = new_src_key;

                new_val_arrays[which_db].size = new_val_arrays[which_db].capacity = 1;
                new_val_arrays[which_db].dbts = new_src_data;
            } else {
                // keys[0..num_dbs-1] are the new keys
                // keys[num_dbs..2*num_dbs-1] are the old keys
                // vals[0..num_dbs-1] are the new vals

                // Generate the old keys
                r = env->i->generate_row_for_put(db, src_db, &keys[which_db + num_dbs], NULL, old_src_key, old_src_data);
                if (r != 0) goto cleanup;

                paranoid_invariant(keys[which_db+num_dbs].size <= keys[which_db+num_dbs].capacity);
                old_key_arrays[which_db] = keys[which_db+num_dbs];

                // Generate the new keys and vals
                r = env->i->generate_row_for_put(db, src_db, &keys[which_db], &vals[which_db], new_src_key, new_src_data);
                if (r != 0) goto cleanup;

                paranoid_invariant(keys[which_db].size <= keys[which_db].capacity);
                paranoid_invariant(vals[which_db].size <= vals[which_db].capacity);
                paranoid_invariant(keys[which_db].size == vals[which_db].size);

                new_key_arrays[which_db] = keys[which_db];
                new_val_arrays[which_db] = vals[which_db];
            }
            DBT_ARRAY &old_keys = old_key_arrays[which_db];
            DBT_ARRAY &new_keys = new_key_arrays[which_db];
            DBT_ARRAY &new_vals = new_val_arrays[which_db];

            uint32_t num_skip = 0;
            uint32_t num_del = 0;
            uint32_t num_put = 0;
            // Next index in old_keys to look at
            uint32_t idx_old = 0;
            // Next index in new_keys/new_vals to look at
            uint32_t idx_new = 0;
            uint32_t idx_old_used = 0;
            uint32_t idx_new_used = 0;
            while (idx_old < old_keys.size || idx_new < new_keys.size) {
                // Check for old key, both, new key
                DBT *curr_old_key = &old_keys.dbts[idx_old];
                DBT *curr_new_key = &new_keys.dbts[idx_new];
                DBT *curr_new_val = &new_vals.dbts[idx_new];

                bool locked_new_key = false;
                int cmp;
                if (idx_new == new_keys.size) {
                    cmp = -1;
                } else if (idx_old == old_keys.size) {
                    cmp = +1;
                } else {
                    const toku::comparator &cmpfn = toku_db_get_comparator(db);
                    cmp = cmpfn(curr_old_key, curr_new_key);
                }

                bool do_del = false;
                bool do_put = false;
                bool do_skip = false;
                if (cmp > 0) { // New key does not exist in old array
                    //Check overwrite constraints only in the case where the keys are not equal
                    //(new key is alone/not equal to old key)
                    // If the keys are equal, then we do not care of the flag is DB_NOOVERWRITE or 0
                    r = db_put_check_overwrite_constraint(db, txn,
                                                          curr_new_key,
                                                          lock_flags[which_db], remaining_flags[which_db]);
                    if (r != 0) goto cleanup;
                    if (remaining_flags[which_db] == DB_NOOVERWRITE) {
                        locked_new_key = true;
                    }

                    if (remaining_flags[which_db] == DB_NOOVERWRITE_NO_ERROR) {
                        //update_multiple does not support delaying the no error, since we would
                        //have to log the flag in the put_multiple.
                        r = EINVAL; goto cleanup;
                    }
                    do_put = true;
                } else if (cmp < 0) {
                    // lock old key only when it does not exist in new array
                    // otherwise locking new key takes care of this
                    if (db->i->lt && !(lock_flags[which_db] & DB_PRELOCKED_WRITE)) {
                        r = toku_db_get_point_write_lock(db, txn, curr_old_key);
                        if (r != 0) goto cleanup;
                    }
                    do_del = true;
                } else {
                    do_put = curr_new_val->size > 0 ||
                                curr_old_key->size != curr_new_key->size ||
                                memcmp(curr_old_key->data, curr_new_key->data, curr_old_key->size);
                    do_skip = !do_put;
                }
                // Check put size constraints and insert new key only if keys are unequal (byte for byte) or there is a val
                // We assume any val.size > 0 as unequal (saves on generating old val)
                //      (allows us to avoid generating the old val)
                // we assume that any new vals with size > 0 are different than the old val
                // if (!key_eq || !(dbt_cmp(&vals[which_db], &vals[which_db + num_dbs]) == 0)) { /* ... */ }
                if (do_put) {
                    r = db_put_check_size_constraints(db, curr_new_key, curr_new_val);
                    if (r != 0) goto cleanup;

                    // lock new key unless already locked
                    if (db->i->lt && !(lock_flags[which_db] & DB_PRELOCKED_WRITE) && !locked_new_key) {
                        r = toku_db_get_point_write_lock(db, txn, curr_new_key);
                        if (r != 0) goto cleanup;
                    }
                }

                // TODO: 26 Add comments explaining squish and why not just use another stack array
                // Add more comments to explain this if elseif else well
                if (do_skip) {
                    paranoid_invariant(cmp == 0);
                    paranoid_invariant(!do_put);
                    paranoid_invariant(!do_del);

                    num_skip++;
                    idx_old++;
                    idx_new++;
                } else if (do_put) {
                    paranoid_invariant(cmp >= 0);
                    paranoid_invariant(!do_skip);
                    paranoid_invariant(!do_del);

                    num_put++;
                    if (idx_new != idx_new_used) {
                        swap_dbts(&new_keys.dbts[idx_new_used], &new_keys.dbts[idx_new]);
                        swap_dbts(&new_vals.dbts[idx_new_used], &new_vals.dbts[idx_new]);
                    }
                    idx_new++;
                    idx_new_used++;
                    if (cmp == 0) {
                        idx_old++;
                    }
                } else {
                    invariant(do_del);
                    paranoid_invariant(cmp < 0);
                    paranoid_invariant(!do_skip);
                    paranoid_invariant(!do_put);

                    num_del++;
                    if (idx_old != idx_old_used) {
                        swap_dbts(&old_keys.dbts[idx_old_used], &old_keys.dbts[idx_old]);
                    }
                    idx_old++;
                    idx_old_used++;
                }
            }
            old_keys.size = idx_old_used;
            new_keys.size = idx_new_used;
            new_vals.size = idx_new_used;

            if (num_del > 0) {
                del_dbs[n_del_dbs] = db;
                del_fts[n_del_dbs] = db->i->ft_handle;
                del_key_arrays[n_del_dbs] = old_keys;
                n_del_dbs++;
            }
            // If we put none, but delete some, but not all, then we need the log_put_multiple to happen.
            // Include this db in the put_dbs so we do log_put_multiple.
            // do_put_multiple will be a no-op for this db.
            if (num_put > 0 || (num_del > 0 && num_skip > 0)) {
                put_dbs[n_put_dbs] = db;
                put_fts[n_put_dbs] = db->i->ft_handle;
                put_key_arrays[n_put_dbs] = new_keys;
                put_val_arrays[n_put_dbs] = new_vals;
                n_put_dbs++;
            }
        }
        if (indexer) {
            // do a cheap check
            if (src_same) {
                bool may_insert =
                    toku_indexer_may_insert(indexer, old_src_key) &&
                    toku_indexer_may_insert(indexer, new_src_key);
                if (!may_insert) {
                    toku_indexer_lock(indexer);
                    indexer_lock_taken = true;
                }
                else {
                    indexer_shortcut = true;
                }
            }
        }
        toku_multi_operation_client_lock();
        if (r == 0 && n_del_dbs > 0) {
            log_del_multiple(txn, src_db, old_src_key, old_src_data, n_del_dbs, del_fts, del_key_arrays);
            r = do_del_multiple(txn, n_del_dbs, del_dbs, del_key_arrays, src_db, old_src_key, indexer_shortcut);
        }

        if (r == 0 && n_put_dbs > 0) {
            // We sometimes skip some keys for del/put during runtime, but during recovery
            // we (may) delete ALL the keys for a given DB.  Therefore we must put ALL the keys during
            // recovery so we don't end up losing data.
            // So unlike env->put_multiple, we ONLY log a 'put_multiple' log entry.
            log_put_multiple(txn, src_db, new_src_key, new_src_data, n_put_dbs, put_fts);
            r = do_put_multiple(txn, n_put_dbs, put_dbs, put_key_arrays, put_val_arrays, nullptr, src_db, new_src_key, indexer_shortcut);
        }
        toku_multi_operation_client_unlock();
        if (indexer_lock_taken) {
            toku_indexer_unlock(indexer);
        }
    }

cleanup:
    if (r == 0)
        STATUS_VALUE(YDB_LAYER_NUM_MULTI_UPDATES) += num_dbs;  // accountability 
    else
        STATUS_VALUE(YDB_LAYER_NUM_MULTI_UPDATES_FAIL) += num_dbs;  // accountability 
    return r;
}

int 
autotxn_db_del(DB* db, DB_TXN* txn, DBT* key, uint32_t flags) {
    bool changed; int r;
    r = toku_db_construct_autotxn(db, &txn, &changed, false);
    if (r!=0) return r;
    r = toku_db_del(db, txn, key, flags, false);
    return toku_db_destruct_autotxn(txn, r, changed);
}

int 
autotxn_db_put(DB* db, DB_TXN* txn, DBT* key, DBT* data, uint32_t flags) {
    //{ unsigned i; printf("put %p keylen=%d key={", db, key->size); for(i=0; i<key->size; i++) printf("%d,", ((char*)key->data)[i]); printf("} datalen=%d data={", data->size); for(i=0; i<data->size; i++) printf("%d,", ((char*)data->data)[i]); printf("}\n"); }
    bool changed; int r;
    r = env_check_avail_fs_space(db->dbenv);
    if (r != 0) { goto cleanup; }
    r = toku_db_construct_autotxn(db, &txn, &changed, false);
    if (r!=0) {
        goto cleanup;
    }
    r = toku_db_put(db, txn, key, data, flags, false);
    r = toku_db_destruct_autotxn(txn, r, changed);
cleanup:
    return r;
}

int
autotxn_db_update(DB *db, DB_TXN *txn,
                  const DBT *key,
                  const DBT *update_function_extra,
                  uint32_t flags) {
    bool changed; int r;
    r = env_check_avail_fs_space(db->dbenv);
    if (r != 0) { goto cleanup; }
    r = toku_db_construct_autotxn(db, &txn, &changed, false);
    if (r != 0) { return r; }
    r = toku_db_update(db, txn, key, update_function_extra, flags);
    r = toku_db_destruct_autotxn(txn, r, changed);
cleanup:
    return r;
}

int
autotxn_db_update_broadcast(DB *db, DB_TXN *txn,
                            const DBT *update_function_extra,
                            uint32_t flags) {
    bool changed; int r;
    r = env_check_avail_fs_space(db->dbenv);
    if (r != 0) { goto cleanup; }
    r = toku_db_construct_autotxn(db, &txn, &changed, false);
    if (r != 0) { return r; }
    r = toku_db_update_broadcast(db, txn, update_function_extra, flags);
    r = toku_db_destruct_autotxn(txn, r, changed);
cleanup:
    return r;
}

int
env_put_multiple(DB_ENV *env, DB *src_db, DB_TXN *txn, const DBT *src_key, const DBT *src_val, uint32_t num_dbs, DB **db_array, DBT_ARRAY *keys, DBT_ARRAY *vals, uint32_t *flags_array) {
    int r = env_check_avail_fs_space(env);
    if (r == 0) {
        r = env_put_multiple_internal(env, src_db, txn, src_key, src_val, num_dbs, db_array, keys, vals, flags_array);
    }
    return r;
}

int
toku_ydb_check_avail_fs_space(DB_ENV *env) {
    int rval = env_check_avail_fs_space(env);
    return rval;
}
#undef STATUS_VALUE

#include <toku_race_tools.h>
void __attribute__((constructor)) toku_ydb_write_helgrind_ignore(void);
void
toku_ydb_write_helgrind_ignore(void) {
    TOKU_VALGRIND_HG_DISABLE_CHECKING(&ydb_write_layer_status, sizeof ydb_write_layer_status);
}