From mboxrd@z Thu Jan  1 00:00:00 1970
Return-Path: <v.shpilevoy@tarantool.org>
Subject: Re: [tarantool-patches] Re: [PATCH v2 1/2] box: create bigrefs for
 tuples
References: <cover.1528388742.git.imeevma@gmail.com>
 <6bfce1d3b6f797a6ebf3e56f81fad4181e855453.1528388742.git.imeevma@gmail.com>
 <ad4a1696-b4a1-16bc-f256-b3dd65abbc82@tarantool.org>
 <bc713600-f06e-e960-4d0d-2947547749d3@tarantool.org>
From: Vladislav Shpilevoy <v.shpilevoy@tarantool.org>
Message-ID: <8ab5a569-a85d-5c3e-68ef-5c392727ac09@tarantool.org>
Date: Fri, 8 Jun 2018 15:03:19 +0300
MIME-Version: 1.0
In-Reply-To: <bc713600-f06e-e960-4d0d-2947547749d3@tarantool.org>
Content-Type: text/plain; charset="utf-8"; format="flowed"
Content-Language: en-US
Content-Transfer-Encoding: 8bit
To: tarantool-patches@freelists.org, Imeev Mergen <imeevma@tarantool.org>, Vladimir Davydov <vdavydov.dev@gmail.com>
List-ID: <tarantool-patches.dev.tarantool.org>

I have force pushed my minor fixes on the branch.
Now the patch LGTM.

Vova, please, review the patchset.

On 08/06/2018 13:46, Imeev Mergen wrote:
> 
> 
> On 06/08/2018 12:24 AM, Vladislav Shpilevoy wrote:
>> Hello. Thanks for the patch! I have pushed some minor fixes. Look at
>> them, squash. See the rest 3 comments below.
>>
>>> diff --git a/test/unit/tuple_bigref.c b/test/unit/tuple_bigref.c
>>> new file mode 100644
>>> index 0000000..fd166fd
>>> --- /dev/null
>>> +++ b/test/unit/tuple_bigref.c
>>> @@ -0,0 +1,156 @@
>>> +#include "vy_iterators_helper.h"
>>> +#include "memory.h"
>>> +#include "fiber.h"
>>> +#include "unit.h"
>>> +#include <msgpuck.h>
>>> +#include "trivia/util.h"
>>> +
>>> +enum
>>> +{
>>> +    BIGREF_DIFF = 10,
>>> +    BIGREF_COUNT = 100003,
>>
>> 1. Please, reduce the count to 70k for example. There is no
>> difference 100k or 70k for bigref list, but the test will be
>> faster.
> Done.
>>
>>> +    BIGREF_CAPACITY = 107,
>>> +};
>>> +
>>> +/**
>>> + * This test checks that indexes are given as intended.
>>
>> 2. Actually this test works exactly like the first one. See the
>> comments below.
>>> + */
>>> +static int
>>> +test_bigrefs_3()
>>> +{
>>> +    struct tuple **tuples = (struct tuple **) malloc(BIGREF_CAPACITY *
>>> +                             sizeof(*tuples));
>>> +    for(int i = 0; i < BIGREF_CAPACITY; ++i) {
>>> +        tuples[i] = create_tuple();
>>> +        tuple_ref(tuples[i]);
>>> +    }
>>> +    for(int i = 0; i < BIGREF_CAPACITY; ++i) {
>>> +        assert(tuples[i]->refs == 1);
>>> +        for(int j = 1; j < BIGREF_COUNT; ++j)
>>> +            box_tuple_ref(tuples[i]);
>>> +        assert(tuples[i]->is_bigref);
>>> +        if(i % BIGREF_DIFF == 0) {
>>> +            for(int j = 1; j < BIGREF_COUNT; ++j)
>>> +                box_tuple_unref(tuples[i]);
>>> +        }
>>
>> 3. Here when i % BIGREF_DIFF, you had added the tuple to bigref list,
>> and then have deleted it immediately. So the list has no gaps and works
>> exactly like in the first test.
>>
>> Please, rework this test to test a list with gaps. You should fill it,
>> then remove some of tuples, for example, each BIGREF_DIFF, and check,
>> that other tuples are not destroyed. And that new tuples occupy the
>> gaps.
> Changed test_bigrefs_3, added gaps in creating and in destroying. New received
> indexes should be equal to index of tuple with refilled refs.
>> And please, use gcov to check coverage of tuple bigrefs. I have run
>> gcov, and saw this:
>>
>>         -:  362:static inline void
>>         -:  363:bigref_list_delete_index(uint16_t index)
>>         -:  364:{
>>       648:  365:    bigref_list.refs[index] = 0;
>>       648:  366:    if (--bigref_list.size == 0) {
>>        15:  367:        bigref_list_reset();
>>        15:  368:        return;
>>         -:  369:    }
>>         -:  370:    /* Drop the 'take' hint. */
>>       633:  371:    bigref_list.hint_index_to_take = 0;
>>      1263:  372:    if (bigref_list.capacity == BIGREF_MIN_CAPACITY ||
>>       630:  373:        bigref_list.size > bigref_list.capacity / BIGREF_FACTOR)
>>       249:  374:        return;
>>         -:  375:
>>       384:  376:    uint16_t top_index = bigref_list.hint_index_to_free;
>>      1086:  377:    while (bigref_list.refs[top_index] == 0)
>>       159:  378:        top_index--;
>>       384:  379:    bigref_list.hint_index_to_free = top_index;
>>         -:  380:
>>       384:  381:    uint16_t needed_capacity = top_index + 1;
>>       384:  382:    if (needed_capacity < BIGREF_MIN_CAPACITY)
>>     #####:  383:        needed_capacity = BIGREF_MIN_CAPACITY;
>>       384:  384:    if (needed_capacity > bigref_list.capacity / BIGREF_FACTOR)
>>       384:  385:        return;
>>         -:  386:    /* Round up capacity to the next highest power of 2. */
>>         -:  387:    assert(sizeof(needed_capacity) == sizeof(uint16_t));
>>     #####:  388:    needed_capacity--;
>>     #####:  389:    needed_capacity |= needed_capacity >> 1;
>>     #####:  390:    needed_capacity |= needed_capacity >> 2;
>>     #####:  391:    needed_capacity |= needed_capacity >> 4;
>>     #####:  392:    needed_capacity |= needed_capacity >> 8;
>>     #####:  393:    assert(needed_capacity < UINT16_MAX);
>>     #####:  394:    needed_capacity++;
>>     #####:  395:    uint32_t *refs =
>>     #####:  396:        (uint32_t *) realloc(bigref_list.refs, needed_capacity *
>>         -:  397:                     sizeof(*bigref_list.refs));
>>     #####:  398:    if (refs == NULL) {
>>     #####:  399:        panic("failed to reallocate %zu bytes: Cannot allocate "\
>>         -:  400:              "memory.", needed_capacity * sizeof(*bigref_list.refs));
>>         -:  401:    }
>>     #####:  402:    bigref_list.refs = refs;
>>     #####:  403:    bigref_list.capacity = needed_capacity;
>>       648:  404:}
>>
> Done. All lines but the ones that contain panic() are used.
>>
>> ##### means this line had never been executed. You should write such a test, that
>> will cover these lines.
>>
>> To use gcov you can build Tarantool with ENABLE_GCOV:
>>
>>     cmake . -DENABLE_GCOV=1
>>     make -j
>>
>> Then you run the test:
>>
>>     ./test/unit/tuple_bigref.test
>>
>> Then you convert .gcda/.gcno files to .gcov files:
>>
>>     gcov src/box/tuple.c -o src/box/CMakeFiles/tuple.dir/tuple.c.gcno
>>
>> Now tuple.c.gcov stores the coverage info.
>> These commands work on Mac. Maybe on Linux it differs. If cmake ENABLE_GCOV
>> does not work, then try this:
>>
>>     @@ -4,10 +4,7 @@ set(ENABLE_GCOV_DEFAULT OFF)
>>      option(ENABLE_GCOV "Enable integration with gcov, a code coverage program" ${ENABLE_GCOV_DEFAULT})
>>
>>      if (ENABLE_GCOV)
>>     -    if (NOT HAVE_GCOV)
>>     -    message (FATAL_ERROR
>>     -         "ENABLE_GCOV option requested but gcov library is not found")
>>     -    endif()
>>
>> and run cmake again.
>>
> 
> commit d7e3481f71882ff02f08f3f66175e6bb3830b6e0
> Author: Mergen Imeev <imeevma@gmail.com>
> Date: Mon, 28 May 2018 19:17:51 +0300
> 
> box: create bigrefs for tuples
> 
> Due to limitation of reference counters for tuple being only
> 65535 it was possible to reach this limitation. This patch
> increases capacity of reference counters to 4 billions.
> 
> Closes #3224
> ---
> 
> diff --git a/src/box/box.cc b/src/box/box.cc
> index c728a4c..4257861 100644
> --- a/src/box/box.cc
> +++ b/src/box/box.cc
> @@ -174,20 +174,22 @@ process_rw(struct request *request, struct space *space, struct tuple **result)
>           txn_rollback_stmt();
>           return -1;
>       }
> +    if (result == NULL)
> +        return txn_commit_stmt(txn, request);
> +    *result = tuple;
> +    if (tuple == NULL)
> +        return txn_commit_stmt(txn, request);
>       /*
>        * Pin the tuple locally before the commit,
>        * otherwise it may go away during yield in
>        * when WAL is written in autocommit mode.
>        */
> -    TupleRefNil ref(tuple);
> -    if (txn_commit_stmt(txn, request) != 0)
> -        return -1;
> -    if (result != NULL) {
> -        if (tuple != NULL && tuple_bless(tuple) == NULL)
> -            return -1;
> -        *result = tuple;
> -    }
> -    return 0;
> +    tuple_ref(tuple);
> +    int rc = txn_commit_stmt(txn, request);
> +    if (rc == 0)
> +        tuple_bless(tuple);
> +    tuple_unref(tuple);
> +    return rc;
>   }
> 
>   void
> diff --git a/src/box/errcode.h b/src/box/errcode.h
> index a0759f8..e009524 100644
> --- a/src/box/errcode.h
> +++ b/src/box/errcode.h
> @@ -138,7 +138,7 @@ struct errcode_record {
>       /* 83 */_(ER_ROLE_EXISTS,        "Role '%s' already exists") \
>       /* 84 */_(ER_CREATE_ROLE,        "Failed to create role '%s': %s") \
>       /* 85 */_(ER_INDEX_EXISTS,        "Index '%s' already exists") \
> -    /* 86 */_(ER_TUPLE_REF_OVERFLOW,    "Tuple reference counter overflow") \
> +    /* 86 */_(ER_UNUSED6,            "") \
>       /* 87 */_(ER_ROLE_LOOP,            "Granting role '%s' to role '%s' would create a loop") \
>       /* 88 */_(ER_GRANT,            "Incorrect grant arguments: %s") \
>       /* 89 */_(ER_PRIV_GRANTED,        "User '%s' already has %s access on %s '%s'") \
> diff --git a/src/box/index.cc b/src/box/index.cc
> index 3c62ec1..f992bc9 100644
> --- a/src/box/index.cc
> +++ b/src/box/index.cc
> @@ -220,8 +220,8 @@ box_index_random(uint32_t space_id, uint32_t index_id, uint32_t rnd,
>       /* No tx management, random() is for approximation anyway. */
>       if (index_random(index, rnd, result) != 0)
>           return -1;
> -    if (*result != NULL && tuple_bless(*result) == NULL)
> -        return -1;
> +    if (*result != NULL)
> +        tuple_bless(*result);
>       return 0;
>   }
> 
> @@ -253,8 +253,8 @@ box_index_get(uint32_t space_id, uint32_t index_id, const char *key,
>       txn_commit_ro_stmt(txn);
>       /* Count statistics. */
>       rmean_collect(rmean_box, IPROTO_SELECT, 1);
> -    if (*result != NULL && tuple_bless(*result) == NULL)
> -        return -1;
> +    if (*result != NULL)
> +        tuple_bless(*result);
>       return 0;
>   }
> 
> @@ -285,8 +285,8 @@ box_index_min(uint32_t space_id, uint32_t index_id, const char *key,
>           return -1;
>       }
>       txn_commit_ro_stmt(txn);
> -    if (*result != NULL && tuple_bless(*result) == NULL)
> -        return -1;
> +    if (*result != NULL)
> +        tuple_bless(*result);
>       return 0;
>   }
> 
> @@ -317,8 +317,8 @@ box_index_max(uint32_t space_id, uint32_t index_id, const char *key,
>           return -1;
>       }
>       txn_commit_ro_stmt(txn);
> -    if (*result != NULL && tuple_bless(*result) == NULL)
> -        return -1;
> +    if (*result != NULL)
> +        tuple_bless(*result);
>       return 0;
>   }
> 
> @@ -397,8 +397,8 @@ box_iterator_next(box_iterator_t *itr, box_tuple_t **result)
>       assert(result != NULL);
>       if (iterator_next(itr, result) != 0)
>           return -1;
> -    if (*result != NULL && tuple_bless(*result) == NULL)
> -        return -1;
> +    if (*result != NULL)
> +        tuple_bless(*result);
>       return 0;
>   }
> 
> diff --git a/src/box/lua/tuple.c b/src/box/lua/tuple.c
> index 8057331..22fe696 100644
> --- a/src/box/lua/tuple.c
> +++ b/src/box/lua/tuple.c
> @@ -496,10 +496,7 @@ luaT_pushtuple(struct lua_State *L, box_tuple_t *tuple)
>           luaL_pushcdata(L, CTID_CONST_STRUCT_TUPLE_REF);
>       *ptr = tuple;
>       /* The order is important - first reference tuple, next set gc */
> -    if (box_tuple_ref(tuple) != 0) {
> -        luaT_error(L);
> -        return;
> -    }
> +    box_tuple_ref(tuple);
>       lua_pushcfunction(L, lbox_tuple_gc);
>       luaL_setcdatagc(L, -2);
>   }
> diff --git a/src/box/port.c b/src/box/port.c
> index 03f6be7..9b6b858 100644
> --- a/src/box/port.c
> +++ b/src/box/port.c
> @@ -45,8 +45,7 @@ port_tuple_add(struct port *base, struct tuple *tuple)
>       struct port_tuple *port = port_tuple(base);
>       struct port_tuple_entry *e;
>       if (port->size == 0) {
> -        if (tuple_ref(tuple) != 0)
> -            return -1;
> +        tuple_ref(tuple);
>           e = &port->first_entry;
>           port->first = port->last = e;
>       } else {
> @@ -55,10 +54,7 @@ port_tuple_add(struct port *base, struct tuple *tuple)
>               diag_set(OutOfMemory, sizeof(*e), "mempool_alloc", "e");
>               return -1;
>           }
> -        if (tuple_ref(tuple) != 0) {
> -            mempool_free(&port_tuple_entry_pool, e);
> -            return -1;
> -        }
> +        tuple_ref(tuple);
>           port->last->next = e;
>           port->last = e;
>       }
> diff --git a/src/box/tuple.c b/src/box/tuple.c
> index 014f374..68540f4 100644
> --- a/src/box/tuple.c
> +++ b/src/box/tuple.c
> @@ -48,6 +48,26 @@ enum {
>       OBJSIZE_MIN = 16,
>   };
> 
> +/**
> + * Container for big reference counters. Contains array of big
> + * reference counters, size of this array and number of non-zero
> + * big reference counters. When reference counter of tuple becomes
> + * more than 32767, field refs of this tuple becomes index of big
> + * reference counter in big reference counter array and field
> + * is_bigref is set true. The moment big reference becomes equal
> + * 32767 it is set to 0, refs of the tuple becomes 32767 and
> + * is_bigref becomes false. Big reference counter can be equal to
> + * 0 or be more than 32767.
> + */
> +static struct bigref_list {
> +    /** Array of big reference counters. */
> +    uint32_t *refs;
> +    /** Number of non-zero elements in the array. */
> +    uint16_t size;
> +    /** Capacity of the array. */
> +    uint16_t capacity;
> +} bigref_list;
> +
>   static const double ALLOC_FACTOR = 1.05;
> 
>   /**
> @@ -151,6 +171,13 @@ tuple_validate_raw(struct tuple_format *format, const char *tuple)
>       return 0;
>   }
> 
> +/** Initialize big references container. */
> +static inline void
> +bigref_list_create()
> +{
> +    memset(&bigref_list, 0, sizeof(bigref_list));
> +}
> +
>   /**
>    * Incremented on every snapshot and is used to distinguish tuples
>    * which were created after start of a snapshot (these tuples can
> @@ -211,6 +238,8 @@ tuple_init(field_name_hash_f hash)
> 
>       box_tuple_last = NULL;
> 
> +    bigref_list_create();
> +
>       if (coll_id_cache_init() != 0)
>           return -1;
> 
> @@ -244,6 +273,132 @@ tuple_arena_create(struct slab_arena *arena, struct quota *quota,
>       }
>   }
> 
> +enum {
> +    BIGREF_FACTOR = 2,
> +    BIGREF_MAX = UINT32_MAX,
> +    BIGREF_MIN_CAPACITY = 16,
> +    /**
> +     * Only 15 bits are available for bigref list index in
> +     * struct tuple.
> +     */
> +    BIGREF_MAX_CAPACITY = UINT16_MAX >> 1
> +};
> +
> +/** Destroy big references and free memory that was allocated. */
> +static inline void
> +bigref_list_reset()
> +{
> +    free(bigref_list.refs);
> +    bigref_list_create();
> +}
> +
> +/**
> + * Return index for new big reference counter and allocate memory
> + * if needed.
> + * @retval index for new big reference counter.
> + */
> +static inline uint16_t
> +bigref_list_new_index()
> +{
> +    if (bigref_list.size < bigref_list.capacity) {
> +        uint16_t vacant_index = 0;
> +        while (bigref_list.refs[vacant_index] != 0)
> +            ++vacant_index;
> +        ++bigref_list.size;
> +        return vacant_index;
> +    }
> +    /* Extend the array. */
> +    uint16_t capacity = bigref_list.capacity;
> +    if (capacity == 0)
> +        capacity = BIGREF_MIN_CAPACITY;
> +    else if (capacity < BIGREF_MAX_CAPACITY)
> +        capacity = MIN(capacity * BIGREF_FACTOR, BIGREF_MAX_CAPACITY);
> +    else
> +        panic("Too many big references");
> +    uint32_t *refs = (uint32_t *) realloc(bigref_list.refs, capacity *
> +                          sizeof(*bigref_list.refs));
> +    if (refs == NULL) {
> +        panic("failed to reallocate %zu bytes: Cannot allocate "\
> +              "memory.", capacity * sizeof(*bigref_list.refs));
> +    }
> +    bigref_list.refs = refs;
> +    memset(bigref_list.refs + bigref_list.capacity, 0, (capacity -
> +           bigref_list.capacity) * sizeof(*bigref_list.refs));
> +    bigref_list.capacity = capacity;
> +    return bigref_list.size++;
> +}
> +
> +void
> +tuple_ref_slow(struct tuple *tuple)
> +{
> +    assert(tuple->is_bigref || tuple->refs == TUPLE_REF_MAX);
> +    if (! tuple->is_bigref) {
> +        tuple->ref_index = bigref_list_new_index();
> +        tuple->is_bigref = true;
> +        bigref_list.refs[tuple->ref_index] = TUPLE_REF_MAX;
> +    } else if (bigref_list.refs[tuple->ref_index] == BIGREF_MAX) {
> +        panic("Tuple big reference counter overflow");
> +    }
> +    bigref_list.refs[tuple->ref_index]++;
> +}
> +
> +/**
> + * Try to decrease allocated memory if it is possible. Free memory
> + * when size == 0.
> + */
> +static inline void
> +bigref_list_delete_index(uint16_t index)
> +{
> +    bigref_list.refs[index] = 0;
> +    if (--bigref_list.size == 0) {
> +        bigref_list_reset();
> +        return;
> +    }
> +    if (bigref_list.capacity == BIGREF_MIN_CAPACITY ||
> +        bigref_list.size > bigref_list.capacity / BIGREF_FACTOR)
> +        return;
> +
> +    uint16_t top_index = bigref_list.capacity - 1;
> +    while (bigref_list.refs[top_index] == 0)
> +        top_index--;
> +
> +    uint16_t needed_capacity = top_index + 1;
> +    if (needed_capacity < BIGREF_MIN_CAPACITY)
> +        needed_capacity = BIGREF_MIN_CAPACITY;
> +    if (needed_capacity > bigref_list.capacity / BIGREF_FACTOR)
> +        return;
> +    /* Round up capacity to the next highest power of 2. */
> +    assert(sizeof(needed_capacity) == sizeof(uint16_t));
> +    needed_capacity--;
> +    needed_capacity |= needed_capacity >> 1;
> +    needed_capacity |= needed_capacity >> 2;
> +    needed_capacity |= needed_capacity >> 4;
> +    needed_capacity |= needed_capacity >> 8;
> +    assert(needed_capacity < UINT16_MAX);
> +    needed_capacity++;
> +    uint32_t *refs =
> +        (uint32_t *) realloc(bigref_list.refs, needed_capacity *
> +                     sizeof(*bigref_list.refs));
> +    if (refs == NULL) {
> +        panic("failed to reallocate %zu bytes: Cannot allocate "\
> +              "memory.", needed_capacity * sizeof(*bigref_list.refs));
> +    }
> +    bigref_list.refs = refs;
> +    bigref_list.capacity = needed_capacity;
> +}
> +
> +void
> +tuple_unref_slow(struct tuple *tuple)
> +{
> +    assert(tuple->is_bigref &&
> +           bigref_list.refs[tuple->ref_index] > TUPLE_REF_MAX);
> +    if(--bigref_list.refs[tuple->ref_index] == TUPLE_REF_MAX) {
> +        bigref_list_delete_index(tuple->ref_index);
> +        tuple->ref_index = TUPLE_REF_MAX;
> +        tuple->is_bigref = false;
> +    }
> +}
> +
>   void
>   tuple_arena_destroy(struct slab_arena *arena)
>   {
> @@ -265,6 +420,8 @@ tuple_free(void)
>       tuple_format_free();
> 
>       coll_id_cache_destroy();
> +
> +    bigref_list_reset();
>   }
> 
>   box_tuple_format_t *
> @@ -288,7 +445,8 @@ int
>   box_tuple_ref(box_tuple_t *tuple)
>   {
>       assert(tuple != NULL);
> -    return tuple_ref(tuple);
> +    tuple_ref(tuple);
> +    return 0;
>   }
> 
>   void
> @@ -357,10 +515,7 @@ box_tuple_iterator(box_tuple_t *tuple)
>                "mempool", "new slab");
>           return NULL;
>       }
> -    if (tuple_ref(tuple) != 0) {
> -        mempool_free(&tuple_iterator_pool, it);
> -        return NULL;
> -    }
> +    tuple_ref(tuple);
>       tuple_rewind(it, tuple);
>       return it;
>   }
> @@ -451,7 +606,6 @@ box_tuple_new(box_tuple_format_t *format, const char *data, const char *end)
>       struct tuple *ret = tuple_new(format, data, end);
>       if (ret == NULL)
>           return NULL;
> -    /* Can't fail on zero refs. */
>       return tuple_bless(ret);
>   }
> 
> diff --git a/src/box/tuple.h b/src/box/tuple.h
> index e2384dd..14dbd40 100644
> --- a/src/box/tuple.h
> +++ b/src/box/tuple.h
> @@ -105,8 +105,7 @@ typedef struct tuple box_tuple_t;
>    * tuple will leak.
>    *
>    * \param tuple a tuple
> - * \retval -1 on error (check box_error_last())
> - * \retval 0 on success
> + * \retval 0 always
>    * \sa box_tuple_unref()
>    */
>   int
> @@ -269,8 +268,7 @@ box_tuple_next(box_tuple_iterator_t *it);
>    * Use box_tuple_format_default() to create space-independent tuple.
>    * \param data tuple data in MsgPack Array format ([field1, field2, ...]).
>    * \param end the end of \a data
> - * \retval NULL on out of memory
> - * \retval tuple otherwise
> + * \retval tuple
>    * \pre data, end is valid MsgPack Array
>    * \sa \code box.tuple.new(data) \endcode
>    */
> @@ -307,9 +305,17 @@ box_tuple_upsert(const box_tuple_t *tuple, const char *expr, const
>    */
>   struct PACKED tuple
>   {
> -    /** reference counter */
> -    uint16_t refs;
> -    /** format identifier */
> +    union {
> +        /** Reference counter. */
> +        uint16_t refs;
> +        struct {
> +            /** Index of big reference counter. */
> +            uint16_t ref_index : 15;
> +            /** Big reference flag. */
> +            bool is_bigref : 1;
> +        };
> +    };
> +    /** Format identifier. */
>       uint16_t format_id;
>       /**
>        * Length of the MessagePack data in raw part of the
> @@ -774,26 +780,36 @@ tuple_field_uuid(const struct tuple *tuple, int fieldno,
>       return 0;
>   }
> 
> -enum { TUPLE_REF_MAX = UINT16_MAX };
> +enum { TUPLE_REF_MAX = UINT16_MAX >> 1 };
> +
> +/**
> + * Increase tuple big reference counter.
> + * @param tuple Tuple to reference.
> + */
> +void
> +tuple_ref_slow(struct tuple *tuple);
> 
>   /**
>    * Increment tuple reference counter.
>    * @param tuple Tuple to reference.
> - * @retval  0 Success.
> - * @retval -1 Too many refs error.
>    */
> -static inline int
> +static inline void
>   tuple_ref(struct tuple *tuple)
>   {
> -    if (tuple->refs + 1 > TUPLE_REF_MAX) {
> -        diag_set(ClientError, ER_TUPLE_REF_OVERFLOW);
> -        return -1;
> -    }
> -    tuple->refs++;
> -    return 0;
> +    if (unlikely(tuple->refs >= TUPLE_REF_MAX))
> +        tuple_ref_slow(tuple);
> +    else
> +        tuple->refs++;
>   }
> 
>   /**
> + * Decrease tuple big reference counter.
> + * @param tuple Tuple to reference.
> + */
> +void
> +tuple_unref_slow(struct tuple *tuple);
> +
> +/**
>    * Decrement tuple reference counter. If it has reached zero, free the tuple.
>    *
>    * @pre tuple->refs + count >= 0
> @@ -802,10 +818,9 @@ static inline void
>   tuple_unref(struct tuple *tuple)
>   {
>       assert(tuple->refs - 1 >= 0);
> -
> -    tuple->refs--;
> -
> -    if (tuple->refs == 0)
> +    if (unlikely(tuple->is_bigref))
> +        tuple_unref_slow(tuple);
> +    else if (--tuple->refs == 0)
>           tuple_delete(tuple);
>   }
> 
> @@ -813,25 +828,18 @@ extern struct tuple *box_tuple_last;
> 
>   /**
>    * Convert internal `struct tuple` to public `box_tuple_t`.
> - * \retval tuple on success
> - * \retval NULL on error, check diag
> + * \retval tuple
>    * \post \a tuple ref counted until the next call.
> - * \post tuple_ref() doesn't fail at least once
>    * \sa tuple_ref
>    */
>   static inline box_tuple_t *
>   tuple_bless(struct tuple *tuple)
>   {
>       assert(tuple != NULL);
> -    /* Ensure tuple can be referenced at least once after return */
> -    if (tuple->refs + 2 > TUPLE_REF_MAX) {
> -        diag_set(ClientError, ER_TUPLE_REF_OVERFLOW);
> -        return NULL;
> -    }
> -    tuple->refs++;
> +    tuple_ref(tuple);
>       /* Remove previous tuple */
>       if (likely(box_tuple_last != NULL))
> -        tuple_unref(box_tuple_last); /* do not throw */
> +        tuple_unref(box_tuple_last);
>       /* Remember current tuple */
>       box_tuple_last = tuple;
>       return tuple;
> @@ -849,41 +857,6 @@ tuple_to_buf(const struct tuple *tuple, char *buf, size_t size);
>   #include "tuple_update.h"
>   #include "errinj.h"
> 
> -/**
> - * \copydoc tuple_ref()
> - * \throws if overflow detected.
> - */
> -static inline void
> -tuple_ref_xc(struct tuple *tuple)
> -{
> -    if (tuple_ref(tuple))
> -        diag_raise();
> -}
> -
> -/**
> - * \copydoc tuple_bless
> - * \throw ER_TUPLE_REF_OVERFLOW
> - */
> -static inline box_tuple_t *
> -tuple_bless_xc(struct tuple *tuple)
> -{
> -    box_tuple_t *blessed = tuple_bless(tuple);
> -    if (blessed == NULL)
> -        diag_raise();
> -    return blessed;
> -}
> -
> -/** Make tuple references exception-friendly in absence of @finally. */
> -struct TupleRefNil {
> -    struct tuple *tuple;
> -    TupleRefNil (struct tuple *arg) :tuple(arg)
> -    { if (tuple) tuple_ref_xc(tuple); }
> -    ~TupleRefNil() { if (tuple) tuple_unref(tuple); }
> -
> -    TupleRefNil(const TupleRefNil&) = delete;
> -    void operator=(const TupleRefNil&) = delete;
> -};
> -
>   /* @copydoc tuple_field_with_type() */
>   static inline const char *
>   tuple_field_with_type_xc(const struct tuple *tuple, uint32_t fieldno,
> diff --git a/src/box/vinyl.c b/src/box/vinyl.c
> index f0d2687..dc0d020 100644
> --- a/src/box/vinyl.c
> +++ b/src/box/vinyl.c
> @@ -3822,7 +3822,6 @@ vinyl_iterator_primary_next(struct iterator *base, struct tuple **ret)
>       assert(base->next = vinyl_iterator_primary_next);
>       struct vinyl_iterator *it = (struct vinyl_iterator *)base;
>       assert(it->lsm->index_id == 0);
> -    struct tuple *tuple;
> 
>       if (it->tx == NULL) {
>           diag_set(ClientError, ER_CURSOR_NO_TRANSACTION);
> @@ -3833,18 +3832,15 @@ vinyl_iterator_primary_next(struct iterator *base, struct tuple **ret)
>           goto fail;
>       }
> 
> -    if (vy_read_iterator_next(&it->iterator, &tuple) != 0)
> +    if (vy_read_iterator_next(&it->iterator, ret) != 0)
>           goto fail;
> -
> -    if (tuple == NULL) {
> +    if (*ret == NULL) {
>           /* EOF. Close the iterator immediately. */
>           vinyl_iterator_close(it);
> -        *ret = NULL;
> -        return 0;
> +    } else {
> +        tuple_bless(*ret);
>       }
> -    *ret = tuple_bless(tuple);
> -    if (*ret != NULL)
> -        return 0;
> +    return 0;
>   fail:
>       vinyl_iterator_close(it);
>       return -1;
> @@ -3890,11 +3886,10 @@ next:
>        * Note, there's no need in vy_tx_track() as the
>        * tuple is already tracked in the secondary index.
>        */
> -    struct tuple *full_tuple;
>       if (vy_point_lookup(it->lsm->pk, it->tx, vy_tx_read_view(it->tx),
> -                tuple, &full_tuple) != 0)
> +                tuple, ret) != 0)
>           goto fail;
> -    if (full_tuple == NULL) {
> +    if (*ret == NULL) {
>           /*
>            * All indexes of a space must be consistent, i.e.
>            * if a tuple is present in one index, it must be
> @@ -3908,10 +3903,9 @@ next:
>                vy_lsm_name(it->lsm), vy_stmt_str(tuple));
>           goto next;
>       }
> -    *ret = tuple_bless(full_tuple);
> -    tuple_unref(full_tuple);
> -    if (*ret != NULL)
> -        return 0;
> +    tuple_bless(*ret);
> +    tuple_unref(*ret);
> +    return 0;
>   fail:
>       vinyl_iterator_close(it);
>       return -1;
> @@ -3997,16 +3991,12 @@ vinyl_index_get(struct index *index, const char *key,
>       const struct vy_read_view **rv = (tx != NULL ? vy_tx_read_view(tx) :
>                         &env->xm->p_global_read_view);
> 
> -    struct tuple *tuple;
> -    if (vy_lsm_full_by_key(lsm, tx, rv, key, part_count, &tuple) != 0)
> +    if (vy_lsm_full_by_key(lsm, tx, rv, key, part_count, ret) != 0)
>           return -1;
> -
> -    if (tuple != NULL) {
> -        *ret = tuple_bless(tuple);
> -        tuple_unref(tuple);
> -        return *ret == NULL ? -1 : 0;
> +    if (*ret != NULL) {
> +        tuple_bless(*ret);
> +        tuple_unref(*ret);
>       }
> -    *ret = NULL;
>       return 0;
>   }
> 
> diff --git a/test/box/misc.result b/test/box/misc.result
> index 8f94f55..f7703ba 100644
> --- a/test/box/misc.result
> +++ b/test/box/misc.result
> @@ -332,7 +332,6 @@ t;
>     - 'box.error.UNKNOWN_UPDATE_OP : 28'
>     - 'box.error.WRONG_COLLATION_OPTIONS : 151'
>     - 'box.error.CURSOR_NO_TRANSACTION : 80'
> -  - 'box.error.TUPLE_REF_OVERFLOW : 86'
>     - 'box.error.ALTER_SEQUENCE : 143'
>     - 'box.error.INVALID_XLOG_NAME : 75'
>     - 'box.error.SAVEPOINT_EMPTY_TX : 60'
> @@ -360,7 +359,7 @@ t;
>     - 'box.error.VINYL_MAX_TUPLE_SIZE : 139'
>     - 'box.error.LOAD_FUNCTION : 99'
>     - 'box.error.INVALID_XLOG : 74'
> -  - 'box.error.PRIV_NOT_GRANTED : 91'
> +  - 'box.error.READ_VIEW_ABORTED : 130'
>     - 'box.error.TRANSACTION_CONFLICT : 97'
>     - 'box.error.GUEST_USER_PASSWORD : 96'
>     - 'box.error.PROC_C : 102'
> @@ -405,7 +404,7 @@ t;
>     - 'box.error.injection : table: <address>
>     - 'box.error.NULLABLE_MISMATCH : 153'
>     - 'box.error.LAST_DROP : 15'
> -  - 'box.error.NO_SUCH_ROLE : 82'
> +  - 'box.error.TUPLE_FORMAT_LIMIT : 16'
>     - 'box.error.DECOMPRESSION : 124'
>     - 'box.error.CREATE_SEQUENCE : 142'
>     - 'box.error.CREATE_USER : 43'
> @@ -414,66 +413,66 @@ t;
>     - 'box.error.SEQUENCE_OVERFLOW : 147'
>     - 'box.error.SYSTEM : 115'
>     - 'box.error.KEY_PART_IS_TOO_LONG : 118'
> -  - 'box.error.TUPLE_FORMAT_LIMIT : 16'
> -  - 'box.error.BEFORE_REPLACE_RET : 53'
> +  - 'box.error.INJECTION : 8'
> +  - 'box.error.INVALID_MSGPACK : 20'
>     - 'box.error.NO_SUCH_SAVEPOINT : 61'
>     - 'box.error.TRUNCATE_SYSTEM_SPACE : 137'
>     - 'box.error.VY_QUOTA_TIMEOUT : 135'
>     - 'box.error.WRONG_INDEX_OPTIONS : 108'
>     - 'box.error.INVALID_VYLOG_FILE : 133'
>     - 'box.error.INDEX_FIELD_COUNT_LIMIT : 127'
> -  - 'box.error.READ_VIEW_ABORTED : 130'
> +  - 'box.error.PRIV_NOT_GRANTED : 91'
>     - 'box.error.USER_MAX : 56'
> -  - 'box.error.PROTOCOL : 104'
> +  - 'box.error.BEFORE_REPLACE_RET : 53'
>     - 'box.error.TUPLE_NOT_ARRAY : 22'
>     - 'box.error.KEY_PART_COUNT : 31'
>     - 'box.error.ALTER_SPACE : 12'
>     - 'box.error.ACTIVE_TRANSACTION : 79'
>     - 'box.error.EXACT_FIELD_COUNT : 38'
>     - 'box.error.DROP_SEQUENCE : 144'
> -  - 'box.error.INVALID_MSGPACK : 20'
>     - 'box.error.MORE_THAN_ONE_TUPLE : 41'
> -  - 'box.error.RTREE_RECT : 101'
> -  - 'box.error.SUB_STMT_MAX : 121'
> +  - 'box.error.UPSERT_UNIQUE_SECONDARY_KEY : 105'
>     - 'box.error.UNKNOWN_REQUEST_TYPE : 48'
> -  - 'box.error.SPACE_EXISTS : 10'
> +  - 'box.error.SUB_STMT_MAX : 121'
>     - 'box.error.PROC_LUA : 32'
> +  - 'box.error.SPACE_EXISTS : 10'
>     - 'box.error.ROLE_NOT_GRANTED : 92'
> +  - 'box.error.UNSUPPORTED : 5'
>     - 'box.error.NO_SUCH_SPACE : 36'
>     - 'box.error.WRONG_INDEX_PARTS : 107'
> -  - 'box.error.DROP_SPACE : 11'
>     - 'box.error.MIN_FIELD_COUNT : 39'
>     - 'box.error.REPLICASET_UUID_MISMATCH : 63'
>     - 'box.error.UPDATE_FIELD : 29'
> +  - 'box.error.INDEX_EXISTS : 85'
>     - 'box.error.COMPRESSION : 119'
>     - 'box.error.INVALID_ORDER : 68'
> -  - 'box.error.INDEX_EXISTS : 85'
>     - 'box.error.SPLICE : 25'
>     - 'box.error.UNKNOWN : 0'
> +  - 'box.error.IDENTIFIER : 70'
>     - 'box.error.DROP_PRIMARY_KEY : 17'
>     - 'box.error.NULLABLE_PRIMARY : 152'
>     - 'box.error.NO_SUCH_SEQUENCE : 145'
>     - 'box.error.RELOAD_CFG : 58'
>     - 'box.error.INVALID_UUID : 64'
> -  - 'box.error.INJECTION : 8'
> +  - 'box.error.DROP_SPACE : 11'
>     - 'box.error.TIMEOUT : 78'
> -  - 'box.error.IDENTIFIER : 70'
>     - 'box.error.ITERATOR_TYPE : 72'
>     - 'box.error.REPLICA_MAX : 73'
> +  - 'box.error.NO_SUCH_ROLE : 82'
>     - 'box.error.MISSING_REQUEST_FIELD : 69'
>     - 'box.error.MISSING_SNAPSHOT : 93'
>     - 'box.error.WRONG_SPACE_OPTIONS : 111'
>     - 'box.error.READONLY : 7'
> -  - 'box.error.UNSUPPORTED : 5'
>     - 'box.error.UPDATE_INTEGER_OVERFLOW : 95'
> +  - 'box.error.RTREE_RECT : 101'
>     - 'box.error.NO_CONNECTION : 77'
>     - 'box.error.INVALID_XLOG_ORDER : 76'
> -  - 'box.error.UPSERT_UNIQUE_SECONDARY_KEY : 105'
> -  - 'box.error.ROLLBACK_IN_SUB_STMT : 123'
>     - 'box.error.WRONG_SCHEMA_VERSION : 109'
> -  - 'box.error.UNSUPPORTED_INDEX_FEATURE : 112'
> -  - 'box.error.INDEX_PART_TYPE_MISMATCH : 24'
> +  - 'box.error.ROLLBACK_IN_SUB_STMT : 123'
> +  - 'box.error.PROTOCOL : 104'
>     - 'box.error.INVALID_XLOG_TYPE : 125'
> +  - 'box.error.INDEX_PART_TYPE_MISMATCH : 24'
> +  - 'box.error.UNSUPPORTED_INDEX_FEATURE : 112'
>   ...
>   test_run:cmd("setopt delimiter ''");
>   ---
> diff --git a/test/box/select.result b/test/box/select.result
> index 4aed706..b3ee6cd 100644
> --- a/test/box/select.result
> +++ b/test/box/select.result
> @@ -619,31 +619,62 @@ collectgarbage('collect')
>   ---
>   - 0
>   ...
> +-- gh-3224 resurrect tuple bigrefs
> +collectgarbage('stop')
> +---
> +- 0
> +...
>   s = box.schema.space.create('select', { temporary = true })
>   ---
>   ...
>   index = s:create_index('primary', { type = 'tree' })
>   ---
>   ...
> -a = s:insert{0}
> +_ = s:insert{0}
> +---
> +...
> +_ = s:insert{1}
> +---
> +...
> +_ = s:insert{2}
> +---
> +...
> +_ = s:insert{3}
> +---
> +...
> +lots_of_links = setmetatable({}, {__mode = 'v'})
>   ---
>   ...
> -lots_of_links = {}
> +i = 0
> +---
> +...
> +while (i < 33000) do table.insert(lots_of_links, s:get{0}) i = i + 1 end
> +---
> +...
> +while (i < 66000) do table.insert(lots_of_links, s:get{1}) i = i + 1 end
> +---
> +...
> +while (i < 100000) do table.insert(lots_of_links, s:get{2}) i = i + 1 end
>   ---
>   ...
>   ref_count = 0
>   ---
>   ...
> -while (true) do table.insert(lots_of_links, s:get{0}) ref_count = ref_count + 1 end
> +for k, v in pairs(lots_of_links) do ref_count = ref_count + 1 end
>   ---
> -- error: Tuple reference counter overflow
>   ...
>   ref_count
>   ---
> -- 65531
> +- 100000
>   ...
> -lots_of_links = {}
> +-- check that tuples are deleted after gc is activated
> +collectgarbage('collect')
>   ---
> +- 0
> +...
> +lots_of_links
> +---
> +- []
>   ...
>   s:drop()
>   ---
> diff --git a/test/box/select.test.lua b/test/box/select.test.lua
> index 54c2ecc..3400bda 100644
> --- a/test/box/select.test.lua
> +++ b/test/box/select.test.lua
> @@ -124,12 +124,25 @@ test.random(s.index[0], 48)
>   s:drop()
> 
>   collectgarbage('collect')
> +
> +-- gh-3224 resurrect tuple bigrefs
> +
> +collectgarbage('stop')
>   s = box.schema.space.create('select', { temporary = true })
>   index = s:create_index('primary', { type = 'tree' })
> -a = s:insert{0}
> -lots_of_links = {}
> +_ = s:insert{0}
> +_ = s:insert{1}
> +_ = s:insert{2}
> +_ = s:insert{3}
> +lots_of_links = setmetatable({}, {__mode = 'v'})
> +i = 0
> +while (i < 33000) do table.insert(lots_of_links, s:get{0}) i = i + 1 end
> +while (i < 66000) do table.insert(lots_of_links, s:get{1}) i = i + 1 end
> +while (i < 100000) do table.insert(lots_of_links, s:get{2}) i = i + 1 end
>   ref_count = 0
> -while (true) do table.insert(lots_of_links, s:get{0}) ref_count = ref_count + 1 end
> +for k, v in pairs(lots_of_links) do ref_count = ref_count + 1 end
>   ref_count
> -lots_of_links = {}
> +-- check that tuples are deleted after gc is activated
> +collectgarbage('collect')
> +lots_of_links
>   s:drop()
> diff --git a/test/unit/CMakeLists.txt b/test/unit/CMakeLists.txt
> index dbc02cd..aef5316 100644
> --- a/test/unit/CMakeLists.txt
> +++ b/test/unit/CMakeLists.txt
> @@ -192,3 +192,6 @@ target_link_libraries(vy_cache.test ${ITERATOR_TEST_LIBS})
> 
>   add_executable(coll.test coll.cpp)
>   target_link_libraries(coll.test core unit ${ICU_LIBRARIES} misc)
> +
> +add_executable(tuple_bigref.test tuple_bigref.c)
> +target_link_libraries(tuple_bigref.test tuple unit)
> diff --git a/test/unit/tuple_bigref.c b/test/unit/tuple_bigref.c
> new file mode 100644
> index 0000000..a24ab91
> --- /dev/null
> +++ b/test/unit/tuple_bigref.c
> @@ -0,0 +1,174 @@
> +#include "vy_iterators_helper.h"
> +#include "memory.h"
> +#include "fiber.h"
> +#include "unit.h"
> +#include <msgpuck.h>
> +#include "trivia/util.h"
> +
> +enum {
> +    BIGREF_DIFF = 10,
> +    BIGREF_COUNT = 70003,
> +    BIGREF_CAPACITY = 107,
> +};
> +
> +static char tuple_buf[64];
> +static char *tuple_end = tuple_buf;
> +
> +/**
> + * This function creates new tuple with refs == 1.
> + */
> +static inline struct tuple *
> +create_tuple()
> +{
> +    struct tuple *ret =
> +        tuple_new(box_tuple_format_default(), tuple_buf, tuple_end);
> +    tuple_ref(ret);
> +    return ret;
> +}
> +
> +/**
> + * This test performs overall check of bigrefs.
> + * What it checks:
> + * 1) Till refs <= TUPLE_REF_MAX it shows number of refs
> + * of tuple and it isn't a bigref.
> + * 2) When refs > TUPLE_REF_MAX first 15 bits of it becomes
> + * index of bigref and the last bit becomes true which
> + * shows that it is bigref.
> + * 3) Each of tuple has its own number of refs, but all
> + * these numbers more than it is needed for getting a bigref.
> + * 4) Indexes of bigrefs are given sequentially.
> + * 5) After some tuples are sequentially deleted all of
> + * others bigrefs are fine. In this test BIGREF_CAPACITY
> + * tuples created and each of their ref counter increased
> + * to (BIGREF_COUNT - index of tuple). Tuples are created
> + * consistently.
> + */
> +static int
> +test_bigrefs_1()
> +{
> +    struct tuple **tuples = (struct tuple **) malloc(BIGREF_CAPACITY *
> +                             sizeof(*tuples));
> +    for(int i = 0; i < BIGREF_CAPACITY; ++i)
> +        tuples[i] = create_tuple();
> +    for(int i = 0; i < BIGREF_CAPACITY; ++i) {
> +        assert(tuples[i]->refs == 1);
> +        for(int j = 1; j < TUPLE_REF_MAX; ++j)
> +            tuple_ref(tuples[i]);
> +        assert(! tuples[i]->is_bigref);
> +        tuple_ref(tuples[i]);
> +        assert(tuples[i]->is_bigref);
> +        for(int j = TUPLE_REF_MAX + 1; j < BIGREF_COUNT - i; ++j)
> +            tuple_ref(tuples[i]);
> +        assert(tuples[i]->is_bigref && tuples[i]->ref_index == i);
> +    }
> +    for(int i = 0; i < BIGREF_CAPACITY; ++i) {
> +        for(int j = 1; j < BIGREF_COUNT - i; ++j)
> +            tuple_unref(tuples[i]);
> +        assert(tuples[i]->refs == 1);
> +        tuple_unref(tuples[i]);
> +    }
> +    free(tuples);
> +    return 0;
> +}
> +
> +/**
> + * This test checks that bigrefs works fine after being
> + * created and destroyed BIGREF_DIFF times.
> + */
> +static int
> +test_bigrefs_2()
> +{
> +    struct tuple *tuple = create_tuple();
> +    for(int i = 0; i < 2; ++i) {
> +        assert(tuple->refs == 1);
> +        for(int j = 1; j < BIGREF_COUNT; ++j)
> +            tuple_ref(tuple);
> +        assert(tuple->is_bigref && tuple->ref_index == 0);
> +        for(int j = 1; j < BIGREF_COUNT; ++j)
> +            tuple_unref(tuple);
> +        assert(!tuple->is_bigref && tuple->refs == 1);
> +    }
> +    tuple_unref(tuple);
> +    return 0;
> +}
> +
> +/**
> + * This test checks that indexes are given and free as
> + * intended.
> + */
> +static int
> +test_bigrefs_3()
> +{
> +    struct tuple **tuples = (struct tuple **) malloc(BIGREF_CAPACITY *
> +                             sizeof(*tuples));
> +    for(int i = 0; i < BIGREF_CAPACITY; ++i)
> +        tuples[i] = create_tuple();
> +    for(int i = 0; i < BIGREF_CAPACITY; ++i) {
> +        assert(tuples[i]->refs == 1);
> +        for(int j = 1; j < BIGREF_COUNT; ++j)
> +            tuple_ref(tuples[i]);
> +        assert(tuples[i]->is_bigref);
> +    }
> +    /** Checks that indexes are given consistently */
> +    for(int i = 0; i < BIGREF_CAPACITY; i = i + BIGREF_DIFF) {
> +        for(int j = 1; j < BIGREF_COUNT; ++j)
> +            tuple_unref(tuples[i]);
> +        assert(!tuples[i]->is_bigref);
> +    }
> +    for(int i = 0; i < BIGREF_CAPACITY; i = i + BIGREF_DIFF) {
> +        assert(tuples[i]->refs == 1);
> +        for(int j = 1; j < BIGREF_COUNT; ++j)
> +            tuple_ref(tuples[i]);
> +        assert(tuples[i]->is_bigref && tuples[i]->ref_index == i);
> +    }
> +
> +    /**
> +     * Checks that capacity of bigref_list decreased
> +     * as intended (for gcov and gdb checks because
> +     * capacity cannot be seen outside of tuple.c).
> +     */
> +    int tmp_indexes[] = {1, 10, 20, 100};
> +    for(int i = BIGREF_CAPACITY - 1; i >= 0; --i) {
> +        if(i == tmp_indexes[0] || i == tmp_indexes[1] ||
> +           i == tmp_indexes[2] || i == tmp_indexes[3])
> +            continue;
> +        for(int j = 1; j < BIGREF_COUNT; ++j)
> +            tuple_unref(tuples[i]);
> +        assert(tuples[i]->refs == 1);
> +        tuple_unref(tuples[i]);
> +    }
> +    for(int i = 3; i >= 0; --i) {
> +        assert(tuples[tmp_indexes[i]]->ref_index == tmp_indexes[i]);
> +        for(int j = 1; j < BIGREF_COUNT; ++j)
> +            tuple_unref(tuples[tmp_indexes[i]]);
> +        assert(tuples[tmp_indexes[i]]->refs == 1);
> +        tuple_unref(tuples[tmp_indexes[i]]);
> +    }
> +    free(tuples);
> +    return 0;
> +}
> +
> +int
> +main()
> +{
> +    header();
> +    plan(3);
> +
> +    memory_init();
> +    fiber_init(fiber_c_invoke);
> +    tuple_init(NULL);
> +
> +    tuple_end = mp_encode_array(tuple_end, 1);
> +    tuple_end = mp_encode_uint(tuple_end, 2);
> +
> +    ok(test_bigrefs_1() == 0, "Overall test passed.");
> +    ok(test_bigrefs_2() == 0, "Create/destroy test passed.");
> +    ok(test_bigrefs_3() == 0, "Non-consistent indexes test passed.");
> +
> +    tuple_free();
> +    fiber_free();
> +    memory_free();
> +
> +    footer();
> +    check_plan();
> +}
> diff --git a/test/unit/tuple_bigref.result b/test/unit/tuple_bigref.result
> new file mode 100644
> index 0000000..91b9a0f
> --- /dev/null
> +++ b/test/unit/tuple_bigref.result
> @@ -0,0 +1,6 @@
> +    *** main ***
> +1..3
> +ok 1 - Overall test passed.
> +ok 2 - Create/destroy test passed.
> +ok 3 - Non-consistent indexes test passed.
> +    *** main: done ***