mdbx: remove extra prep_backlog_data() inside update_gc().

Change-Id: Ic6250159c5abe03ed0705c7a432c5eb5b17b790e
mdbx-test: partialy revert long_stochastic.sh changes.
2025-12-18 17:52:22 +08:00 · 2019-07-09 15:58:48 +03:00 · 2019-07-09 14:31:27 +03:00 · 2019-07-09 03:27:23 +03:00 · 2019-07-09 01:39:32 +03:00 · 2019-07-09 00:41:04 +03:00
36 changed files with 1960 additions and 657 deletions
--- a/.travis.yml
+++ b/.travis.yml
@@ -1,6 +1,5 @@
 language: c
-sudo: required
-dist: trusty
+dist: xenial

 compiler:
 - gcc
@@ -26,5 +25,5 @@ addons:
      description: "Build submitted via Travis CI"
    notification_email: leo@yuriev.ru
    build_command_prepend: "make clean"
-    build_command:   "make all -j 4"
+    build_command: "make all -j 2"
    branch_pattern: coverity_scan
--- a/4
+++ b/4
@@ -82,11 +82,11 @@ clean:
 	rm -rf $(TOOLS) mdbx_test @* *.[ao] *.[ls]o *~ tmp.db/* *.gcov *.log *.err src/*.o test/*.o

 check:	all
-	rm -f $(TESTDB) $(TESTLOG) && (set -o pipefail; ./mdbx_test --pathname=$(TESTDB) --dont-cleanup-after basic | tee -a $(TESTLOG) | tail -n 42) \
+	rm -f $(TESTDB) $(TESTLOG) && (set -o pipefail; ./mdbx_test --repeat=42 --pathname=$(TESTDB) --dont-cleanup-after basic | tee -a $(TESTLOG) | tail -n 42) \
 	&& ./mdbx_chk -vvn $(TESTDB) && ./mdbx_chk -vvn $(TESTDB)-copy

 check-singleprocess:	all
-	rm -f $(TESTDB) $(TESTLOG) && (set -o pipefail; ./mdbx_test --pathname=$(TESTDB) --dont-cleanup-after --hill --copy | tee -a $(TESTLOG) | tail -n 42) \
+	rm -f $(TESTDB) $(TESTLOG) && (set -o pipefail; ./mdbx_test --repeat=42 --pathname=$(TESTDB) --dont-cleanup-after --hill --copy | tee -a $(TESTLOG) | tail -n 42) \
 	&& ./mdbx_chk -vvn $(TESTDB) && ./mdbx_chk -vvn $(TESTDB)-copy

 check-fault:	all
--- a/README-RU.md
+++ b/README-RU.md
@@ -171,31 +171,18 @@ Amplification Factor) и RAF (Read Amplification Factor) также Olog(N).
 Доработки и усовершенствования относительно LMDB
 ================================================

-1. Утилита `mdbx_chk` для проверки целостности структуры БД.
-
-2. Автоматическое динамическое управление размером БД согласно
+1. Автоматическое динамическое управление размером БД согласно
 параметрам задаваемым функцией `mdbx_env_set_geometry()`, включая шаг
 приращения и порог уменьшения размера БД, а также выбор размера
 страницы. Соответственно, это позволяет снизить фрагментированность
 файла БД на диске и освободить место, в том числе в **Windows**.

-3. Автоматическая без-затратная компактификация БД путем возврата
+2. Автоматическая без-затратная компактификация БД путем возврата
 освобождающихся страниц в область нераспределенного резерва в конце
 файла данных. При этом уменьшается количество страниц находящихся в
 памяти и участвующих в в обмене с диском.

-4. Поддержка ключей и значений нулевой длины, включая сортированные
-дубликаты.
-
-5. Возможность связать с каждой завершаемой транзакцией до 3
-дополнительных маркеров посредством `mdbx_canary_put()`, и прочитать их
-в транзакции чтения посредством `mdbx_canary_get()`.
-
-6. Возможность посредством `mdbx_replace()` обновить или удалить запись
-с получением предыдущего значения данных, а также адресно изменить
-конкретное multi-значение.
-
-7. Режим `LIFO RECLAIM`.
+3. Режим `LIFO RECLAIM`.

    Для повторного использования выбираются не самые старые, а
    самые новые страницы из доступных. За счет этого цикл
@@ -209,9 +196,27 @@ Amplification Factor) и RAF (Read Amplification Factor) также Olog(N).
    многократное увеличение производительности по записи
    (обновлению данных).

-8. Генерация последовательностей посредством `mdbx_dbi_sequence()`.
+4. Быстрая оценка количества элементов попадающих в запрашиваемый
+диапазон значений ключа посредством функций `mdbx_estimate_range()`,
+`mdbx_estimate_move()` и `mdbx_estimate_distance()` для выбора
+оптимального плана выполнения запроса.

-9. Обработчик `OOM-KICK`.
+5. Утилита `mdbx_chk` для проверки целостности структуры БД.
+
+6. Поддержка ключей и значений нулевой длины, включая сортированные
+дубликаты.
+
+7. Возможность связать с каждой завершаемой транзакцией до 3
+дополнительных маркеров посредством `mdbx_canary_put()`, и прочитать их
+в транзакции чтения посредством `mdbx_canary_get()`.
+
+8. Возможность посредством `mdbx_replace()` обновить или удалить запись
+с получением предыдущего значения данных, а также адресно изменить
+конкретное multi-значение.
+
+9. Генерация последовательностей посредством `mdbx_dbi_sequence()`.
+
+10. Обработчик `OOM-KICK`.

    Посредством `mdbx_env_set_oomfunc()` может быть установлен
    внешний обработчик (callback), который будет вызван при
@@ -232,83 +237,83 @@ Amplification Factor) и RAF (Read Amplification Factor) также Olog(N).
    * прервать текущую операцию изменения данных с возвратом кода
      ошибки.

-10. Возможность открыть БД в эксклюзивном режиме посредством флага
-`MDBX_EXCLUSIVE`.
+11. Возможность открыть БД в эксклюзивном режиме посредством флага
+    `MDBX_EXCLUSIVE`, в том числе на сетевом носителе.

-11. Возможность получить отставание текущей транзакции чтения от
+12. Возможность получить отставание текущей транзакции чтения от
 последней версии данных в БД посредством `mdbx_txn_straggler()`.

-12. Возможность явно запросить обновление существующей записи, без
+13. Возможность явно запросить обновление существующей записи, без
 создания новой посредством флажка `MDBX_CURRENT` для `mdbx_put()`.

-13. Исправленный вариант `mdbx_cursor_count()`, возвращающий корректное
+14. Исправленный вариант `mdbx_cursor_count()`, возвращающий корректное
 количество дубликатов для всех типов таблиц и любого положения курсора.

-14. Возможность получить посредством `mdbx_env_info()` дополнительную
+15. Возможность получить посредством `mdbx_env_info()` дополнительную
 информацию, включая номер самой старой версии БД (снимка данных),
 который используется одним из читателей.

-15. Функция `mdbx_del()` не игнорирует дополнительный (уточняющий)
+16. Функция `mdbx_del()` не игнорирует дополнительный (уточняющий)
 аргумент `data` для таблиц без дубликатов (без флажка `MDBX_DUPSORT`), а
 при его ненулевом значении всегда использует его для сверки с удаляемой
 записью.

-16. Возможность открыть dbi-таблицу, одновременно с установкой
+17. Возможность открыть dbi-таблицу, одновременно с установкой
 компараторов для ключей и данных, посредством `mdbx_dbi_open_ex()`.

-17. Возможность посредством `mdbx_is_dirty()` определить находятся ли
+18. Возможность посредством `mdbx_is_dirty()` определить находятся ли
 некоторый ключ или данные в "грязной" странице БД. Таким образом,
 избегая лишнего копирования данных перед выполнением модифицирующих
 операций (значения, размещенные в "грязных" страницах, могут быть
 перезаписаны при изменениях, иначе они будут неизменны).

-18. Корректное обновление текущей записи, в том числе сортированного
+19. Корректное обновление текущей записи, в том числе сортированного
 дубликата, при использовании режима `MDBX_CURRENT` в
 `mdbx_cursor_put()`.

-19. Возможность узнать есть ли за текущей позицией курсора строка данных
+20. Возможность узнать есть ли за текущей позицией курсора строка данных
 посредством `mdbx_cursor_eof()`.

-20. Дополнительный код ошибки `MDBX_EMULTIVAL`, который возвращается из
+21. Дополнительный код ошибки `MDBX_EMULTIVAL`, который возвращается из
 `mdbx_put()` и `mdbx_replace()` при попытке выполнить неоднозначное
 обновление или удаления одного из нескольких значений с одним ключом.

-21. Возможность посредством `mdbx_get_ex()` получить значение по
+22. Возможность посредством `mdbx_get_ex()` получить значение по
 заданному ключу, одновременно с количеством дубликатов.

-22. Наличие функций `mdbx_cursor_on_first()` и `mdbx_cursor_on_last()`,
+23. Наличие функций `mdbx_cursor_on_first()` и `mdbx_cursor_on_last()`,
 которые позволяют быстро выяснить стоит ли курсор на первой/последней
 позиции.

-23. Возможность автоматического формирования контрольных точек (сброса
+24. Возможность автоматического формирования контрольных точек (сброса
 данных на диск) при накоплении заданного объёма изменений,
 устанавливаемого функцией `mdbx_env_set_syncbytes()`.

-24. Управление отладкой и получение отладочных сообщений посредством
+25. Управление отладкой и получение отладочных сообщений посредством
 `mdbx_setup_debug()`.

-25. Функция `mdbx_env_pgwalk()` для обхода всех страниц БД.
+26. Функция `mdbx_env_pgwalk()` для обхода всех страниц БД.

-26. Три мета-страницы вместо двух, что позволяет гарантированно
+27. Три мета-страницы вместо двух, что позволяет гарантированно
 консистентно обновлять слабые контрольные точки фиксации без риска
 повредить крайнюю сильную точку фиксации.

-27. Гарантия сохранности БД в режиме `WRITEMAP+MAPSYNC`.
+28. Гарантия сохранности БД в режиме `WRITEMAP+MAPSYNC`.
  > В текущей версии _libmdbx_ вам предоставляется выбор между безопасным
  > режимом (по умолчанию) асинхронной фиксации, и режимом `UTTERLY_NOSYNC`
  > когда при системной аварии есть шанс полного разрушения БД как в LMDB.
  > Для подробностей смотрите раздел
  > [Сохранность данных в режиме асинхронной фиксации](#Сохранность-данных-в-режиме-асинхронной-фиксации).

-28. Возможность закрыть БД в "грязном" состоянии (без сброса данных и
+29. Возможность закрыть БД в "грязном" состоянии (без сброса данных и
 формирования сильной точки фиксации) посредством `mdbx_env_close_ex()`.

-29. При завершении читающих транзакций, открытые в них DBI-хендлы не
+30. При завершении читающих транзакций, открытые в них DBI-хендлы не
 закрываются и не теряются при завершении таких транзакций посредством
 `mdbx_txn_abort()` или `mdbx_txn_reset()`. Что позволяет избавится от ряда
 сложно обнаруживаемых ошибок.

-30. Все курсоры, как в транзакциях только для чтения, так и в пишущих,
+31. Все курсоры, как в транзакциях только для чтения, так и в пишущих,
 могут быть переиспользованы посредством `mdbx_cursor_renew()` и ДОЛЖНЫ
 ОСВОБОЖДАТЬСЯ ЯВНО.
  >
@@ -682,11 +687,11 @@ $ objdump -f -h -j .text libmdbx.so
 libmdbx.so:     file format elf64-x86-64
 architecture: i386:x86-64, flags 0x00000150:
 HAS_SYMS, DYNAMIC, D_PAGED
-start address 0x000030e0
+start address 0x0000000000003870

 Sections:
 Idx Name          Size      VMA               LMA               File off  Algn
- 11 .text         00014d84  00000000000030e0  00000000000030e0  000030e0  2**4
+ 11 .text         000173d4  0000000000003870  0000000000003870  00003870  2**4
                  CONTENTS, ALLOC, LOAD, READONLY, CODE

 ```
--- a/README.md
+++ b/README.md
@@ -170,30 +170,18 @@ regular maintenance. Backups can be made on the fly on working DB
 Improvements over LMDB
 ======================

-1. `mdbx_chk` tool for DB integrity check.
-
-2. Automatic dynamic DB size management according to the parameters
-specified by `mdbx_env_set_geometry()` function. Including including
+1. Automatic dynamic DB size management according to the parameters
+specified by `mdbx_env_set_geometry()` function. Including
 growth step and truncation threshold, as well as the choice of page
 size.

-3. Automatic returning of freed pages into unallocated space at the end
-of database file with optionally automatic shrinking it. This reduces
+2. Automatic returning of freed pages into unallocated space at the end
+of database file, with optionally automatic shrinking it. This reduces
 amount of pages resides in RAM and circulated in disk I/O. In fact
 _libmdbx_ constantly performs DB compactification, without spending
 additional resources for that.

-4. Support for keys and values of zero length, including sorted
-duplicates.
-
-5. Ability to assign up to 3 markers to commiting transaction with
-`mdbx_canary_put()` and then get them in read transaction by
-`mdbx_canary_get()`.
-
-6. Ability to update or delete record and get previous value via
-`mdbx_replace()` Also can update specific multi-value.
-
-7. `LIFO RECLAIM` mode:
+3. `LIFO RECLAIM` mode:

    The newest pages are picked for reuse instead of the oldest. This allows
    to minimize reclaim loop and make it execution time independent of total
@@ -204,9 +192,25 @@ duplicates.
    [BBWC](https://en.wikipedia.org/wiki/Disk_buffer#Write_acceleration)
    this may greatly improve write performance.

-8. Sequence generation via `mdbx_dbi_sequence()`.
+4. Fast estimation of range query result size via functions `mdbx_estimate_range()`,
+`mdbx_estimate_move()` and `mdbx_estimate_distance()`. E.g. for selection the
+optimal query execution plan.

-9. `OOM-KICK` callback.
+5. `mdbx_chk` tool for DB integrity check.
+
+6. Support for keys and values of zero length, including sorted
+duplicates.
+
+7. Ability to assign up to 3 markers to commiting transaction with
+`mdbx_canary_put()` and then get them in read transaction by
+`mdbx_canary_get()`.
+
+8. Ability to update or delete record and get previous value via
+`mdbx_replace()`. Also can update specific multi-value.
+
+9. Sequence generation via `mdbx_dbi_sequence()`.
+
+10. `OOM-KICK` callback.

    `mdbx_env_set_oomfunc()` allows to set a callback, which will be called
    in the event of DB space exhausting during long-time read transaction in
@@ -224,75 +228,75 @@ duplicates.

    * abort current write transaction with returning error code.

-10. Ability to open DB in exclusive mode with `MDBX_EXCLUSIVE` flag.
+11. Ability to open DB in exclusive mode with `MDBX_EXCLUSIVE` flag.

-11. Ability to get how far current read-only snapshot is from latest
+12. Ability to get how far current read-only snapshot is from latest
 version of the DB by `mdbx_txn_straggler()`.

-12. Ability to explicitly request update of present record without
+13. Ability to explicitly request update of present record without
 creating new record. Implemented as `MDBX_CURRENT` flag for
 `mdbx_put()`.

-13. Fixed `mdbx_cursor_count()`, which returns correct count of
+14. Fixed `mdbx_cursor_count()`, which returns correct count of
 duplicated for all table types and any cursor position.

-14. `mdbx_env_info()` to getting additional info, including number of
+15. `mdbx_env_info()` to getting additional info, including number of
 the oldest snapshot of DB, which is used by one of the readers.

-15. `mdbx_del()` doesn't ignore additional argument (specifier) `data`
+16. `mdbx_del()` doesn't ignore additional argument (specifier) `data`
 for tables without duplicates (without flag `MDBX_DUPSORT`), if `data`
 is not null then always uses it to verify record, which is being
 deleted.

-16. Ability to open dbi-table with simultaneous setup of comparators for
+17. Ability to open dbi-table with simultaneous setup of comparators for
 keys and values, via `mdbx_dbi_open_ex()`.

-17. `mdbx_is_dirty()`to find out if key or value is on dirty page, that
+18. `mdbx_is_dirty()`to find out if key or value is on dirty page, that
 useful to avoid copy-out before updates.

-18. Correct update of current record in `MDBX_CURRENT` mode of
+19. Correct update of current record in `MDBX_CURRENT` mode of
 `mdbx_cursor_put()`, including sorted duplicated.

-19. Check if there is a row with data after current cursor position via
+20. Check if there is a row with data after current cursor position via
 `mdbx_cursor_eof()`.

-20. Additional error code `MDBX_EMULTIVAL`, which is returned by
+21. Additional error code `MDBX_EMULTIVAL`, which is returned by
 `mdbx_put()` and `mdbx_replace()` in case is ambiguous update or delete.

-21. Ability to get value by key and duplicates count by `mdbx_get_ex()`.
+22. Ability to get value by key and duplicates count by `mdbx_get_ex()`.

-22. Functions `mdbx_cursor_on_first()` and `mdbx_cursor_on_last()`,
+23. Functions `mdbx_cursor_on_first()` and `mdbx_cursor_on_last()`,
 which allows to know if cursor is currently on first or last position
 respectively.

-23. Automatic creation of synchronization points (flush changes to
+24. Automatic creation of synchronization points (flush changes to
 persistent storage) when changes reach set threshold (threshold can be
 set by `mdbx_env_set_syncbytes()`).

-24. Control over debugging and receiving of debugging messages via
+25. Control over debugging and receiving of debugging messages via
 `mdbx_setup_debug()`.

-25. Function `mdbx_env_pgwalk()` for page-walking all pages in DB.
+26. Function `mdbx_env_pgwalk()` for page-walking all pages in DB.

-26. Three meta-pages instead of two, this allows to guarantee
+27. Three meta-pages instead of two, this allows to guarantee
 consistently update weak sync-points without risking to corrupt last
 steady sync-point.

-27. Guarantee of DB integrity in `WRITEMAP+MAPSYNC` mode:
+28. Guarantee of DB integrity in `WRITEMAP+MAPSYNC` mode:
  > Current _libmdbx_ gives a choice of safe async-write mode (default)
  > and `UTTERLY_NOSYNC` mode which may result in full
  > DB corruption during system crash as with LMDB. For details see
  > [Data safety in async-write mode](#data-safety-in-async-write-mode).

-28. Ability to close DB in "dirty" state (without data flush and
+29. Ability to close DB in "dirty" state (without data flush and
 creation of steady synchronization point) via `mdbx_env_close_ex()`.

-29. If read transaction is aborted via `mdbx_txn_abort()` or
+30. If read transaction is aborted via `mdbx_txn_abort()` or
 `mdbx_txn_reset()` then DBI-handles, which were opened in it, aren't
 closed or deleted. This allows to avoid several types of hard-to-debug
 errors.

-30. All cursors in all read and write transactions can be reused by
+31. All cursors in all read and write transactions can be reused by
 `mdbx_cursor_renew()` and MUST be freed explicitly.
  > ## Caution, please pay attention!
  >
@@ -594,11 +598,11 @@ $ objdump -f -h -j .text libmdbx.so
 libmdbx.so:     file format elf64-x86-64
 architecture: i386:x86-64, flags 0x00000150:
 HAS_SYMS, DYNAMIC, D_PAGED
-start address 0x000030e0
+start address 0x0000000000003870

 Sections:
 Idx Name          Size      VMA               LMA               File off  Algn
- 11 .text         00014d84  00000000000030e0  00000000000030e0  000030e0  2**4
+ 11 .text         000173d4  0000000000003870  0000000000003870  00003870  2**4
                  CONTENTS, ALLOC, LOAD, READONLY, CODE

 ```
--- a/libmdbx.files
+++ b/libmdbx.files
@@ -3,7 +3,9 @@ README-RU.md
 pcrf_test/CMakeLists.txt
 src/tools/CMakeLists.txt
 test/CMakeLists.txt
+test/append.cc
 test/copy.cc
+test/ttl.cc
 tutorial/CMakeLists.txt
 tutorial/sample-mdbx.c
 AUTHORS
--- a/mdbx.h
+++ b/mdbx.h
@@ -168,7 +168,7 @@ typedef pthread_t mdbx_tid_t;
 /*--------------------------------------------------------------------------*/

 #define MDBX_VERSION_MAJOR 0
-#define MDBX_VERSION_MINOR 2
+#define MDBX_VERSION_MINOR 3

 #if defined(LIBMDBX_EXPORTS)
 #define LIBMDBX_API __dll_export
@@ -1666,11 +1666,16 @@ typedef enum {
  MDBX_subpage_dupfixed_leaf
 } MDBX_page_type_t;

-typedef int MDBX_pgvisitor_func(uint64_t pgno, unsigned number, void *ctx,
-                                int deep, const char *dbi, size_t page_size,
-                                MDBX_page_type_t type, size_t nentries,
-                                size_t payload_bytes, size_t header_bytes,
-                                size_t unused_bytes);
+#define MDBX_PGWALK_MAIN ((const char *)((ptrdiff_t)0))
+#define MDBX_PGWALK_GC ((const char *)((ptrdiff_t)-1))
+#define MDBX_PGWALK_META ((const char *)((ptrdiff_t)-2))
+
+typedef int
+MDBX_pgvisitor_func(const uint64_t pgno, const unsigned number, void *const ctx,
+                    const int deep, const char *const dbi,
+                    const size_t page_size, const MDBX_page_type_t type,
+                    const size_t nentries, const size_t payload_bytes,
+                    const size_t header_bytes, const size_t unused_bytes);
 LIBMDBX_API int mdbx_env_pgwalk(MDBX_txn *txn, MDBX_pgvisitor_func *visitor,
                                void *ctx);

@@ -1695,6 +1700,54 @@ LIBMDBX_API int mdbx_cursor_on_first(MDBX_cursor *mc);
 /* Returns: MDBX_RESULT_TRUE, MDBX_RESULT_FALSE or Error code. */
 LIBMDBX_API int mdbx_cursor_on_last(MDBX_cursor *mc);

+/* Estimates the distance between cursors as the number of elements.
+ * Both cursors must be initialized for the same DBI.
+ *
+ * [in] cursor_a          The first cursor for estimation.
+ * [in] cursor_b          The second cursor for estimation.
+ * [out] distance_items   A pointer to store estimated distance value,
+ *                        i.e. *distance_items = distance(a - b).
+ *
+ * Returns A non-zero error value on failure and 0 on success. */
+LIBMDBX_API int mdbx_estimate_distance(const MDBX_cursor *first,
+                                       const MDBX_cursor *last,
+                                       ptrdiff_t *distance_items);
+
+/* Estimates the move distance, i.e. between the current cursor position and
+ * next position after the specified move-operation with given key and data.
+ * Current cursor position and state are preserved.
+ *
+ * [in] cursor            Cursor for estimation.
+ * [in,out] key           The key for a retrieved item.
+ * [in,out] data          The data of a retrieved item.
+ * [in] op                A cursor operation MDBX_cursor_op.
+ * [out] distance_items   A pointer to store estimated move distance
+ *                        as the number of elements.
+ *
+ * Returns A non-zero error value on failure and 0 on success. */
+LIBMDBX_API int mdbx_estimate_move(const MDBX_cursor *cursor, MDBX_val *key,
+                                   MDBX_val *data, MDBX_cursor_op move_op,
+                                   ptrdiff_t *distance_items);
+
+/* Estimates the size of a range in the number of elements.
+ *
+ * [in] txn               A transaction handle returned by mdbx_txn_begin().
+ * [in] dbi               A database handle returned by mdbx_dbi_open().
+ * [in] begin_key         The key of range beginning or NULL for explicit FIRST.
+ * [in] begin_data        Optional additional data to seeking among sorted
+ *                        duplicates. Only for MDBX_DUPSORT, NULL otherwise.
+ * [in] end_key           The key of range ending or NULL for explicit LAST.
+ * [in] end_data          Optional additional data to seeking among sorted
+ *                        duplicates. Only for MDBX_DUPSORT, NULL otherwise.
+ * [out] distance_items   A pointer to store range estimation result.
+ *
+ * Returns A non-zero error value on failure and 0 on success. */
+#define MDBX_EPSILON ((MDBX_val *)((ptrdiff_t)-1))
+LIBMDBX_API int mdbx_estimate_range(MDBX_txn *txn, MDBX_dbi dbi,
+                                    MDBX_val *begin_key, MDBX_val *begin_data,
+                                    MDBX_val *end_key, MDBX_val *end_data,
+                                    ptrdiff_t *size_items);
+
 LIBMDBX_API int mdbx_replace(MDBX_txn *txn, MDBX_dbi dbi, MDBX_val *key,
                             MDBX_val *new_data, MDBX_val *old_data,
                             unsigned flags);
--- a/src/bits.h
+++ b/src/bits.h
@@ -29,6 +29,12 @@
 #ifndef _GNU_SOURCE
 #   define _GNU_SOURCE 1
 #endif
+#ifndef _POSIX_C_SOURCE
+#   define _POSIX_C_SOURCE 200112L
+#endif
+#ifndef _XOPEN_SOURCE
+#   define _XOPEN_SOURCE 500
+#endif
 #ifndef _FILE_OFFSET_BITS
 #   define _FILE_OFFSET_BITS 64
 #endif
@@ -253,11 +259,14 @@ typedef struct MDBX_reader {
  volatile mdbx_pid_t mr_pid;
  /* The thread ID of the thread owning this txn. */
  volatile mdbx_tid_t mr_tid;
+  /* The number of pages used in the reader's MVCC snapshot,
+   * i.e. the value of meta->mm_geo.next and txn->mt_next_pgno */
+  volatile pgno_t mr_snapshot_pages;

  /* cache line alignment */
-  uint8_t pad[MDBX_CACHELINE_SIZE -
-              (sizeof(txnid_t) + sizeof(mdbx_pid_t) + sizeof(mdbx_tid_t)) %
-                  MDBX_CACHELINE_SIZE];
+  uint8_t pad[MDBX_CACHELINE_SIZE - (sizeof(txnid_t) + sizeof(mdbx_pid_t) +
+                                     sizeof(mdbx_tid_t) + sizeof(pgno_t)) %
+                                        MDBX_CACHELINE_SIZE];
 } MDBX_reader;

 /* Information about a single database in the environment. */
@@ -1149,7 +1158,7 @@ typedef struct MDBX_node {
 #define LEAF2KEY(p, i, ks) ((char *)(p) + PAGEHDRSZ + ((i) * (ks)))

 /* Set the node's key into keyptr, if requested. */
-#define MDBX_GET_KEY(node, keyptr)                                             \
+#define MDBX_GET_MAYNULL_KEYPTR(node, keyptr)                                  \
  do {                                                                         \
    if ((keyptr) != NULL) {                                                    \
      (keyptr)->iov_len = NODEKSZ(node);                                       \
@@ -1158,7 +1167,7 @@ typedef struct MDBX_node {
  } while (0)

 /* Set the node's key into key. */
-#define MDBX_GET_KEY2(node, key)                                               \
+#define MDBX_GET_KEYVALUE(node, key)                                           \
  do {                                                                         \
    key.iov_len = NODEKSZ(node);                                               \
    key.iov_base = NODEKEY(node);                                              \
@@ -1206,7 +1215,7 @@ static __inline pgno_t pgno_sub(pgno_t base, pgno_t subtrahend) {
 }

 static __inline void mdbx_jitter4testing(bool tiny) {
-#ifndef NDEBUG
+#if MDBX_DEBUG
  if (MDBX_DBG_JITTER & mdbx_runtime_flags)
    mdbx_osal_jitter(tiny);
 #else
--- a/src/defs.h
+++ b/src/defs.h
@@ -303,7 +303,7 @@
 #endif /* __flatten */

 #ifndef likely
-#   if defined(__GNUC__) || defined(__clang__)
+#   if (defined(__GNUC__) || defined(__clang__)) && !defined(__COVERITY__)
 #       define likely(cond) __builtin_expect(!!(cond), 1)
 #   else
 #       define likely(x) (x)
@@ -311,13 +311,24 @@
 #endif /* likely */

 #ifndef unlikely
-#   if defined(__GNUC__) || defined(__clang__)
+#   if (defined(__GNUC__) || defined(__clang__)) && !defined(__COVERITY__)
 #       define unlikely(cond) __builtin_expect(!!(cond), 0)
 #   else
 #       define unlikely(x) (x)
 #   endif
 #endif /* unlikely */

+/* Workaround for Coverity Scan */
+#if defined(__COVERITY__) && __GNUC_PREREQ(7, 0) && !defined(__cplusplus)
+typedef float _Float32;
+typedef double _Float32x;
+typedef double _Float64;
+typedef long double _Float64x;
+typedef float _Float128 __attribute__((__mode__(__TF__)));
+typedef __complex__ float __cfloat128 __attribute__ ((__mode__ (__TC__)));
+typedef _Complex float __cfloat128 __attribute__ ((__mode__ (__TC__)));
+#endif /* Workaround for Coverity Scan */
+
 /* Wrapper around __func__, which is a C99 feature */
 #if defined(__STDC_VERSION__) && __STDC_VERSION__ >= 199901L
 #   define mdbx_func_ __func__
--- a/src/lck-posix.c
+++ b/src/lck-posix.c
@@ -44,7 +44,8 @@ static __cold __attribute__((destructor)) void mdbx_global_destructor(void) {
 /* lck */

 #ifndef OFF_T_MAX
-#define OFF_T_MAX (sizeof(off_t) > 4 ? INT64_MAX : INT32_MAX)
+#define OFF_T_MAX                                                              \
+  ((sizeof(off_t) > 4 ? INT64_MAX : INT32_MAX) & ~(size_t)0xffff)
 #endif
 #define LCK_WHOLE OFF_T_MAX

--- a/src/mdbx.c
+++ b/src/mdbx.c
--- a/src/osal.c
+++ b/src/osal.c
@@ -1,4 +1,4 @@
-/* https://en.wikipedia.org/wiki/Operating_system_abstraction_layer */
+/* https://en.wikipedia.org/wiki/Operating_system_abstraction_layer */

 /*
 * Copyright 2015-2019 Leonid Yuriev <leo@yuriev.ru>
@@ -153,8 +153,10 @@ typedef struct _FILE_PROVIDER_EXTERNAL_INFO_V1 {

 /*----------------------------------------------------------------------------*/

-#ifndef _MSC_VER
-/* Prototype should match libc runtime. ISO POSIX (2003) & LSB 3.1 */
+#if !defined(_MSC_VER) &&                                                      \
+    /* workaround for avoid musl libc wrong prototype */ (                     \
+        defined(__GLIBC__) || defined(__GNU_LIBRARY__))
+/* Prototype should match libc runtime. ISO POSIX (2003) & LSB 1.x-3.x */
 __nothrow __noreturn void __assert_fail(const char *assertion, const char *file,
                                        unsigned line, const char *function);
 #endif /* _MSC_VER */
@@ -542,7 +544,6 @@ int mdbx_pread(mdbx_filehandle_t fd, void *buf, size_t bytes, uint64_t offset) {
  if (bytes > MAX_WRITE)
    return MDBX_EINVAL;
 #if defined(_WIN32) || defined(_WIN64)
-
  OVERLAPPED ov;
  ov.hEvent = 0;
  ov.Offset = (DWORD)offset;
@@ -567,21 +568,21 @@ int mdbx_pread(mdbx_filehandle_t fd, void *buf, size_t bytes, uint64_t offset) {

 int mdbx_pwrite(mdbx_filehandle_t fd, const void *buf, size_t bytes,
                uint64_t offset) {
-#if defined(_WIN32) || defined(_WIN64)
-  if (bytes > MAX_WRITE)
-    return ERROR_INVALID_PARAMETER;
-
-  OVERLAPPED ov;
-  ov.hEvent = 0;
-  ov.Offset = (DWORD)offset;
-  ov.OffsetHigh = HIGH_DWORD(offset);
-
-  DWORD written;
-  if (likely(WriteFile(fd, buf, (DWORD)bytes, &written, &ov)))
-    return (bytes == written) ? MDBX_SUCCESS : MDBX_EIO /* ERROR_WRITE_FAULT */;
-  return GetLastError();
-#else
  while (true) {
+#if defined(_WIN32) || defined(_WIN64)
+    OVERLAPPED ov;
+    ov.hEvent = 0;
+    ov.Offset = (DWORD)offset;
+    ov.OffsetHigh = HIGH_DWORD(offset);
+
+    DWORD written;
+    if (unlikely(!WriteFile(fd, buf,
+                            (bytes <= MAX_WRITE) ? (DWORD)bytes : MAX_WRITE,
+                            &written, &ov)))
+      return GetLastError();
+    if (likely(bytes == written))
+      return MDBX_SUCCESS;
+#else
    STATIC_ASSERT_MSG(sizeof(off_t) >= sizeof(size_t),
                      "libmdbx requires 64-bit file I/O on 64-bit systems");
    const intptr_t written =
@@ -592,15 +593,13 @@ int mdbx_pwrite(mdbx_filehandle_t fd, const void *buf, size_t bytes,
      const int rc = errno;
      if (rc != EINTR)
        return rc;
-    } else if (written > 0) {
-      bytes -= written;
-      offset += written;
-      buf = (char *)buf + written;
-    } else {
-      return -1;
+      continue;
    }
-  }
 #endif
+    bytes -= written;
+    offset += written;
+    buf = (char *)buf + written;
+  }
 }

 int mdbx_pwritev(mdbx_filehandle_t fd, struct iovec *iov, int iovcnt,
@@ -631,84 +630,31 @@ int mdbx_pwritev(mdbx_filehandle_t fd, struct iovec *iov, int iovcnt,
 #endif
 }

-int mdbx_write(mdbx_filehandle_t fd, const void *buf, size_t bytes) {
-#ifdef SIGPIPE
-  sigset_t set, old;
-  sigemptyset(&set);
-  sigaddset(&set, SIGPIPE);
-  int rc = pthread_sigmask(SIG_BLOCK, &set, &old);
-  if (rc != 0)
-    return rc;
-#endif
-
-  const char *ptr = buf;
-  for (;;) {
-    size_t chunk = (MAX_WRITE < bytes) ? MAX_WRITE : bytes;
-#if defined(_WIN32) || defined(_WIN64)
-    DWORD written;
-    if (unlikely(!WriteFile(fd, ptr, (DWORD)chunk, &written, NULL)))
-      return GetLastError();
-#else
-    intptr_t written = write(fd, ptr, chunk);
-    if (written < 0) {
-      int rc = errno;
-#ifdef SIGPIPE
-      if (rc == EPIPE) {
-        /* Collect the pending SIGPIPE, otherwise at least OS X
-         * gives it to the process on thread-exit (ITS#8504). */
-        int tmp;
-        sigwait(&set, &tmp);
-        written = 0;
-        continue;
-      }
-      pthread_sigmask(SIG_SETMASK, &old, NULL);
-#endif
-      return rc;
-    }
-#endif
-    if (likely(bytes == (size_t)written)) {
-#ifdef SIGPIPE
-      pthread_sigmask(SIG_SETMASK, &old, NULL);
-#endif
-      return MDBX_SUCCESS;
-    }
-    ptr += written;
-    bytes -= written;
-  }
-}
-
 int mdbx_filesync(mdbx_filehandle_t fd, bool filesize_changed) {
 #if defined(_WIN32) || defined(_WIN64)
  (void)filesize_changed;
  return FlushFileBuffers(fd) ? MDBX_SUCCESS : GetLastError();
-#elif __GLIBC_PREREQ(2, 16) || _BSD_SOURCE || _XOPEN_SOURCE ||                 \
-    (__GLIBC_PREREQ(2, 8) && _POSIX_C_SOURCE >= 200112L)
-  for (;;) {
-/* LY: It is no reason to use fdatasync() here, even in case
- * no such bug in a kernel. Because "no-bug" mean that a kernel
- * internally do nearly the same, e.g. fdatasync() == fsync()
- * when no-kernel-bug and file size was changed.
- *
- * So, this code is always safe and without appreciable
- * performance degradation.
- *
- * For more info about of a corresponding fdatasync() bug
- * see http://www.spinics.net/lists/linux-ext4/msg33714.html */
-#if _POSIX_C_SOURCE >= 199309L || _XOPEN_SOURCE >= 500 ||                      \
-    defined(_POSIX_SYNCHRONIZED_IO)
-    if (!filesize_changed && fdatasync(fd) == 0)
-      return MDBX_SUCCESS;
+#else
+  int rc;
+  do {
+#if defined(_POSIX_SYNCHRONIZED_IO) && _POSIX_SYNCHRONIZED_IO > 0
+    /* LY: This code is always safe and without appreciable performance
+     * degradation, even on a kernel with fdatasync's bug.
+     *
+     * For more info about of a corresponding fdatasync() bug
+     * see http://www.spinics.net/lists/linux-ext4/msg33714.html */
+    if (!filesize_changed) {
+      if (fdatasync(fd) == 0)
+        return MDBX_SUCCESS;
+    } else
 #else
    (void)filesize_changed;
 #endif
-    if (fsync(fd) == 0)
+        if (fsync(fd) == 0)
      return MDBX_SUCCESS;
-    int rc = errno;
-    if (rc != EINTR)
-      return rc;
-  }
-#else
-#error FIXME
+    rc = errno;
+  } while (rc == EINTR);
+  return rc;
 #endif
 }

--- a/src/osal.h
+++ b/src/osal.h
@@ -1,4 +1,4 @@
-/* https://en.wikipedia.org/wiki/Operating_system_abstraction_layer */
+/* https://en.wikipedia.org/wiki/Operating_system_abstraction_layer */

 /*
 * Copyright 2015-2019 Leonid Yuriev <leo@yuriev.ru>
@@ -494,7 +494,6 @@ int mdbx_pwritev(mdbx_filehandle_t fd, struct iovec *iov, int iovcnt,
 int mdbx_pread(mdbx_filehandle_t fd, void *buf, size_t count, uint64_t offset);
 int mdbx_pwrite(mdbx_filehandle_t fd, const void *buf, size_t count,
                uint64_t offset);
-int mdbx_write(mdbx_filehandle_t fd, const void *buf, size_t count);

 int mdbx_thread_create(mdbx_thread_t *thread,
                       THREAD_RESULT(THREAD_CALL *start_routine)(void *),
--- a/src/tools/mdbx_chk.c
+++ b/src/tools/mdbx_chk.c
@@ -92,7 +92,8 @@ MDBX_envinfo envinfo;
 MDBX_stat envstat;
 size_t maxkeysize, userdb_count, skipped_subdb;
 uint64_t reclaimable_pages, gc_pages, lastpgno, unused_pages;
-unsigned verbose, quiet;
+unsigned verbose;
+char ignore_wrong_order, quiet;
 const char *only_subdb;

 struct problem {
@@ -153,6 +154,13 @@ static void pagemap_cleanup(void) {
 static walk_dbi_t *pagemap_lookup_dbi(const char *dbi_name, bool silent) {
  static walk_dbi_t *last;

+  if (dbi_name == MDBX_PGWALK_MAIN)
+    return &dbi_main;
+  if (dbi_name == MDBX_PGWALK_GC)
+    return &dbi_free;
+  if (dbi_name == MDBX_PGWALK_META)
+    return &dbi_meta;
+
  if (last && strcmp(last->name, dbi_name) == 0)
    return last;

@@ -165,7 +173,7 @@ static walk_dbi_t *pagemap_lookup_dbi(const char *dbi_name, bool silent) {
  }

  dbi->name = mdbx_strdup(dbi_name);
-  if (verbose > 1 && !silent) {
+  if (verbose > 0 && !silent) {
    print(" - found '%s' area\n", dbi_name);
    fflush(NULL);
  }
@@ -244,36 +252,34 @@ static size_t problems_pop(struct problem *list) {
  return count;
 }

-static int pgvisitor(uint64_t pgno, unsigned pgnumber, void *ctx, int deep,
-                     const char *dbi_name, size_t page_size,
-                     MDBX_page_type_t pagetype, size_t nentries,
-                     size_t payload_bytes, size_t header_bytes,
-                     size_t unused_bytes) {
+static int pgvisitor(const uint64_t pgno, const unsigned pgnumber,
+                     void *const ctx, const int deep,
+                     const char *const dbi_name_or_tag, const size_t page_size,
+                     const MDBX_page_type_t pagetype, const size_t nentries,
+                     const size_t payload_bytes, const size_t header_bytes,
+                     const size_t unused_bytes) {
  (void)ctx;
+  if (deep > 42) {
+    problem_add("deep", deep, "too large", nullptr);
+    return MDBX_CORRUPTED /* avoid infinite loop/recursion */;
+  }
+
  if (pagetype == MDBX_page_void)
    return MDBX_SUCCESS;

-  walk_dbi_t fake, *dbi = &fake;
-  if (deep > 0) {
-    dbi = pagemap_lookup_dbi(dbi_name, false);
-    if (!dbi)
-      return MDBX_ENOMEM;
-  } else if (deep == 0 && strcmp(dbi_name, dbi_main.name) == 0)
-    dbi = &dbi_main;
-  else if (deep == -1 && strcmp(dbi_name, dbi_free.name) == 0)
-    dbi = &dbi_free;
-  else if (deep == -2 && strcmp(dbi_name, dbi_meta.name) == 0)
-    dbi = &dbi_meta;
-  else
-    problem_add("deep", deep, "unknown area", "%s", dbi_name);
+  walk_dbi_t *dbi = pagemap_lookup_dbi(dbi_name_or_tag, false);
+  if (!dbi)
+    return MDBX_ENOMEM;

  const size_t page_bytes = payload_bytes + header_bytes + unused_bytes;
  walk.pgcount += pgnumber;

  const char *pagetype_caption;
+  bool branch = false;
  switch (pagetype) {
  default:
-    problem_add("page", pgno, "unknown page-type", "%u", (unsigned)pagetype);
+    problem_add("page", pgno, "unknown page-type", "type %u, deep %i",
+                (unsigned)pagetype, deep);
    pagetype_caption = "unknown";
    dbi->pages.other += pgnumber;
    break;
@@ -289,6 +295,7 @@ static int pgvisitor(uint64_t pgno, unsigned pgnumber, void *ctx, int deep,
  case MDBX_page_branch:
    pagetype_caption = "branch";
    dbi->pages.branch += pgnumber;
+    branch = true;
    break;
  case MDBX_page_leaf:
    pagetype_caption = "leaf";
@@ -309,15 +316,39 @@ static int pgvisitor(uint64_t pgno, unsigned pgnumber, void *ctx, int deep,
  }

  if (pgnumber) {
-    if (verbose > 3 && (!only_subdb || strcmp(only_subdb, dbi_name) == 0)) {
+    if (verbose > 3 && (!only_subdb || strcmp(only_subdb, dbi->name) == 0)) {
      if (pgnumber == 1)
        print("     %s-page %" PRIu64, pagetype_caption, pgno);
      else
        print("     %s-span %" PRIu64 "[%u]", pagetype_caption, pgno, pgnumber);
      print(" of %s: header %" PRIiPTR ", payload %" PRIiPTR
-            ", unused %" PRIiPTR "\n",
-            dbi_name, header_bytes, payload_bytes, unused_bytes);
+            ", unused %" PRIiPTR ", deep %i\n",
+            dbi->name, header_bytes, payload_bytes, unused_bytes, deep);
    }
+
+    bool already_used = false;
+    for (unsigned n = 0; n < pgnumber; ++n) {
+      uint64_t spanpgno = pgno + n;
+      if (spanpgno >= lastpgno)
+        problem_add("page", spanpgno, "wrong page-no",
+                    "%s-page: %" PRIu64 " > %" PRIu64 ", deep %i",
+                    pagetype_caption, spanpgno, lastpgno, deep);
+      else if (walk.pagemap[spanpgno]) {
+        walk_dbi_t *coll_dbi = &walk.dbi[walk.pagemap[spanpgno] - 1];
+        problem_add("page", spanpgno,
+                    (branch && coll_dbi == dbi) ? "loop" : "already used",
+                    "%s-page: by %s, deep %i", pagetype_caption, coll_dbi->name,
+                    deep);
+        already_used = true;
+      } else {
+        walk.pagemap[spanpgno] = (short)(dbi - walk.dbi + 1);
+        dbi->pages.total += 1;
+      }
+    }
+
+    if (already_used)
+      return branch ? MDBX_RESULT_TRUE /* avoid infinite loop/recursion */
+                    : MDBX_SUCCESS;
  }

  if (unused_bytes > page_size)
@@ -342,8 +373,9 @@ static int pgvisitor(uint64_t pgno, unsigned pgnumber, void *ctx, int deep,
      } */
    } else {
      problem_add("page", pgno, "empty",
-                  "%s-page: payload %" PRIuPTR " bytes, %" PRIuPTR " entries",
-                  pagetype_caption, payload_bytes, nentries);
+                  "%s-page: payload %" PRIuPTR " bytes, %" PRIuPTR
+                  " entries, deep %i",
+                  pagetype_caption, payload_bytes, nentries, deep);
      dbi->pages.empty += 1;
    }
  }
@@ -352,9 +384,9 @@ static int pgvisitor(uint64_t pgno, unsigned pgnumber, void *ctx, int deep,
    if (page_bytes != page_size) {
      problem_add("page", pgno, "misused",
                  "%s-page: %" PRIuPTR " != %" PRIuPTR " (%" PRIuPTR
-                  "h + %" PRIuPTR "p + %" PRIuPTR "u)",
+                  "h + %" PRIuPTR "p + %" PRIuPTR "u), deep %i",
                  pagetype_caption, page_size, page_bytes, header_bytes,
-                  payload_bytes, unused_bytes);
+                  payload_bytes, unused_bytes, deep);
      if (page_size > page_bytes)
        dbi->lost_bytes += page_size - page_bytes;
    } else {
@@ -363,23 +395,6 @@ static int pgvisitor(uint64_t pgno, unsigned pgnumber, void *ctx, int deep,
    }
  }

-  if (pgnumber) {
-    do {
-      if (pgno >= lastpgno)
-        problem_add("page", pgno, "wrong page-no",
-                    "%s-page: %" PRIu64 " > %" PRIu64, pagetype_caption, pgno,
-                    lastpgno);
-      else if (walk.pagemap[pgno])
-        problem_add("page", pgno, "already used", "%s-page: by %s",
-                    pagetype_caption, walk.dbi[walk.pagemap[pgno] - 1].name);
-      else {
-        walk.pagemap[pgno] = (short)(dbi - walk.dbi + 1);
-        dbi->pages.total += 1;
-      }
-      ++pgno;
-    } while (--pgnumber);
-  }
-
  return user_break ? MDBX_EINTR : MDBX_SUCCESS;
 }

@@ -615,6 +630,8 @@ static int process_db(MDBX_dbi dbi_handle, char *dbi_name, visitor *handler,

  saved_list = problems_push();
  prev_key.iov_base = NULL;
+  prev_key.iov_len = 0;
+  prev_data.iov_base = NULL;
  prev_data.iov_len = 0;
  rc = mdbx_cursor_get(mc, &key, &data, MDBX_FIRST);
  while (rc == MDBX_SUCCESS) {
@@ -625,40 +642,55 @@ static int process_db(MDBX_dbi dbi_handle, char *dbi_name, visitor *handler,
      goto bailout;
    }

+    bool bad_key = false;
    if (key.iov_len > maxkeysize) {
      problem_add("entry", record_count, "key length exceeds max-key-size",
                  "%" PRIuPTR " > %" PRIuPTR, key.iov_len, maxkeysize);
+      bad_key = true;
    } else if ((flags & MDBX_INTEGERKEY) && key.iov_len != sizeof(uint64_t) &&
               key.iov_len != sizeof(uint32_t)) {
      problem_add("entry", record_count, "wrong key length",
                  "%" PRIuPTR " != 4or8", key.iov_len);
+      bad_key = true;
    }

+    bool bad_data = false;
    if ((flags & MDBX_INTEGERDUP) && data.iov_len != sizeof(uint64_t) &&
        data.iov_len != sizeof(uint32_t)) {
      problem_add("entry", record_count, "wrong data length",
                  "%" PRIuPTR " != 4or8", data.iov_len);
+      bad_data = true;
    }

-    if (prev_key.iov_base) {
+    if (prev_key.iov_base && !bad_data) {
      if ((flags & MDBX_DUPFIXED) && prev_data.iov_len != data.iov_len) {
        problem_add("entry", record_count, "different data length",
                    "%" PRIuPTR " != %" PRIuPTR, prev_data.iov_len,
                    data.iov_len);
+        bad_data = true;
      }

-      int cmp = mdbx_cmp(txn, dbi_handle, &prev_key, &key);
-      if (cmp > 0) {
-        problem_add("entry", record_count, "broken ordering of entries", NULL);
-      } else if (cmp == 0) {
-        ++dups;
-        if (!(flags & MDBX_DUPSORT))
-          problem_add("entry", record_count, "duplicated entries", NULL);
-        else if (flags & MDBX_INTEGERDUP) {
-          cmp = mdbx_dcmp(txn, dbi_handle, &prev_data, &data);
-          if (cmp > 0)
-            problem_add("entry", record_count,
-                        "broken ordering of multi-values", NULL);
+      if (!bad_key) {
+        int cmp = mdbx_cmp(txn, dbi_handle, &prev_key, &key);
+        if (cmp == 0) {
+          ++dups;
+          if ((flags & MDBX_DUPSORT) == 0) {
+            problem_add("entry", record_count, "duplicated entries", NULL);
+            if (data.iov_len == prev_data.iov_len &&
+                memcmp(data.iov_base, prev_data.iov_base, data.iov_len) == 0) {
+              problem_add("entry", record_count, "complete duplicate", NULL);
+            }
+          } else if (!bad_data) {
+            cmp = mdbx_dcmp(txn, dbi_handle, &prev_data, &data);
+            if (cmp == 0) {
+              problem_add("entry", record_count, "complete duplicate", NULL);
+            } else if (cmp > 0 && !ignore_wrong_order) {
+              problem_add("entry", record_count, "wrong order of multi-values",
+                          NULL);
+            }
+          }
+        } else if (cmp > 0 && !ignore_wrong_order) {
+          problem_add("entry", record_count, "wrong order of entries", NULL);
        }
      }
    } else if (verbose) {
@@ -670,7 +702,7 @@ static int process_db(MDBX_dbi dbi_handle, char *dbi_name, visitor *handler,

    if (handler) {
      rc = handler(record_count, &key, &data);
-      if (rc)
+      if (MDBX_IS_ERROR(rc))
        goto bailout;
    }

@@ -678,8 +710,10 @@ static int process_db(MDBX_dbi dbi_handle, char *dbi_name, visitor *handler,
    key_bytes += key.iov_len;
    data_bytes += data.iov_len;

-    prev_key = key;
-    prev_data = data;
+    if (!bad_key)
+      prev_key = key;
+    if (!bad_data)
+      prev_data = data;
    rc = mdbx_cursor_get(mc, &key, &data, MDBX_NEXT);
  }
  if (rc != MDBX_NOTFOUND)
@@ -701,7 +735,7 @@ bailout:
  }

  mdbx_cursor_close(mc);
-  return rc || problems_count;
+  return (rc || problems_count) ? MDBX_RESULT_TRUE : MDBX_SUCCESS;
 }

 static void usage(char *prog) {
@@ -714,7 +748,8 @@ static void usage(char *prog) {
          "  -w\t\tlock DB for writing while checking\n"
          "  -d\t\tdisable page-by-page traversal of b-tree\n"
          "  -s subdb\tprocess a specific subdatabase only\n"
-          "  -c\t\tforce cooperative mode (don't try exclusive)\n",
+          "  -c\t\tforce cooperative mode (don't try exclusive)\n"
+          "  -i\t\tignore wrong order errors (for custom comparators case)\n",
          prog);
  exit(EXIT_INTERRUPTED);
 }
@@ -898,7 +933,7 @@ int main(int argc, char *argv[]) {
    usage(prog);
  }

-  for (int i; (i = getopt(argc, argv, "Vvqnwcds:")) != EOF;) {
+  for (int i; (i = getopt(argc, argv, "Vvqnwcdsi:")) != EOF;) {
    switch (i) {
    case 'V':
      printf("%s (%s, build %s)\n", mdbx_version.git.describe,
@@ -928,6 +963,9 @@ int main(int argc, char *argv[]) {
        usage(prog);
      only_subdb = optarg;
      break;
+    case 'i':
+      ignore_wrong_order = 1;
+      break;
    default:
      usage(prog);
    }
@@ -1142,7 +1180,9 @@ int main(int argc, char *argv[]) {
          uint64_t all_leaf = dbi->pages.leaf + dbi->pages.leaf_dupfixed;
          if (all_leaf) {
            print(", leaf %" PRIu64, all_leaf);
-            if (verbose > 2)
+            if (verbose > 2 &&
+                (dbi->pages.leaf_dupfixed | dbi->pages.subleaf_dupsort |
+                 dbi->pages.subleaf_dupsort))
              print(" (usual %" PRIu64 ", sub-dupsort %" PRIu64
                    ", dupfixed %" PRIu64 ", sub-dupfixed %" PRIu64 ")",
                    dbi->pages.leaf, dbi->pages.subleaf_dupsort,
@@ -1163,20 +1203,22 @@ int main(int argc, char *argv[]) {
                  total_page_bytes);
      if (verbose > 2) {
        for (walk_dbi_t *dbi = walk.dbi; dbi < walk.dbi + MAX_DBI && dbi->name;
-             ++dbi) {
-          uint64_t dbi_bytes = dbi->pages.total * envstat.ms_psize;
-          print("     %s: subtotal %" PRIu64 " bytes (%.1f%%),"
-                " payload %" PRIu64 " (%.1f%%), unused %" PRIu64 " (%.1f%%)",
-                dbi->name, dbi_bytes, dbi_bytes * 100.0 / total_page_bytes,
-                dbi->payload_bytes, dbi->payload_bytes * 100.0 / dbi_bytes,
-                dbi_bytes - dbi->payload_bytes,
-                (dbi_bytes - dbi->payload_bytes) * 100.0 / dbi_bytes);
-          if (dbi->pages.empty)
-            print(", %" PRIu64 " empty pages", dbi->pages.empty);
-          if (dbi->lost_bytes)
-            print(", %" PRIu64 " bytes lost", dbi->lost_bytes);
-          print("\n");
-        }
+             ++dbi)
+          if (dbi->pages.total) {
+            uint64_t dbi_bytes = dbi->pages.total * envstat.ms_psize;
+            print("     %s: subtotal %" PRIu64 " bytes (%.1f%%),"
+                  " payload %" PRIu64 " (%.1f%%), unused %" PRIu64 " (%.1f%%)",
+                  dbi->name, dbi_bytes, dbi_bytes * 100.0 / total_page_bytes,
+                  dbi->payload_bytes, dbi->payload_bytes * 100.0 / dbi_bytes,
+                  dbi_bytes - dbi->payload_bytes,
+                  (dbi_bytes - dbi->payload_bytes) * 100.0 / dbi_bytes);
+            if (dbi->pages.empty)
+              print(", %" PRIu64 " empty pages", dbi->pages.empty);
+            if (dbi->lost_bytes)
+              print(", %" PRIu64 " bytes lost", dbi->lost_bytes);
+            print("\n");
+          } else
+            print("     %s: empty\n", dbi->name);
      }
      print(" - summary: average fill %.1f%%",
            walk.total_payload_bytes * 100.0 / total_page_bytes);
--- a/src/tools/mdbx_load.1
+++ b/src/tools/mdbx_load.1
@@ -39,6 +39,13 @@ option below.
 .BR \-V
 Write the library version number to the standard output, and exit.
 .TP
+.BR \-a
+Append all records in the order they appear in the input. The input is assumed to already be
+in correctly sorted order and no sorting or checking for redundant values will be performed.
+This option must be used to reload data that was produced by running
+.B mdbx_dump
+on a database that uses custom compare functions.
+.TP
 .BR \-f \ file
 Read from the specified file instead of from the standard input.
 .TP
--- a/src/tools/mdbx_load.c
+++ b/src/tools/mdbx_load.c
@@ -1,4 +1,4 @@
-/* mdbx_load.c - memory-mapped database load tool */
+/* mdbx_load.c - memory-mapped database load tool */

 /*
 * Copyright 2015-2019 Leonid Yuriev <leo@yuriev.ru>
@@ -57,6 +57,7 @@ static int Eof;

 static MDBX_envinfo envinfo;
 static MDBX_val kbuf, dbuf;
+static MDBX_val k0buf;

 #define STRLENOF(s) (sizeof(s) - 1)

@@ -304,11 +305,18 @@ static int readline(MDBX_val *out, MDBX_val *buf) {
 }

 static void usage(void) {
-  fprintf(stderr, "usage: %s [-V] [-f input] [-n] [-s name] [-N] [-T] dbpath\n",
+  fprintf(stderr,
+          "usage: %s [-V] [-a] [-f input] [-n] [-s name] [-N] [-T] dbpath\n",
          prog);
  exit(EXIT_FAILURE);
 }

+static int anyway_greater(const MDBX_val *a, const MDBX_val *b) {
+  (void)a;
+  (void)b;
+  return 1;
+}
+
 int main(int argc, char *argv[]) {
  int i, rc;
  MDBX_env *env = NULL;
@@ -316,28 +324,32 @@ int main(int argc, char *argv[]) {
  MDBX_cursor *mc = NULL;
  MDBX_dbi dbi;
  char *envname = NULL;
-  int envflags = 0, putflags = 0;
+  int envflags = MDBX_UTTERLY_NOSYNC, putflags = 0;
+  int append = 0;
+  MDBX_val prevk;

  prog = argv[0];
-
-  if (argc < 2) {
+  if (argc < 2)
    usage();
-  }

-  /* -f: load file instead of stdin
+  /* -a: append records in input order
+   * -f: load file instead of stdin
   * -n: use NOSUBDIR flag on env_open
   * -s: load into named subDB
   * -N: use NOOVERWRITE on puts
   * -T: read plaintext
   * -V: print version and exit
   */
-  while ((i = getopt(argc, argv, "f:ns:NTV")) != EOF) {
+  while ((i = getopt(argc, argv, "af:ns:NTV")) != EOF) {
    switch (i) {
    case 'V':
      printf("%s (%s, build %s)\n", mdbx_version.git.describe,
             mdbx_version.git.datetime, mdbx_build.datetime);
      exit(EXIT_SUCCESS);
      break;
+    case 'a':
+      append = 1;
+      break;
    case 'f':
      if (freopen(optarg, "r", stdin) == NULL) {
        fprintf(stderr, "%s: %s: reopen: %s\n", prog, optarg,
@@ -381,6 +393,7 @@ int main(int argc, char *argv[]) {
  dbuf.iov_len = 4096;
  dbuf.iov_base = mdbx_malloc(dbuf.iov_len);

+  /* read first header for mapsize= */
  if (!(mode & NOHDR))
    readhdr();

@@ -418,8 +431,17 @@ int main(int argc, char *argv[]) {
    goto env_close;
  }

-  kbuf.iov_len = mdbx_env_get_maxkeysize(env) * 2 + 2;
-  kbuf.iov_base = mdbx_malloc(kbuf.iov_len);
+  kbuf.iov_len = mdbx_env_get_maxkeysize(env);
+  if (kbuf.iov_len >= SIZE_MAX / 4) {
+    fprintf(stderr, "mdbx_env_get_maxkeysize failed, returns %zu\n",
+            kbuf.iov_len);
+    goto env_close;
+  }
+  kbuf.iov_len = (kbuf.iov_len + 1) * 2;
+  kbuf.iov_base = malloc(kbuf.iov_len * 2);
+  k0buf.iov_len = kbuf.iov_len;
+  k0buf.iov_base = (char *)kbuf.iov_base + kbuf.iov_len;
+  prevk.iov_base = k0buf.iov_base;

  while (!Eof) {
    if (user_break) {
@@ -427,9 +449,6 @@ int main(int argc, char *argv[]) {
      break;
    }

-    MDBX_val key, data;
-    int batch = 0;
-
    rc = mdbx_txn_begin(env, NULL, 0, &txn);
    if (rc) {
      fprintf(stderr, "mdbx_txn_begin failed, error %d %s\n", rc,
@@ -437,7 +456,9 @@ int main(int argc, char *argv[]) {
      goto env_close;
    }

-    rc = mdbx_dbi_open(txn, subname, dbi_flags | MDBX_CREATE, &dbi);
+    rc = mdbx_dbi_open_ex(txn, subname, dbi_flags | MDBX_CREATE, &dbi,
+                          append ? anyway_greater : NULL,
+                          append ? anyway_greater : NULL);
    if (rc) {
      fprintf(stderr, "mdbx_open failed, error %d %s\n", rc, mdbx_strerror(rc));
      goto txn_abort;
@@ -450,11 +471,15 @@ int main(int argc, char *argv[]) {
      goto txn_abort;
    }

+    int batch = 0;
+    prevk.iov_len = 0;
    while (1) {
+      MDBX_val key;
      rc = readline(&key, &kbuf);
      if (rc) /* rc == EOF */
        break;

+      MDBX_val data;
      rc = readline(&data, &dbuf);
      if (rc) {
        fprintf(stderr, "%s: line %" PRIiSIZE ": failed to read key value\n",
@@ -462,7 +487,18 @@ int main(int argc, char *argv[]) {
        goto txn_abort;
      }

-      rc = mdbx_cursor_put(mc, &key, &data, putflags);
+      int appflag = 0;
+      if (append) {
+        appflag = MDBX_APPEND;
+        if (dbi_flags & MDBX_DUPSORT) {
+          if (prevk.iov_len == key.iov_len &&
+              memcmp(prevk.iov_base, key.iov_base, key.iov_len) == 0)
+            appflag = MDBX_APPEND | MDBX_APPENDDUP;
+          else
+            memcpy(prevk.iov_base, key.iov_base, prevk.iov_len = key.iov_len);
+        }
+      }
+      rc = mdbx_cursor_put(mc, &key, &data, putflags | appflag);
      if (rc == MDBX_KEYEXIST && putflags)
        continue;
      if (rc) {
@@ -501,6 +537,8 @@ int main(int argc, char *argv[]) {
      goto env_close;
    }
    mdbx_dbi_close(env, dbi);
+
+    /* try read next header */
    if (!(mode & NOHDR))
      readhdr();
  }
--- a/src/version.c
+++ b/src/version.c
@@ -14,11 +14,11 @@

 #include "./bits.h"

-#if MDBX_VERSION_MAJOR != 0 || MDBX_VERSION_MINOR != 2
+#if MDBX_VERSION_MAJOR != 0 || MDBX_VERSION_MINOR != 3
 #error "API version mismatch!"
 #endif

-#define MDBX_VERSION_RELEASE 2
+#define MDBX_VERSION_RELEASE 0
 #define MDBX_VERSION_REVISION 0

 /*LIBMDBX_EXPORTS*/ const mdbx_version_info mdbx_version = {
--- a/test/CMakeLists.txt
+++ b/test/CMakeLists.txt
@@ -27,6 +27,8 @@ add_executable(${TARGET}
  try.cc
  utils.cc
  utils.h
+  append.cc
+  ttl.cc
  )

 target_link_libraries(${TARGET}
--- a/test/append.cc
+++ b/test/append.cc
@@ -0,0 +1,165 @@
+/*
+ * Copyright 2017-2019 Leonid Yuriev <leo@yuriev.ru>
+ * and other libmdbx authors: please see AUTHORS file.
+ * All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted only as authorized by the OpenLDAP
+ * Public License.
+ *
+ * A copy of this license is available in the file LICENSE in the
+ * top-level directory of the distribution or, alternatively, at
+ * <http://www.OpenLDAP.org/license.html>.
+ */
+
+#include "test.h"
+
+bool testcase_append::run() {
+  MDBX_dbi dbi;
+  int err = db_open__begin__table_create_open_clean(dbi);
+  if (unlikely(err != MDBX_SUCCESS)) {
+    log_notice("append: bailout-prepare due '%s'", mdbx_strerror(err));
+    return true;
+  }
+
+  keyvalue_maker.setup(config.params, config.actor_id, 0 /* thread_number */);
+  /* LY: тест наполнения таблиц в append-режиме,
+   * при котором записи добавляются строго в конец (в порядке сортировки) */
+  const unsigned flags = (config.params.table_flags & MDBX_DUPSORT)
+                             ? MDBX_APPEND | MDBX_APPENDDUP
+                             : MDBX_APPEND;
+  keyvalue_maker.make_ordered();
+
+  key = keygen::alloc(config.params.keylen_max);
+  data = keygen::alloc(config.params.datalen_max);
+  keygen::buffer last_key = keygen::alloc(config.params.keylen_max);
+  keygen::buffer last_data = keygen::alloc(config.params.datalen_max);
+  last_key->value.iov_base = last_key->bytes;
+  last_key->value.iov_len = 0;
+  last_data->value.iov_base = last_data->bytes;
+  last_data->value.iov_len = 0;
+
+  simple_checksum inserted_checksum;
+  uint64_t inserted_number = 0;
+  uint64_t serial_count = 0;
+
+  unsigned txn_nops = 0;
+  uint64_t commited_inserted_number = inserted_number;
+  simple_checksum commited_inserted_checksum = inserted_checksum;
+  while (should_continue()) {
+    const keygen::serial_t serial = serial_count;
+    if (!keyvalue_maker.increment(serial_count, 1)) {
+      // дошли до границы пространства ключей
+      break;
+    }
+
+    log_trace("append: append-a %" PRIu64, serial);
+    generate_pair(serial, key, data);
+    int cmp = inserted_number ? mdbx_cmp(txn_guard.get(), dbi, &key->value,
+                                         &last_key->value)
+                              : 1;
+    if (cmp == 0 && (config.params.table_flags & MDBX_DUPSORT))
+      cmp = mdbx_dcmp(txn_guard.get(), dbi, &data->value, &last_data->value);
+
+    err = mdbx_put(txn_guard.get(), dbi, &key->value, &data->value, flags);
+    if (err == MDBX_MAP_FULL && config.params.ignore_dbfull) {
+      log_notice("append: bailout-insert due '%s'", mdbx_strerror(err));
+      txn_end(true);
+      inserted_number = commited_inserted_number;
+      inserted_checksum = commited_inserted_checksum;
+      break;
+    }
+
+    if (cmp > 0) {
+      if (unlikely(err != MDBX_SUCCESS))
+        failure_perror("mdbx_put(appenda-a)", err);
+
+      memcpy(last_key->value.iov_base, key->value.iov_base,
+             last_key->value.iov_len = key->value.iov_len);
+      memcpy(last_data->value.iov_base, data->value.iov_base,
+             last_data->value.iov_len = data->value.iov_len);
+      ++inserted_number;
+      inserted_checksum.push((uint32_t)inserted_number, key->value);
+      inserted_checksum.push(10639, data->value);
+    } else {
+      if (unlikely(err != MDBX_EKEYMISMATCH))
+        failure_perror("mdbx_put(appenda-a) != MDBX_EKEYMISMATCH", err);
+    }
+
+    if (++txn_nops >= config.params.batch_write) {
+      err = breakable_restart();
+      if (unlikely(err != MDBX_SUCCESS)) {
+        log_notice("append: bailout-commit due '%s'", mdbx_strerror(err));
+        inserted_number = commited_inserted_number;
+        inserted_checksum = commited_inserted_checksum;
+        break;
+      }
+      commited_inserted_number = inserted_number;
+      commited_inserted_checksum = inserted_checksum;
+      txn_nops = 0;
+    }
+
+    report(1);
+  }
+
+  if (txn_guard) {
+    err = breakable_commit();
+    if (unlikely(err != MDBX_SUCCESS)) {
+      log_notice("append: bailout-commit due '%s'", mdbx_strerror(err));
+      inserted_number = commited_inserted_number;
+      inserted_checksum = commited_inserted_checksum;
+    }
+  }
+  //----------------------------------------------------------------------------
+  txn_begin(true);
+  cursor_open(dbi);
+
+  MDBX_val check_key, check_data;
+  err =
+      mdbx_cursor_get(cursor_guard.get(), &check_key, &check_data, MDBX_FIRST);
+  if (likely(inserted_number)) {
+    if (unlikely(err != MDBX_SUCCESS))
+      failure_perror("mdbx_cursor_get(MDBX_FIRST)", err);
+  }
+
+  simple_checksum read_checksum;
+  uint64_t read_count = 0;
+  while (err == MDBX_SUCCESS) {
+    ++read_count;
+    read_checksum.push((uint32_t)read_count, check_key);
+    read_checksum.push(10639, check_data);
+
+    err =
+        mdbx_cursor_get(cursor_guard.get(), &check_key, &check_data, MDBX_NEXT);
+  }
+
+  if (unlikely(err != MDBX_NOTFOUND))
+    failure_perror("mdbx_cursor_get(MDBX_NEXT) != EOF", err);
+
+  if (unlikely(read_count != inserted_number))
+    failure("read_count(%" PRIu64 ") != inserted_number(%" PRIu64 ")",
+            read_count, inserted_number);
+
+  if (unlikely(read_checksum.value != inserted_checksum.value))
+    failure("read_checksum(0x%016" PRIu64 ") "
+            "!= inserted_checksum(0x%016" PRIu64 ")",
+            read_checksum.value, inserted_checksum.value);
+
+  cursor_close();
+  txn_end(true);
+  //----------------------------------------------------------------------------
+
+  if (dbi) {
+    if (config.params.drop_table && !mode_readonly()) {
+      txn_begin(false);
+      db_table_drop(dbi);
+      err = breakable_commit();
+      if (unlikely(err != MDBX_SUCCESS)) {
+        log_notice("append: bailout-clean due '%s'", mdbx_strerror(err));
+        return true;
+      }
+    } else
+      db_table_close(dbi);
+  }
+  return true;
+}
--- a/test/cases.cc
+++ b/test/cases.cc
@@ -63,12 +63,13 @@ void testcase_setup(const char *casename, actor_params &params,
    log_notice(">>> testcase_setup(%s)", casename);
    configure_actor(last_space_id, ac_jitter, nullptr, params);
    configure_actor(last_space_id, ac_hill, nullptr, params);
+    configure_actor(last_space_id, ac_ttl, nullptr, params);
    configure_actor(last_space_id, ac_jitter, nullptr, params);
    configure_actor(last_space_id, ac_hill, nullptr, params);
-    configure_actor(last_space_id, ac_jitter, nullptr, params);
-    configure_actor(last_space_id, ac_hill, nullptr, params);
+    configure_actor(last_space_id, ac_ttl, nullptr, params);
    configure_actor(last_space_id, ac_try, nullptr, params);
    configure_actor(last_space_id, ac_copy, nullptr, params);
+    configure_actor(last_space_id, ac_append, nullptr, params);
    log_notice("<<< testcase_setup(%s): done", casename);
  } else {
    failure("unknown testcase `%s`", casename);
--- a/test/config.cc
+++ b/test/config.cc
@@ -1,4 +1,4 @@
-/*
+/*
 * Copyright 2017-2019 Leonid Yuriev <leo@yuriev.ru>
 * and other libmdbx authors: please see AUTHORS file.
 * All rights reserved.
@@ -412,6 +412,8 @@ void dump(const char *title) {
             i->params.max_tables);

    log_info("drop table: %s\n", i->params.drop_table ? "Yes" : "No");
+    log_info("ignore MDBX_MAP_FULL error: %s\n",
+             i->params.ignore_dbfull ? "Yes" : "No");
    indent.pop();
  }

--- a/test/config.h
+++ b/test/config.h
@@ -27,7 +27,9 @@ enum actor_testcase {
  ac_deadwrite,
  ac_jitter,
  ac_try,
-  ac_copy
+  ac_copy,
+  ac_append,
+  ac_ttl
 };

 enum actor_status {
@@ -36,7 +38,8 @@ enum actor_status {
  as_running,
  as_successful,
  as_killed,
-  as_failed
+  as_failed,
+  as_coredump,
 };

 const char *testcase2str(const actor_testcase);
@@ -246,6 +249,7 @@ struct actor_params_pod {
  keygen_params_pod keygen;

  bool drop_table;
+  bool ignore_dbfull;
 };

 struct actor_config_pod {
--- a/test/gc.sh
+++ b/test/gc.sh
@@ -1,66 +0,0 @@
-#!/bin/bash
-set -euo pipefail
-make check
-TESTDB_PREFIX=${1:-/dev/shm/mdbx-gc-test}.
-
-function rep9 { printf "%*s" $1 '' | tr ' ' '9'; }
-function join { local IFS="$1"; shift; echo "$*"; }
-function bit2option { local -n arr=$1; (( ($2&(1<<$3)) != 0 )) && echo -n '+' || echo -n '-'; echo "${arr[$3]}"; }
-
-options=(writemap coalesce lifo)
-
-function bits2list {
-	local -n arr=$1
-	local i
-	local list=()
-	for ((i=0; i<${#arr[@]}; ++i)) do
-		list[$i]=$(bit2option $1 $2 $i)
-	done
-	join , "${list[@]}"
-}
-
-function probe {
-	echo "=============================================== $(date)"
-	echo "${caption}: $*"
-	rm -f ${TESTDB_PREFIX}* \
-		&& ./mdbx_test --pathname=${TESTDB_PREFIX}db "$@" | lz4 > ${TESTDB_PREFIX}log.lz4 \
-		&& ./mdbx_chk -nvvv ${TESTDB_PREFIX}db | tee ${TESTDB_PREFIX}chk \
-		&& ./mdbx_chk -nvvv ${TESTDB_PREFIX}db-copy | tee ${TESTDB_PREFIX}chk-copy \
-		|| (echo "FAILED"; exit 1)
-}
-
-###############################################################################
-
-count=0
-for nops in {2..7}; do
-	for ((wbatch=nops-1; wbatch > 0; --wbatch)); do
-		loops=$(((3333 >> nops) / nops + 1))
-		for ((rep=0; rep++ < loops; )); do
-			for ((bits=2**${#options[@]}; --bits >= 0; )); do
-				seed=$(date +%N)
-				caption="Probe #$((++count)) int-key,w/o-dups, repeat ${rep} of ${loops}" probe \
-					--pagesize=min --size=6G --table=+key.integer,-data.dups --keylen.min=min --keylen.max=max --datalen.min=min --datalen.max=1111 \
-					--nops=$( rep9 $nops ) --batch.write=$( rep9 $wbatch ) --mode=$(bits2list options $bits) \
-					--keygen.seed=${seed} basic
-				caption="Probe #$((++count)) int-key,with-dups, repeat ${rep} of ${loops}" probe \
-					--pagesize=min --size=6G --table=+key.integer,+data.dups --keylen.min=min --keylen.max=max --datalen.min=min --datalen.max=max \
-					--nops=$( rep9 $nops ) --batch.write=$( rep9 $wbatch ) --mode=$(bits2list options $bits) \
-					--keygen.seed=${seed} basic
-				caption="Probe #$((++count)) int-key,int-data, repeat ${rep} of ${loops}" probe \
-					--pagesize=min --size=6G --table=+key.integer,+data.integer --keylen.min=min --keylen.max=max --datalen.min=min --datalen.max=max \
-					--nops=$( rep9 $nops ) --batch.write=$( rep9 $wbatch ) --mode=$(bits2list options $bits) \
-					--keygen.seed=${seed} basic
-				caption="Probe #$((++count)) w/o-dups, repeat ${rep} of ${loops}" probe \
-					--pagesize=min --size=6G --table=-data.dups --keylen.min=min --keylen.max=max --datalen.min=min --datalen.max=1111 \
-					--nops=$( rep9 $nops ) --batch.write=$( rep9 $wbatch ) --mode=$(bits2list options $bits) \
-					--keygen.seed=${seed} basic
-				caption="Probe #$((++count)) with-dups, repeat ${rep} of ${loops}" probe \
-					--pagesize=min --size=6G --table=+data.dups --keylen.min=min --keylen.max=max --datalen.min=min --datalen.max=max \
-					--nops=$( rep9 $nops ) --batch.write=$( rep9 $wbatch ) --mode=$(bits2list options $bits) \
-					--keygen.seed=${seed} basic
-			done
-		done
-	done
-done
-
-echo "=== ALL DONE ====================== $(date)"
--- a/test/hill.cc
+++ b/test/hill.cc
@@ -15,11 +15,12 @@
 #include "test.h"

 bool testcase_hill::run() {
-  db_open();
-
-  txn_begin(false);
-  MDBX_dbi dbi = db_table_open(true);
-  txn_end(false);
+  MDBX_dbi dbi;
+  int err = db_open__begin__table_create_open_clean(dbi);
+  if (unlikely(err != MDBX_SUCCESS)) {
+    log_notice("hill: bailout-prepare due '%s'", mdbx_strerror(err));
+    return true;
+  }

  /* LY: тест "холмиком":
   *  - сначала наполняем таблицу циклическими CRUD-манипуляциями,
@@ -59,14 +60,15 @@ bool testcase_hill::run() {
          : MDBX_NODUPDATA;

  uint64_t serial_count = 0;
+  uint64_t commited_serial = serial_count;
  unsigned txn_nops = 0;
-  if (!txn_guard)
-    txn_begin(false);

  while (should_continue()) {
    const keygen::serial_t a_serial = serial_count;
-    if (unlikely(!keyvalue_maker.increment(serial_count, 1)))
-      failure("uphill: unexpected key-space overflow");
+    if (unlikely(!keyvalue_maker.increment(serial_count, 1))) {
+      log_notice("uphill: unexpected key-space overflow");
+      break;
+    }

    const keygen::serial_t b_serial = serial_count;
    assert(b_serial > a_serial);
@@ -76,26 +78,52 @@ bool testcase_hill::run() {
    log_trace("uphill: insert-a (age %" PRIu64 ") %" PRIu64, age_shift,
              a_serial);
    generate_pair(a_serial, a_key, a_data_1, age_shift);
-    int rc = mdbx_put(txn_guard.get(), dbi, &a_key->value, &a_data_1->value,
-                      insert_flags);
-    if (unlikely(rc != MDBX_SUCCESS))
-      failure_perror("mdbx_put(insert-a.1)", rc);
+    err = mdbx_put(txn_guard.get(), dbi, &a_key->value, &a_data_1->value,
+                   insert_flags);
+    if (unlikely(err != MDBX_SUCCESS)) {
+      if (err == MDBX_MAP_FULL && config.params.ignore_dbfull) {
+        log_notice("uphill: bailout at insert-a due '%s'", mdbx_strerror(err));
+        txn_restart(true, false);
+        serial_count = commited_serial;
+        break;
+      }
+      failure_perror("mdbx_put(insert-a.1)", err);
+    }

    if (++txn_nops >= config.params.batch_write) {
-      txn_restart(false, false);
+      err = breakable_restart();
+      if (unlikely(err != MDBX_SUCCESS)) {
+        log_notice("uphill: bailout at commit due '%s'", mdbx_strerror(err));
+        serial_count = commited_serial;
+        break;
+      }
+      commited_serial = a_serial;
      txn_nops = 0;
    }

    // создаем вторую запись из пары
    log_trace("uphill: insert-b %" PRIu64, b_serial);
    generate_pair(b_serial, b_key, b_data, 0);
-    rc = mdbx_put(txn_guard.get(), dbi, &b_key->value, &b_data->value,
-                  insert_flags);
-    if (unlikely(rc != MDBX_SUCCESS))
-      failure_perror("mdbx_put(insert-b)", rc);
+    err = mdbx_put(txn_guard.get(), dbi, &b_key->value, &b_data->value,
+                   insert_flags);
+    if (unlikely(err != MDBX_SUCCESS)) {
+      if (err == MDBX_MAP_FULL && config.params.ignore_dbfull) {
+        log_notice("uphill: bailout at insert-b due '%s'", mdbx_strerror(err));
+        txn_restart(true, false);
+        serial_count = commited_serial;
+        break;
+      }
+      failure_perror("mdbx_put(insert-b)", err);
+    }

    if (++txn_nops >= config.params.batch_write) {
-      txn_restart(false, false);
+      err = breakable_restart();
+      if (unlikely(err != MDBX_SUCCESS)) {
+        log_notice("uphill: bailout at commit due '%s'", mdbx_strerror(err));
+        serial_count = commited_serial;
+        break;
+      }
+      commited_serial = a_serial;
      txn_nops = 0;
    }

@@ -104,25 +132,51 @@ bool testcase_hill::run() {
              a_serial);
    generate_pair(a_serial, a_key, a_data_0, 0);
    checkdata("uphill: update-a", dbi, a_key->value, a_data_1->value);
-    rc = mdbx_replace(txn_guard.get(), dbi, &a_key->value, &a_data_0->value,
-                      &a_data_1->value, update_flags);
-    if (unlikely(rc != MDBX_SUCCESS))
-      failure_perror("mdbx_replace(update-a: 1->0)", rc);
+    err = mdbx_replace(txn_guard.get(), dbi, &a_key->value, &a_data_0->value,
+                       &a_data_1->value, update_flags);
+    if (unlikely(err != MDBX_SUCCESS)) {
+      if (err == MDBX_MAP_FULL && config.params.ignore_dbfull) {
+        log_notice("uphill: bailout at update-a due '%s'", mdbx_strerror(err));
+        txn_restart(true, false);
+        serial_count = commited_serial;
+        break;
+      }
+      failure_perror("mdbx_replace(update-a: 1->0)", err);
+    }

    if (++txn_nops >= config.params.batch_write) {
-      txn_restart(false, false);
+      err = breakable_restart();
+      if (unlikely(err != MDBX_SUCCESS)) {
+        log_notice("uphill: bailout at commit due '%s'", mdbx_strerror(err));
+        serial_count = commited_serial;
+        break;
+      }
+      commited_serial = a_serial;
      txn_nops = 0;
    }

    // удаляем вторую запись
    log_trace("uphill: delete-b %" PRIu64, b_serial);
    checkdata("uphill: delete-b", dbi, b_key->value, b_data->value);
-    rc = mdbx_del(txn_guard.get(), dbi, &b_key->value, &b_data->value);
-    if (unlikely(rc != MDBX_SUCCESS))
-      failure_perror("mdbx_del(b)", rc);
+    err = mdbx_del(txn_guard.get(), dbi, &b_key->value, &b_data->value);
+    if (unlikely(err != MDBX_SUCCESS)) {
+      if (err == MDBX_MAP_FULL && config.params.ignore_dbfull) {
+        log_notice("uphill: bailout at delete-b due '%s'", mdbx_strerror(err));
+        txn_restart(true, false);
+        serial_count = commited_serial;
+        break;
+      }
+      failure_perror("mdbx_del(b)", err);
+    }

    if (++txn_nops >= config.params.batch_write) {
-      txn_restart(false, false);
+      err = breakable_restart();
+      if (unlikely(err != MDBX_SUCCESS)) {
+        log_notice("uphill: bailout at commit due '%s'", mdbx_strerror(err));
+        serial_count = commited_serial;
+        break;
+      }
+      commited_serial = a_serial;
      txn_nops = 0;
    }

@@ -134,7 +188,7 @@ bool testcase_hill::run() {
    }
  }

-  while (serial_count > 0) {
+  while (serial_count > 1) {
    if (unlikely(!keyvalue_maker.increment(serial_count, -2)))
      failure("downhill: unexpected key-space underflow");

@@ -150,26 +204,48 @@ bool testcase_hill::run() {
    generate_pair(a_serial, a_key, a_data_0, 0);
    generate_pair(a_serial, a_key, a_data_1, age_shift);
    checkdata("downhill: update-a", dbi, a_key->value, a_data_0->value);
-    int rc = mdbx_replace(txn_guard.get(), dbi, &a_key->value, &a_data_1->value,
-                          &a_data_0->value, update_flags);
-    if (unlikely(rc != MDBX_SUCCESS))
-      failure_perror("mdbx_put(update-a: 0->1)", rc);
+    err = mdbx_replace(txn_guard.get(), dbi, &a_key->value, &a_data_1->value,
+                       &a_data_0->value, update_flags);
+    if (unlikely(err != MDBX_SUCCESS)) {
+      if (err == MDBX_MAP_FULL && config.params.ignore_dbfull) {
+        log_notice("downhill: bailout at update-a due '%s'",
+                   mdbx_strerror(err));
+        txn_end(true);
+        break;
+      }
+      failure_perror("mdbx_put(update-a: 0->1)", err);
+    }

    if (++txn_nops >= config.params.batch_write) {
-      txn_restart(false, false);
+      err = breakable_restart();
+      if (unlikely(err != MDBX_SUCCESS)) {
+        log_notice("downhill: bailout at commit due '%s'", mdbx_strerror(err));
+        break;
+      }
      txn_nops = 0;
    }

    // создаем вторую запись из пары
    log_trace("downhill: insert-b %" PRIu64, b_serial);
    generate_pair(b_serial, b_key, b_data, 0);
-    rc = mdbx_put(txn_guard.get(), dbi, &b_key->value, &b_data->value,
-                  insert_flags);
-    if (unlikely(rc != MDBX_SUCCESS))
-      failure_perror("mdbx_put(insert-b)", rc);
+    err = mdbx_put(txn_guard.get(), dbi, &b_key->value, &b_data->value,
+                   insert_flags);
+    if (unlikely(err != MDBX_SUCCESS)) {
+      if (err == MDBX_MAP_FULL && config.params.ignore_dbfull) {
+        log_notice("downhill: bailout at insert-a due '%s'",
+                   mdbx_strerror(err));
+        txn_end(true);
+        break;
+      }
+      failure_perror("mdbx_put(insert-b)", err);
+    }

    if (++txn_nops >= config.params.batch_write) {
-      txn_restart(false, false);
+      err = breakable_restart();
+      if (unlikely(err != MDBX_SUCCESS)) {
+        log_notice("downhill: bailout at commit due '%s'", mdbx_strerror(err));
+        break;
+      }
      txn_nops = 0;
    }

@@ -177,38 +253,67 @@ bool testcase_hill::run() {
    log_trace("downhill: delete-a (age %" PRIu64 ") %" PRIu64, age_shift,
              a_serial);
    checkdata("downhill: delete-a", dbi, a_key->value, a_data_1->value);
-    rc = mdbx_del(txn_guard.get(), dbi, &a_key->value, &a_data_1->value);
-    if (unlikely(rc != MDBX_SUCCESS))
-      failure_perror("mdbx_del(a)", rc);
+    err = mdbx_del(txn_guard.get(), dbi, &a_key->value, &a_data_1->value);
+    if (unlikely(err != MDBX_SUCCESS)) {
+      if (err == MDBX_MAP_FULL && config.params.ignore_dbfull) {
+        log_notice("downhill: bailout at delete-a due '%s'",
+                   mdbx_strerror(err));
+        txn_end(true);
+        break;
+      }
+      failure_perror("mdbx_del(a)", err);
+    }

    if (++txn_nops >= config.params.batch_write) {
-      txn_restart(false, false);
+      err = breakable_restart();
+      if (unlikely(err != MDBX_SUCCESS)) {
+        log_notice("downhill: bailout at commit due '%s'", mdbx_strerror(err));
+        break;
+      }
      txn_nops = 0;
    }

    // удаляем вторую запись
    log_trace("downhill: delete-b %" PRIu64, b_serial);
    checkdata("downhill: delete-b", dbi, b_key->value, b_data->value);
-    rc = mdbx_del(txn_guard.get(), dbi, &b_key->value, &b_data->value);
-    if (unlikely(rc != MDBX_SUCCESS))
-      failure_perror("mdbx_del(b)", rc);
+    err = mdbx_del(txn_guard.get(), dbi, &b_key->value, &b_data->value);
+    if (unlikely(err != MDBX_SUCCESS)) {
+      if (err == MDBX_MAP_FULL && config.params.ignore_dbfull) {
+        log_notice("downhill: bailout at delete-b due '%s'",
+                   mdbx_strerror(err));
+        txn_end(true);
+        break;
+      }
+      failure_perror("mdbx_del(b)", err);
+    }

    if (++txn_nops >= config.params.batch_write) {
-      txn_restart(false, false);
+      err = breakable_restart();
+      if (unlikely(err != MDBX_SUCCESS)) {
+        log_notice("downhill: bailout at commit due '%s'", mdbx_strerror(err));
+        break;
+      }
      txn_nops = 0;
    }

    report(1);
  }

-  if (txn_guard)
-    txn_end(false);
+  if (txn_guard) {
+    err = breakable_commit();
+    if (unlikely(err != MDBX_SUCCESS))
+      log_notice("downhill: bailout at commit due '%s'", mdbx_strerror(err));
+  }

  if (dbi) {
    if (config.params.drop_table && !mode_readonly()) {
      txn_begin(false);
      db_table_drop(dbi);
-      txn_end(false);
+      err = breakable_commit();
+      if (unlikely(err != MDBX_SUCCESS)) {
+        log_notice("hill: bailout-clean due '%s'", mdbx_strerror(err));
+        return true;
+      }
    } else
      db_table_close(dbi);
  }
--- a/test/keygen.cc
+++ b/test/keygen.cc
@@ -1,4 +1,4 @@
-/*
+/*
 * Copyright 2017-2019 Leonid Yuriev <leo@yuriev.ru>
 * and other libmdbx authors: please see AUTHORS file.
 * All rights reserved.
@@ -167,7 +167,16 @@ void maker::setup(const config::actor_params_pod &actor, unsigned actor_id,
  base = 0;
 }

-bool maker::increment(serial_t &serial, int delta) {
+void maker::make_ordered() {
+  mapping.mesh = 0;
+  mapping.rotate = 0;
+}
+
+bool maker::is_unordered() const {
+  return (mapping.mesh >= serial_minwith || mapping.rotate) != 0;
+}
+
+bool maker::increment(serial_t &serial, int delta) const {
  if (serial > mask(mapping.width)) {
    log_extra("keygen-increment: %" PRIu64 " > %" PRIu64 ", overflow", serial,
              mask(mapping.width));
@@ -175,7 +184,8 @@ bool maker::increment(serial_t &serial, int delta) {
  }

  serial_t target = serial + (int64_t)delta;
-  if (target > mask(mapping.width)) {
+  if (target > mask(mapping.width) ||
+      ((delta > 0) ? target < serial : target > serial)) {
    log_extra("keygen-increment: %" PRIu64 "%-d => %" PRIu64 ", overflow",
              serial, delta, target);
    return false;
--- a/test/keygen.h
+++ b/test/keygen.h
@@ -1,4 +1,4 @@
-/*
+/*
 * Copyright 2017-2019 Leonid Yuriev <leo@yuriev.ru>
 * and other libmdbx authors: please see AUTHORS file.
 * All rights reserved.
@@ -121,8 +121,10 @@ public:
            serial_t value_age);
  void setup(const config::actor_params_pod &actor, unsigned actor_id,
             unsigned thread_number);
+  void make_ordered();
+  bool is_unordered() const;

-  bool increment(serial_t &serial, int delta);
+  bool increment(serial_t &serial, int delta) const;
 };

 } /* namespace keygen */
--- a/test/long_stochastic.sh
+++ b/test/long_stochastic.sh
@@ -0,0 +1,108 @@
+#!/bin/bash
+set -euo pipefail
+TESTDB_PREFIX=${1:-/dev/shm/mdbx-gc-test}.
+
+rm -f $(dirname ${TESTDB_PREFIX})/*
+
+if LC_ALL=C free | grep -q -i available; then
+	ram_avail_mb=$(($(LC_ALL=C free | grep -i Mem: | tr -s [:blank:] ' ' | cut -d ' ' -f 7) / 1024))
+else
+	ram_avail_mb=$(($(LC_ALL=C free | grep -i Mem: | tr -s [:blank:] ' ' | cut -d ' ' -f 4) / 1024))
+fi
+
+ram_reserve4logs_mb=3333
+if [ ${ram_avail_mb} -lt ${ram_reserve4logs_mb} ]; then
+	echo "=== At least ${ram_reserve4logs_mb}Mb RAM required"
+	exit -2
+fi
+
+#
+# В режимах отличных от MDBX_WRITEMAP изменения до записи в файл
+# будут накапливаться в памяти, что может потребовать свободной
+# памяти размером с БД. Кроме этого, в тест входит сценарий
+# создания копия БД на ходу. Поэтому БД не может быть больше 1/3
+# от доступной памяти. Однако, следует учесть что malloc() будет
+# не сразу возвращать выделенную память системе, а также
+# предусмотреть места для логов.
+#
+# In non-MDBX_WRITEMAP modes, updates (dirty pages) will
+# accumulate in memory before writing to the disk, which may
+# require a free memory up to the size of a whole database. In
+# addition, the test includes a script create a copy of the
+# database on the go. Therefore, the database cannot be more 1/3
+# of available memory. Moreover, should be taken into account
+# that malloc() will not return the allocated memory to the
+# system immediately, as well some space is required for logs.
+#
+db_size_mb=$(expr '(' ${ram_avail_mb} - ${ram_reserve4logs_mb} ')' / 4)
+if [ ${db_size_mb} -gt 3072 ]; then
+db_size_mb=3072
+fi
+echo "=== ${ram_avail_mb}M RAM available, use ${db_size_mb}M for DB"
+
+make check
+rm -f $(dirname ${TESTDB_PREFIX})/*
+
+###############################################################################
+
+function rep9 { printf "%*s" $1 '' | tr ' ' '9'; }
+function join { local IFS="$1"; shift; echo "$*"; }
+function bit2option { local -n arr=$1; (( ($2&(1<<$3)) != 0 )) && echo -n '+' || echo -n '-'; echo "${arr[$3]}"; }
+
+options=(writemap coalesce lifo)
+
+function bits2list {
+	local -n arr=$1
+	local i
+	local list=()
+	for ((i=0; i<${#arr[@]}; ++i)) do
+		list[$i]=$(bit2option $1 $2 $i)
+	done
+	join , "${list[@]}"
+}
+
+function probe {
+	echo "=============================================== $(date)"
+	echo "${caption}: $*"
+	rm -f ${TESTDB_PREFIX}* \
+		&& ./mdbx_test --ignore-dbfull --repeat=42 --pathname=${TESTDB_PREFIX}db "$@" | lz4 > ${TESTDB_PREFIX}log.lz4 \
+		&& ./mdbx_chk -nvvv ${TESTDB_PREFIX}db | tee ${TESTDB_PREFIX}chk \
+		&& ([ ! -e ${TESTDB_PREFIX}db-copy ] || ./mdbx_chk -nvvv ${TESTDB_PREFIX}db-copy | tee ${TESTDB_PREFIX}chk-copy) \
+		|| (echo "FAILED"; exit 1)
+}
+
+###############################################################################
+
+count=0
+for nops in $(seq 2 6); do
+	for ((wbatch=nops-1; wbatch > 0; --wbatch)); do
+		loops=$(((111 >> nops) / nops + 3))
+		for ((rep=0; rep++ < loops; )); do
+			for ((bits=2**${#options[@]}; --bits >= 0; )); do
+				seed=$(date +%N)
+				caption="Probe #$((++count)) int-key,w/o-dups, repeat ${rep} of ${loops}" probe \
+					--pagesize=min --size-upper=${db_size_mb}M --table=+key.integer,-data.dups --keylen.min=min --keylen.max=max --datalen.min=min --datalen.max=1111 \
+					--nops=$( rep9 $nops ) --batch.write=$( rep9 $wbatch ) --mode=$(bits2list options $bits) \
+					--keygen.seed=${seed} basic
+				caption="Probe #$((++count)) int-key,with-dups, repeat ${rep} of ${loops}" probe \
+					--pagesize=min --size-upper=${db_size_mb}M --table=+key.integer,+data.dups --keylen.min=min --keylen.max=max --datalen.min=min --datalen.max=max \
+					--nops=$( rep9 $nops ) --batch.write=$( rep9 $wbatch ) --mode=$(bits2list options $bits) \
+					--keygen.seed=${seed} basic
+				caption="Probe #$((++count)) int-key,int-data, repeat ${rep} of ${loops}" probe \
+					--pagesize=min --size-upper=${db_size_mb}M --table=+key.integer,+data.integer --keylen.min=min --keylen.max=max --datalen.min=min --datalen.max=max \
+					--nops=$( rep9 $nops ) --batch.write=$( rep9 $wbatch ) --mode=$(bits2list options $bits) \
+					--keygen.seed=${seed} basic
+				caption="Probe #$((++count)) w/o-dups, repeat ${rep} of ${loops}" probe \
+					--pagesize=min --size-upper=${db_size_mb}M --table=-data.dups --keylen.min=min --keylen.max=max --datalen.min=min --datalen.max=1111 \
+					--nops=$( rep9 $nops ) --batch.write=$( rep9 $wbatch ) --mode=$(bits2list options $bits) \
+					--keygen.seed=${seed} basic
+				caption="Probe #$((++count)) with-dups, repeat ${rep} of ${loops}" probe \
+					--pagesize=min --size-upper=${db_size_mb}M --table=+data.dups --keylen.min=min --keylen.max=max --datalen.min=min --datalen.max=max \
+					--nops=$( rep9 $nops ) --batch.write=$( rep9 $wbatch ) --mode=$(bits2list options $bits) \
+					--keygen.seed=${seed} basic
+			done
+		done
+	done
+done
+
+echo "=== ALL DONE ====================== $(date)"
--- a/test/loop.bat
+++ b/test/loop.bat
@@ -1,15 +0,0 @@
-@echo off
-
-del test.db test.db-lck
-
-:loop
-
-mdbx_test.exe --pathname=test.db --dont-cleanup-after basic > test.log
-if errorlevel 1 goto fail
-
-mdbx_chk.exe -nvvv test.db > chk.log
-if errorlevel 1 goto fail
-goto loop
-
-:fail
-echo FAILED
--- a/test/main.cc
+++ b/test/main.cc
@@ -27,6 +27,8 @@ void actor_params::set_defaults(const std::string &tmpdir) {
  loglevel =
 #ifdef NDEBUG
      logging::info;
+#elif defined(_WIN32) || defined(_WIN64)
+      logging::verbose;
 #else
      logging::trace;
 #endif
@@ -37,7 +39,8 @@ void actor_params::set_defaults(const std::string &tmpdir) {
  table_flags = MDBX_DUPSORT;

  size_lower = -1;
-  size_now = 1024 * 1024 * ((table_flags & MDBX_DUPSORT) ? 4 : 256);
+  size_now =
+      intptr_t(1024) * 1024 * ((table_flags & MDBX_DUPSORT) ? 256 : 1024);
  size_upper = -1;
  shrink_threshold = -1;
  growth_step = -1;
@@ -61,14 +64,15 @@ void actor_params::set_defaults(const std::string &tmpdir) {
  datalen_min = mdbx_datalen_min();
  datalen_max = std::min(mdbx_datalen_max(), 256u * 1024 + 42);

-  batch_read = 4;
-  batch_write = 4;
+  batch_read = 42;
+  batch_write = 42;

  delaystart = 0;
  waitfor_nops = 0;
  inject_writefaultn = 0;

  drop_table = false;
+  ignore_dbfull = false;

  max_readers = 42;
  max_tables = 42;
@@ -287,6 +291,9 @@ int main(int argc, char *const argv[]) {
      continue;
    if (config::parse_option(argc, argv, narg, "drop", params.drop_table))
      continue;
+    if (config::parse_option(argc, argv, narg, "ignore-dbfull",
+                             params.ignore_dbfull))
+      continue;
    if (config::parse_option(argc, argv, narg, "dump-config",
                             global::config::dump_config))
      continue;
@@ -341,6 +348,14 @@ int main(int argc, char *const argv[]) {
      configure_actor(last_space_id, ac_copy, value, params);
      continue;
    }
+    if (config::parse_option(argc, argv, narg, "append", nullptr)) {
+      configure_actor(last_space_id, ac_append, value, params);
+      continue;
+    }
+    if (config::parse_option(argc, argv, narg, "ttl", nullptr)) {
+      configure_actor(last_space_id, ac_ttl, value, params);
+      continue;
+    }
    if (config::parse_option(argc, argv, narg, "failfast",
                             global::config::failfast))
      continue;
--- a/test/osal-unix.cc
+++ b/test/osal-unix.cc
@@ -199,7 +199,9 @@ retry:
    if (WIFEXITED(status))
      childs[pid] =
          (WEXITSTATUS(status) == EXIT_SUCCESS) ? as_successful : as_failed;
-    else if (WIFSIGNALED(status) || WCOREDUMP(status))
+    else if (WCOREDUMP(status))
+      childs[pid] = as_coredump;
+    else if (WIFSIGNALED(status))
      childs[pid] = as_killed;
    else if (WIFSTOPPED(status))
      childs[pid] = as_debuging;
@@ -216,7 +218,7 @@ retry:
    if (ts.tv_sec == 0 && ts.tv_nsec == 0)
      ts.tv_nsec = 1;
    if (nanosleep(&ts, &ts) == 0) {
-      /* timeout and no signal fomr child */
+      /* timeout and no signal from child */
      pid = 0;
      return 0;
    }
--- a/test/osal-windows.cc
+++ b/test/osal-windows.cc
@@ -312,14 +312,22 @@ actor_status osal_actor_info(const mdbx_pid_t pid) {
  case EXIT_SUCCESS:
    status = as_successful;
    break;
-  // case EXCEPTION_BREAKPOINT:
+  case EXCEPTION_BREAKPOINT:
  case EXCEPTION_SINGLE_STEP:
    status = as_debuging;
    break;
  case STATUS_CONTROL_C_EXIT:
-  case EXCEPTION_NONCONTINUABLE_EXCEPTION:
    status = as_killed;
    break;
+  case EXCEPTION_ACCESS_VIOLATION:
+  case EXCEPTION_ARRAY_BOUNDS_EXCEEDED:
+  case EXCEPTION_DATATYPE_MISALIGNMENT:
+  case EXCEPTION_STACK_OVERFLOW:
+  case EXCEPTION_INVALID_DISPOSITION:
+  case EXCEPTION_ILLEGAL_INSTRUCTION:
+  case EXCEPTION_NONCONTINUABLE_EXCEPTION:
+    status = as_coredump;
+    break;
  default:
    status = as_failed;
    break;
--- a/test/test.cc
+++ b/test/test.cc
@@ -33,6 +33,10 @@ const char *testcase2str(const actor_testcase testcase) {
    return "try";
  case ac_copy:
    return "copy";
+  case ac_append:
+    return "append";
+  case ac_ttl:
+    return "ttl";
  }
 }

@@ -51,6 +55,8 @@ const char *status2str(actor_status status) {
    return "killed";
  case as_failed:
    return "failed";
+  case as_coredump:
+    return "coredump";
  }
 }

@@ -133,6 +139,8 @@ void testcase::db_open() {

  if (!db_guard)
    db_prepare();
+
+  jitter_delay(true);
  int rc = mdbx_env_open(db_guard.get(), config.params.pathname_db.c_str(),
                         (unsigned)config.params.mode_flags, 0640);
  if (unlikely(rc != MDBX_SUCCESS))
@@ -166,28 +174,84 @@ void testcase::txn_begin(bool readonly, unsigned flags) {
            flags);
 }

+int testcase::breakable_commit() {
+  int rc = MDBX_SUCCESS;
+  log_trace(">> txn_commit");
+  assert(txn_guard);
+
+  MDBX_txn *txn = txn_guard.release();
+  txn_inject_writefault(txn);
+  int err = mdbx_txn_commit(txn);
+  if (unlikely(err != MDBX_SUCCESS)) {
+    if (err == MDBX_MAP_FULL && config.params.ignore_dbfull) {
+      rc = err;
+      err = mdbx_txn_abort(txn);
+      if (unlikely(err != MDBX_SUCCESS && err != MDBX_THREAD_MISMATCH))
+        failure_perror("mdbx_txn_abort()", err);
+    } else
+      failure_perror("mdbx_txn_commit()", err);
+  }
+
+  log_trace("<< txn_commit: %s", rc ? "failed" : "Ok");
+  return rc;
+}
+
 void testcase::txn_end(bool abort) {
  log_trace(">> txn_end(%s)", abort ? "abort" : "commit");
  assert(txn_guard);

  MDBX_txn *txn = txn_guard.release();
  if (abort) {
-    int rc = mdbx_txn_abort(txn);
-    if (unlikely(rc != MDBX_SUCCESS))
-      failure_perror("mdbx_txn_abort()", rc);
+    int err = mdbx_txn_abort(txn);
+    if (unlikely(err != MDBX_SUCCESS && err != MDBX_THREAD_MISMATCH))
+      failure_perror("mdbx_txn_abort()", err);
  } else {
    txn_inject_writefault(txn);
-    int rc = mdbx_txn_commit(txn);
-    if (unlikely(rc != MDBX_SUCCESS))
-      failure_perror("mdbx_txn_commit()", rc);
+    int err = mdbx_txn_commit(txn);
+    if (unlikely(err != MDBX_SUCCESS))
+      failure_perror("mdbx_txn_commit()", err);
  }

  log_trace("<< txn_end(%s)", abort ? "abort" : "commit");
 }

+void testcase::cursor_open(unsigned dbi) {
+  log_trace(">> cursor_open(%u)", dbi);
+  assert(!cursor_guard);
+  assert(txn_guard);
+
+  MDBX_cursor *cursor = nullptr;
+  int rc = mdbx_cursor_open(txn_guard.get(), dbi, &cursor);
+  if (unlikely(rc != MDBX_SUCCESS))
+    failure_perror("mdbx_cursor_open()", rc);
+  cursor_guard.reset(cursor);
+
+  log_trace("<< cursor_open(%u)", dbi);
+}
+
+void testcase::cursor_close() {
+  log_trace(">> cursor_close()");
+  assert(cursor_guard);
+  MDBX_cursor *cursor = cursor_guard.release();
+  mdbx_cursor_close(cursor);
+  log_trace("<< cursor_close()");
+}
+
+int testcase::breakable_restart() {
+  int rc = MDBX_SUCCESS;
+  if (txn_guard)
+    rc = breakable_commit();
+  if (cursor_guard)
+    cursor_close();
+  txn_begin(false, 0);
+  return rc;
+}
+
 void testcase::txn_restart(bool abort, bool readonly, unsigned flags) {
  if (txn_guard)
    txn_end(abort);
+  if (cursor_guard)
+    cursor_close();
  txn_begin(readonly, flags);
 }

@@ -296,6 +360,7 @@ bool testcase::setup() {
    return false;

  start_timestamp = chrono::now_motonic();
+  nops_completed = 0;
  return true;
 }

@@ -365,6 +430,28 @@ void testcase::update_canary(uint64_t increment) {
  log_trace("<< update_canary: sequence = %" PRIu64, canary_now.y);
 }

+int testcase::db_open__begin__table_create_open_clean(MDBX_dbi &dbi) {
+  db_open();
+
+  int err, retry_left = 42;
+  for (;;) {
+    txn_begin(false);
+    dbi = db_table_open(true);
+    db_table_clear(dbi);
+    err = breakable_commit();
+    if (likely(err == MDBX_SUCCESS)) {
+      txn_begin(false);
+      return MDBX_SUCCESS;
+    }
+    if (--retry_left == 0)
+      break;
+    jitter_delay(true);
+  }
+  log_notice("db_begin_table_create_open_clean: bailout due '%s'",
+             mdbx_strerror(err));
+  return err;
+}
+
 MDBX_dbi testcase::db_table_open(bool create) {
  log_trace(">> testcase::db_table_create");

@@ -396,13 +483,21 @@ void testcase::db_table_drop(MDBX_dbi handle) {
  if (config.params.drop_table) {
    int rc = mdbx_drop(txn_guard.get(), handle, true);
    if (unlikely(rc != MDBX_SUCCESS))
-      failure_perror("mdbx_drop()", rc);
+      failure_perror("mdbx_drop(delete=true)", rc);
    log_trace("<< testcase::db_table_drop");
  } else {
    log_trace("<< testcase::db_table_drop: not needed");
  }
 }

+void testcase::db_table_clear(MDBX_dbi handle) {
+  log_trace(">> testcase::db_table_clear, handle %u", handle);
+  int rc = mdbx_drop(txn_guard.get(), handle, false);
+  if (unlikely(rc != MDBX_SUCCESS))
+    failure_perror("mdbx_drop(delete=false)", rc);
+  log_trace("<< testcase::db_table_clear");
+}
+
 void testcase::db_table_close(MDBX_dbi handle) {
  log_trace(">> testcase::db_table_close, handle %u", handle);
  assert(!txn_guard);
@@ -424,8 +519,9 @@ void testcase::checkdata(const char *step, MDBX_dbi handle, MDBX_val key2check,

 //-----------------------------------------------------------------------------

-bool test_execute(const actor_config &config) {
+bool test_execute(const actor_config &config_const) {
  const mdbx_pid_t pid = osal_getpid();
+  actor_config config = config_const;

  if (global::singlemode) {
    logging::setup(format("single_%s", testcase2str(config.testcase)));
@@ -458,23 +554,48 @@ bool test_execute(const actor_config &config) {
    case ac_copy:
      test.reset(new testcase_copy(config, pid));
      break;
+    case ac_append:
+      test.reset(new testcase_append(config, pid));
+      break;
+    case ac_ttl:
+      test.reset(new testcase_ttl(config, pid));
+      break;
    default:
      test.reset(new testcase(config, pid));
      break;
    }

-    if (!test->setup())
-      log_notice("test setup failed");
-    else if (!test->run())
-      log_notice("test failed");
-    else if (!test->teardown())
-      log_notice("test teardown failed");
-    else {
-      log_info("test successed");
-      return true;
-    }
+    size_t iter = 0;
+    do {
+      iter++;
+      if (!test->setup()) {
+        log_notice("test setup failed");
+        return false;
+      }
+      if (!test->run()) {
+        log_notice("test failed");
+        return false;
+      }
+      if (!test->teardown()) {
+        log_notice("test teardown failed");
+        return false;
+      }
+
+      if (config.params.nrepeat == 1)
+        log_info("test successed");
+      else {
+        if (config.params.nrepeat)
+          log_info("test successed (iteration %zi of %zi)", iter,
+                   size_t(config.params.nrepeat));
+        else
+          log_info("test successed (iteration %zi)", iter);
+        config.params.keygen.seed += INT32_C(0xA4F4D37B);
+      }
+
+    } while (config.params.nrepeat == 0 || iter < config.params.nrepeat);
+    return true;
  } catch (const std::exception &pipets) {
    failure("***** Exception: %s *****", pipets.what());
+    return false;
  }
-  return false;
 }
--- a/test/test.h
+++ b/test/test.h
@@ -105,8 +105,12 @@ protected:
  void db_open();
  void db_close();
  void txn_begin(bool readonly, unsigned flags = 0);
+  int breakable_commit();
  void txn_end(bool abort);
+  int breakable_restart();
  void txn_restart(bool abort, bool readonly, unsigned flags = 0);
+  void cursor_open(unsigned dbi);
+  void cursor_close();
  void txn_inject_writefault(void);
  void txn_inject_writefault(MDBX_txn *txn);
  void fetch_canary();
@@ -117,7 +121,9 @@ protected:

  MDBX_dbi db_table_open(bool create);
  void db_table_drop(MDBX_dbi handle);
+  void db_table_clear(MDBX_dbi handle);
  void db_table_close(MDBX_dbi handle);
+  int db_open__begin__table_create_open_clean(MDBX_dbi &dbi);

  bool wait4start();
  void report(size_t nops_done);
@@ -151,6 +157,13 @@ public:
  virtual ~testcase() {}
 };

+class testcase_ttl : public testcase {
+public:
+  testcase_ttl(const actor_config &config, const mdbx_pid_t pid)
+      : testcase(config, pid) {}
+  bool run();
+};
+
 class testcase_hill : public testcase {
 public:
  testcase_hill(const actor_config &config, const mdbx_pid_t pid)
@@ -158,6 +171,13 @@ public:
  bool run();
 };

+class testcase_append : public testcase {
+public:
+  testcase_append(const actor_config &config, const mdbx_pid_t pid)
+      : testcase(config, pid) {}
+  bool run();
+};
+
 class testcase_deadread : public testcase {
 public:
  testcase_deadread(const actor_config &config, const mdbx_pid_t pid)
--- a/test/test.vcxproj
+++ b/test/test.vcxproj
@@ -181,6 +181,8 @@
    <ClInclude Include="utils.h" />
  </ItemGroup>
  <ItemGroup>
+    <ClCompile Include="ttl.cc" />
+    <ClCompile Include="append.cc" />
    <ClCompile Include="cases.cc" />
    <ClCompile Include="chrono.cc" />
    <ClCompile Include="config.cc" />
--- a/test/ttl.cc
+++ b/test/ttl.cc
@@ -0,0 +1,164 @@
+/*
+ * Copyright 2017-2019 Leonid Yuriev <leo@yuriev.ru>
+ * and other libmdbx authors: please see AUTHORS file.
+ * All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted only as authorized by the OpenLDAP
+ * Public License.
+ *
+ * A copy of this license is available in the file LICENSE in the
+ * top-level directory of the distribution or, alternatively, at
+ * <http://www.OpenLDAP.org/license.html>.
+ */
+
+#include "test.h"
+#include <cmath>
+#include <deque>
+
+static unsigned edge2window(uint64_t edge, unsigned window_max) {
+  const double rnd = u64_to_double1(bleach64(edge));
+  const unsigned window = window_max - std::lrint(std::pow(window_max, rnd));
+  return window;
+}
+
+static unsigned edge2count(uint64_t edge, unsigned count_max) {
+  const double rnd = u64_to_double1(prng64_map1_white(edge));
+  const unsigned count = std::lrint(std::pow(count_max, rnd));
+  return count;
+}
+
+bool testcase_ttl::run() {
+  MDBX_dbi dbi;
+  int err = db_open__begin__table_create_open_clean(dbi);
+  if (unlikely(err != MDBX_SUCCESS)) {
+    log_notice("ttl: bailout-prepare due '%s'", mdbx_strerror(err));
+    return true;
+  }
+
+  /* LY: тест "эмуляцией time-to-live":
+   *  - организуется "скользящее окно", которое двигается вперед вдоль
+   *    числовой оси каждую транзакцию.
+   *  - по переднему краю "скользящего окна" записи добавляются в таблицу,
+   *    а по заднему удаляются.
+   *  - количество добавляемых/удаляемых записей псевдослучайно зависит
+   *    от номера транзакции, но с экспоненциальным распределением.
+   *  - размер "скользящего окна" также псевдослучайно зависит от номера
+   *    транзакции с "отрицательным" экспоненциальным распределением
+   *    MAX_WIDTH - exp(rnd(N)), при уменьшении окна сдвигается задний
+   *    край и удаляются записи позади него.
+   *
+   *  Таким образом имитируется поведение таблицы с TTL: записи стохастически
+   *  добавляются и удаляются, но изредка происходят массивные удаления.
+   */
+
+  /* LY: для параметризации используем подходящие параметры, которые не имеют
+   * здесь смысла в первоначальном значении */
+  const unsigned window_max =
+      (config.params.batch_read > 999) ? config.params.batch_read : 1000;
+  const unsigned count_max =
+      (config.params.batch_write > 999) ? config.params.batch_write : 1000;
+  log_info("ttl: using `batch_read` value %u for window_max", window_max);
+  log_info("ttl: using `batch_write` value %u for count_max", count_max);
+
+  uint64_t seed =
+      prng64_map2_white(config.params.keygen.seed) + config.actor_id;
+  keyvalue_maker.setup(config.params, config.actor_id, 0 /* thread_number */);
+  key = keygen::alloc(config.params.keylen_max);
+  data = keygen::alloc(config.params.datalen_max);
+  const unsigned insert_flags = (config.params.table_flags & MDBX_DUPSORT)
+                                    ? MDBX_NODUPDATA
+                                    : MDBX_NODUPDATA | MDBX_NOOVERWRITE;
+
+  std::deque<std::pair<uint64_t, unsigned>> fifo;
+  uint64_t serial = 0;
+  while (should_continue()) {
+    const uint64_t salt = prng64_white(seed) /* mdbx_txn_id(txn_guard.get()) */;
+
+    const unsigned window_width = edge2window(salt, window_max);
+    unsigned head_count = edge2count(salt, count_max);
+    log_verbose("ttl: step #%zu (serial %" PRIu64
+                ", window %u, count %u) salt %" PRIu64,
+                nops_completed, serial, window_width, head_count, salt);
+
+    if (window_width) {
+      while (fifo.size() > window_width) {
+        uint64_t tail_serial = fifo.back().first;
+        const unsigned tail_count = fifo.back().second;
+        log_trace("ttl: pop-tail (serial %" PRIu64 ", count %u)", tail_serial,
+                  tail_count);
+        fifo.pop_back();
+        for (unsigned n = 0; n < tail_count; ++n) {
+          log_trace("ttl: remove-tail %" PRIu64, serial);
+          generate_pair(tail_serial);
+          err = mdbx_del(txn_guard.get(), dbi, &key->value, &data->value);
+          if (unlikely(err != MDBX_SUCCESS)) {
+            if (err == MDBX_MAP_FULL && config.params.ignore_dbfull) {
+              log_notice("ttl: tail-bailout due '%s'", mdbx_strerror(err));
+              goto bailout;
+            }
+            failure_perror("mdbx_del(tail)", err);
+          }
+          if (unlikely(!keyvalue_maker.increment(tail_serial, 1)))
+            failure("ttl: unexpected key-space overflow on the tail");
+        }
+      }
+    } else {
+      log_trace("ttl: purge state");
+      db_table_clear(dbi);
+      fifo.clear();
+    }
+
+    err = breakable_restart();
+    if (unlikely(err != MDBX_SUCCESS)) {
+      log_notice("ttl: bailout at commit due '%s'", mdbx_strerror(err));
+      break;
+    }
+    fifo.push_front(std::make_pair(serial, head_count));
+  retry:
+    for (unsigned n = 0; n < head_count; ++n) {
+      log_trace("ttl: insert-head %" PRIu64, serial);
+      generate_pair(serial);
+      err = mdbx_put(txn_guard.get(), dbi, &key->value, &data->value,
+                     insert_flags);
+      if (unlikely(err != MDBX_SUCCESS)) {
+        if (err == MDBX_MAP_FULL && config.params.ignore_dbfull) {
+          log_notice("ttl: head-insert skip due '%s'", mdbx_strerror(err));
+          txn_restart(true, false);
+          serial = fifo.front().first;
+          fifo.front().second = head_count = n;
+          goto retry;
+        }
+        failure_perror("mdbx_put(head)", err);
+      }
+
+      if (unlikely(!keyvalue_maker.increment(serial, 1))) {
+        log_notice("ttl: unexpected key-space overflow");
+        goto bailout;
+      }
+    }
+    err = breakable_restart();
+    if (unlikely(err != MDBX_SUCCESS)) {
+      log_notice("ttl: head-commit skip due '%s'", mdbx_strerror(err));
+      serial = fifo.front().first;
+      fifo.pop_front();
+    }
+    report(1);
+  }
+
+bailout:
+  txn_end(true);
+  if (dbi) {
+    if (config.params.drop_table && !mode_readonly()) {
+      txn_begin(false);
+      db_table_drop(dbi);
+      err = breakable_commit();
+      if (unlikely(err != MDBX_SUCCESS)) {
+        log_notice("ttl: bailout-clean due '%s'", mdbx_strerror(err));
+        return true;
+      }
+    } else
+      db_table_close(dbi);
+  }
+  return true;
+}
--- a/test/utils.cc
+++ b/test/utils.cc
@@ -252,21 +252,8 @@ uint64_t entropy_ticks(void) {

 //-----------------------------------------------------------------------------

-static __inline uint64_t bleach64(uint64_t dirty) {
-  return mul_64x64_high(bswap64(dirty), UINT64_C(17048867929148541611));
-}
-
-static __inline uint32_t bleach32(uint32_t dirty) {
-  return (uint32_t)((bswap32(dirty) * UINT64_C(2175734609)) >> 32);
-}
-
-uint64_t prng64_careless(uint64_t &state) {
-  state = state * UINT64_C(6364136223846793005) + 1;
-  return state;
-}
-
 uint64_t prng64_white(uint64_t &state) {
-  state = state * UINT64_C(6364136223846793005) + UINT64_C(1442695040888963407);
+  state = prng64_map2_careless(state);
  return bleach64(state);
 }

@@ -366,7 +353,7 @@ void jitter_delay(bool extra) {
        cpu_relax();
        if (dice > 2) {
          unsigned us = entropy_white() &
-                        (extra ? 0xfffff /* 1.05 s */ : 0x3ff /* 1 ms */);
+                        (extra ? 0xffff /* 656 ms */ : 0x3ff /* 1 ms */);
          log_trace("== jitter.delay: %0.6f", us / 1000000.0);
          osal_udelay(us);
        }
--- a/test/utils.h
+++ b/test/utils.h
@@ -1,4 +1,4 @@
-/*
+/*
 * Copyright 2017-2019 Leonid Yuriev <leo@yuriev.ru>
 * and other libmdbx authors: please see AUTHORS file.
 * All rights reserved.
@@ -288,6 +288,7 @@ struct simple_checksum {
  void push(uint32_t data) {
    value += data * UINT64_C(9386433910765580089) + 1;
    value ^= value >> 41;
+    value *= UINT64_C(0xBD9CACC22C6E9571);
  }

  void push(uint64_t data) {
@@ -304,11 +305,15 @@ struct simple_checksum {
  }

  void push(const double &data) { push(&data, sizeof(double)); }
-
  void push(const char *cstr) { push(cstr, strlen(cstr)); }
-
  void push(const std::string &str) { push(str.data(), str.size()); }

+  void push(unsigned salt, const MDBX_val &val) {
+    push(val.iov_len);
+    push(salt);
+    push(val.iov_base, val.iov_len);
+  }
+
 #if defined(_WIN32) || defined(_WIN64) || defined(_WINDOWS)
  void push(const HANDLE &handle) { push(&handle, sizeof(handle)); }
 #endif /* _WINDOWS */
@@ -322,7 +327,60 @@ std::string format(const char *fmt, ...);

 uint64_t entropy_ticks(void);
 uint64_t entropy_white(void);
-uint64_t prng64_careless(uint64_t &state);
+static inline uint64_t bleach64(uint64_t v) {
+  // Tommy Ettinger, https://www.blogger.com/profile/04953541827437796598
+  // http://mostlymangling.blogspot.com/2019/01/better-stronger-mixer-and-test-procedure.html
+  v ^= rot64(v, 25) ^ rot64(v, 50);
+  v *= UINT64_C(0xA24BAED4963EE407);
+  v ^= rot64(v, 24) ^ rot64(v, 49);
+  v *= UINT64_C(0x9FB21C651E98DF25);
+  return v ^ v >> 28;
+}
+
+static inline uint32_t bleach32(uint32_t x) {
+  // https://github.com/skeeto/hash-prospector
+  // exact bias: 0.17353355999581582
+  x ^= x >> 16;
+  x *= UINT32_C(0x7feb352d);
+  x ^= 0x3027C563 ^ (x >> 15);
+  x *= UINT32_C(0x846ca68b);
+  x ^= x >> 16;
+  return x;
+}
+
+static inline uint64_t prng64_map1_careless(uint64_t state) {
+  return state * UINT64_C(6364136223846793005) + 1;
+}
+
+static inline uint64_t prng64_map2_careless(uint64_t state) {
+  return (state + UINT64_C(1442695040888963407)) *
+         UINT64_C(6364136223846793005);
+}
+
+static inline uint64_t prng64_map1_white(uint64_t state) {
+  return bleach64(prng64_map1_careless(state));
+}
+
+static inline uint64_t prng64_map2_white(uint64_t state) {
+  return bleach64(prng64_map2_careless(state));
+}
+
+static inline uint64_t prng64_careless(uint64_t &state) {
+  state = prng64_map1_careless(state);
+  return state;
+}
+
+static inline double u64_to_double1(uint64_t v) {
+  union {
+    uint64_t u64;
+    double d;
+  } casting;
+
+  casting.u64 = UINT64_C(0x3ff) << 52 | (v >> 12);
+  assert(casting.d >= 1.0 && casting.d < 2.0);
+  return casting.d - 1.0;
+}
+
 uint64_t prng64_white(uint64_t &state);
 uint32_t prng32(uint64_t &state);
 void prng_fill(uint64_t &state, void *ptr, size_t bytes);
Author	SHA1	Message	Date
Leonid Yuriev	eefbb7f063	mdbx: remove extra prep_backlog_data() inside update_gc(). Change-Id: Ic6250159c5abe03ed0705c7a432c5eb5b17b790e	2019-07-09 15:58:48 +03:00
Leonid Yuriev	9dc0b6a3f3	mdbx-test: partialy revert long_stochastic.sh changes. Change-Id: I1fd4d601a91b50e82bbf8f183c968d31d671957e	2019-07-09 14:31:27 +03:00
Leonid Yuriev	4ecf78163d	mdbx: fix/refine mdbx_find_largest(). Change-Id: Ib40ca0014da18793be0bdae5133806304ad42819	2019-07-09 03:27:23 +03:00
Leonid Yuriev	421f4b01e6	mdbx-test: refine internals (delays, logging). Change-Id: Ife19e3bc1b124629891ba03fd9e1bf3b2ad3e301	2019-07-09 01:39:32 +03:00
Leonid Yuriev	53563e75bb	mdbx: fix DB-shrinking race with copy-asis & readers. Change-Id: I1e05616de03d814850a1c8ad20e83941b2d1c911	2019-07-09 00:41:04 +03:00
Leonid Yuriev	bd5078347b	mdbx: refine backlog preparation inside update_gc(). Change-Id: Ib18842c2922afba794d6ab69337580bcea29bfe6	2019-07-09 00:41:04 +03:00
Leonid Yuriev	06f16464aa	mdbx-test: adds more jitter while DB opening. Change-Id: Ic0f03f4fbe064ea00f719baa11b28cc24aafcb5b	2019-07-07 21:21:54 +03:00
Leonid Yuriev	1f89c494ff	mdbx: refines OFF_T_MAX if it is not defined. Change-Id: Ic453140e01725493cad0c98e73244c747d5f490e	2019-07-07 20:04:39 +03:00
Leonid Yuriev	8606803344	mdbx: adds paranoid mvcc-checks into read transactions. Change-Id: I7f0f05010306b02b3b1ab8478217de70ffd211a0	2019-07-07 19:51:06 +03:00
Leonid Yuriev	41de36e93d	mdbx: avoids infinite copy_file_range-loop when file is unexpectedly truncated. Change-Id: I5a542a49fa4ac9ced8ca60266a084d52d3618804	2019-07-07 18:42:14 +03:00
Leonid Yuriev	30ac62ceec	mdbx-test: add `coredump` status. Change-Id: I9003a891744eb78d91a3e279e20efcb4c1ad736c	2019-07-07 02:33:39 +03:00
Leonid Yuriev	7e1e142104	mdbx-chk: fix/refine error handling for sub-DBs. Change-Id: I42460793bbe47815add1b3f61f3746f671a749d9	2019-07-07 01:47:10 +03:00
Leonid Yuriev	e2f37908b9	mdbx: fix/refine error handling while DB-pages walking. Change-Id: If910e96cc7c30577531aab24b5a9573e5fe9126d	2019-07-06 21:25:11 +03:00
Leonid Yuriev	6d4e151ba8	mdbx-test: fix comment typo (minor). Change-Id: I7bd25ccfc52371c78f9011ccd45203e44e619142	2019-07-02 00:21:51 +03:00
Leonid Yuriev	40112ebd62	mdbx: engage copy_file_range() for env_copy_as_is. Change-Id: I7ea17914d80500ffa70451b80920d726f0e9c2f8	2019-07-01 18:37:36 +03:00
Leonid Yuriev	6960c45e59	mdbx: avoids EFAULT "Bad address" while copy-as-is if DB is swapped-out from RAM. Change-Id: I711efc1c54a04745bd561bc5e1db5e6f6d8b7115	2019-07-01 17:55:58 +03:00
Leonid Yuriev	2e60256978	mdbx: drop EPIPE/SIGPIPE handling (since pipes are disallowed). Change-Id: I56d4539333edea93cc1a2c3606cf959c82b98b19	2019-07-01 16:39:18 +03:00
Leonid Yuriev	38110579ba	mdbx: drops mdbx_write(), using mdbx_pwrite() instead of. Change-Id: Iff3de2d5ef3fa2e92607d46b96d4526e464e593b	2019-07-01 16:39:18 +03:00
Leonid Yuriev	08c334c8bc	mdbx-test: limit DB-size 3Gb for long_stochastic. Change-Id: I52bc6cd4aea7a41d5b1bdabc2c8e4bd6e34a78a9	2019-06-27 10:53:35 +03:00
Leonid Yuriev	aaf49bb816	mdbx-test: don't fail on key-space overflow. Change-Id: I22a8cb359849c4c02cd393047cb7ea33974607fd	2019-06-27 10:40:00 +03:00
Leonid Yuriev	fa3adb759a	mdbx-test: reduce upper txn-volume inside long_stochastic.	2019-06-26 10:11:24 +03:00
Leonid Yuriev	834f6d0784	mdbx-test: fix iteration count. Change-Id: Iaf19af417e54ee4ad968722c94d377dab29be149	2019-06-25 15:48:03 +03:00
Leonid Yuriev	41d8f65e1e	mdbx-test: biggest case depending on the DB size inside long_stochastic. Change-Id: Ia8c09dd6b4240d76de1356aa1eecaa884636086a	2019-06-25 14:26:57 +03:00
Leonid Yuriev	c9c985ae5d	mdbx-test: more for `--ignore-dbfull` option. Change-Id: I92c284edd889455eefefec12d3315b6f5d37cdc5	2019-06-24 02:55:26 +03:00
Leonid Yuriev	da99dcdb87	mdbx-test: long_stochastic. Change-Id: I28248a8af9041dfa62388a3b4ded7e2a4fdc07a9	2019-06-24 02:14:50 +03:00
Leonid Yuriev	5b88fe819c	mdbx: fix pwrite() for WRITE_MAX. Change-Id: If4924d20c1e267c2d3a190c860b89fc2fda0d517	2019-06-24 02:14:50 +03:00
Leonid Yuriev	f627b33379	mdbx-test: fix dbsize-options handling. Change-Id: Ia51f802ac1ad4e8b1b059a3f3b38214bda6b43fc	2019-06-24 02:14:50 +03:00
Leonid Yuriev	cf004dddbc	mdbx: use single cursor instance inside mdbx_env_walk(). Change-Id: I72cade64468a42fd27ebb4955d71ecbbabe64987	2019-06-24 02:14:50 +03:00
Leonid Yuriev	2d5a3ebd8f	mdbx-test: add `--ignore-dbfull` option (major). Change-Id: I252f9c3679a371722a780913ba994ca3dee9b90a	2019-06-24 02:14:50 +03:00
Leonid Yuriev	728f98d3de	mdbx-check: refine leaf-pages info (cosmetics). Change-Id: I0fdb467f1c1d51bfcdcef5edfe99c8e9ad66037e	2019-06-23 14:26:11 +03:00
Leonid Yuriev	bbf8ef0a4b	mdbx-chk: fix space-usage statistics info. Change-Id: I0cbbbc481f2e6dc37b29f6603ec1ead43b5d1864	2019-06-23 14:07:13 +03:00
Leonid Yuriev	05cf301774	mdbx: bump version to v0.3.0 Change-Id: Ic09361eda834c75d4fdb37dcbe5e8edbf0317f9b	2019-06-22 22:23:25 +03:00
Leonid Yuriev	47beba1782	mdbx: more 'unlikely'. Change-Id: I472e4a922590cd4680a48416611cfd894fa120db	2019-06-22 22:23:17 +03:00
Leonid Yuriev	e3f8dc5501	mdbx: rework RECLAIMING inside update_gc(). Change-Id: I9cf592476780bfdb346472baa12497d68a3d5aad	2019-06-22 22:23:17 +03:00
Leonid Yuriev	15403aadad	mdbx-test: re-seed keygen over iterations. Change-Id: I2cfd635fc46c808dd8431217b75a30780e0c3958	2019-06-22 22:23:17 +03:00
Leonid Yuriev	e6ad443178	mdbx-test: refine 'ttl' testcase. Change-Id: Ic4d759cfa29496bd46fa50fef1e974847b52bb41	2019-06-22 13:16:54 +03:00
Leonid Yuriev	243b01dd63	mdbx-test: refine 'append' testcast (minor). Change-Id: I79ea16046713a085e62e01eeb0978fc4e6766750	2019-06-22 13:16:54 +03:00
Leonid Yuriev	3fc610f860	mdbx-test: use common keygen-seed for `ttl` testcase. Change-Id: I921fff0ee28df8a18b6a38801c275de3fa2563ab	2019-06-22 13:16:54 +03:00
Leonid Yuriev	bfa9fc25d6	mdbx-test: 5-repeats inside gc-test script.	2019-06-22 13:16:54 +03:00
Leonid Yuriev	2219802bca	mdbx-test: more for `ttl` testcase. Change-Id: I8a01963345a2e815ebb39a98939420b8edb53968	2019-06-22 02:00:52 +03:00
Leonid Yuriev	be0ec1d38d	mdbx: fix GC corruption due deep recursive rebalance from update_gc(). Change-Id: I810250deb25cd625e737000282b434e3158ef8cc	2019-06-22 02:00:47 +03:00
Leonid Yuriev	9cf9d6eac2	mdbx-test: add `ttl` testcase. Change-Id: Ia5d164fde250e959226a53c63fcaf024ffe965a2	2019-06-22 00:53:52 +03:00
Leonid Yuriev	eecec74e21	mdbx: more `unlikely` (minor). Change-Id: Id6139473b3e6a7c3f099acc64db3180448294d0a	2019-06-21 02:11:28 +03:00
Leonid Yuriev	52bc4a7f41	mdbx-check: minor refine to avoid deeply recursion. Change-Id: I67f83a232ef47899f43c242b7e6295de4d7ec909	2019-06-20 21:08:31 +03:00
Leonid Yuriev	49d0e872a1	mdbx: iterate & check before recursion inside mdbx_env_pgwalk(). Change-Id: I27058a33f6dece0c3f206283a42ff74e5727417f	2019-06-20 21:04:47 +03:00
Leonid Yuriev	c91cc85c1f	mdbx-chk: log sub-DBs when verbose > 0 (cosmetic). Change-Id: Ie73c0773929b51eb11fb02afe18bb01b59fb2612	2019-06-20 10:12:26 +03:00
Leonid Yuriev	86cfd86cda	mdbx-test: support for `repeat` parameter. Change-Id: I6de52cd21314935c123ac51537e1b893c39dd5ed	2019-06-20 02:41:22 +03:00
Leonid Yuriev	cd75c4f081	mdbx-chk: avoid continuing to check bad records. Change-Id: I03b9d425c8413d6cacc1b67ed4a8253a10a9d603	2019-06-20 00:52:00 +03:00
Leonid Yuriev	2bea60a1a4	mdbx-chk: avoid infinite loop/recursion while checking corrupted DB. Change-Id: I3edb053e4baedced8ce8e8cfa25f9851eaca35d1	2019-06-19 15:08:50 +03:00
Leonid Yuriev	c05702eacf	mdbx: add MDBX_PGWALK-tags and refine pgwalk internals. Change-Id: I1f4eb79463dc6eec3d94d43baab0b28ceefa8c03	2019-06-19 15:05:10 +03:00
Leonid Yuriev	ce0e5d67f5	mdbx-tools: avoid output NaN from mdbx_chk for empty tables. Change-Id: Ie1ff87da3a5e5e124eac1dafd7d5b456f8bde6e3	2019-06-10 13:28:31 +03:00
Leonid Yuriev	48655b41fb	mdbx-ci: switch to Xenial. Change-Id: Ibb4e397d1d405add92c6252fd31080197efeb9a2	2019-05-28 21:23:49 +03:00
Leonid Yuriev	b443477869	mdbx: workaround for Coverity Scan. Change-Id: I0e2d22bbbd38ac7978fb8879219ded79a5be1b0a	2019-05-28 21:19:54 +03:00
Leonid Yuriev	870c2a6f9c	mdbx: symmetrical/invariant mdbx_estimate_range() for MDBX_EPSILON order. Change-Id: Ida7e07d6429576c457bcd4d877a3c38c88dc2771	2019-05-26 00:59:35 +03:00
Leonid Yuriev	e26b7501eb	mdbx: add MDBX_EPSILON support for mdbx_estimate_range(). Change-Id: I2d89a9f20bfa16c8f35a4381709bc54f86f0ff67	2019-05-25 19:10:38 +03:00
Leonid Yuriev	b4002a8484	libmdbx: fix TAGRET typo (minor). Change-Id: Iffafbed7fdad3492aeb51f17caf8109a5b3e35c0	2019-05-02 16:46:05 +03:00
Leonid Yuriev	6e3725457d	mdbx: fix minor comment typo. Change-Id: I56a465e820a49d13c49fb3bd05add970b0eebb14	2019-03-14 00:13:13 +03:00
Leo Yuriev	73f8839a97	mdbx: minor refine/clarify estimation internals.	2019-03-06 16:45:49 +03:00
Leo Yuriev	501eb8c6ad	mdbx: more __hot/__cold attributes for functions.	2019-03-06 16:45:01 +03:00
Leonid Yuriev	7f8cd66e11	mdbx: add notes about range query estimation into READMEs. Change-Id: Ia9e0b7e393082115839483ea7a3b37fb37ba0308	2019-03-06 00:40:53 +03:00
Leo Yuriev	ee899a21ed	mdbx: treat pagesize == 0/INTPTR_MAX as aliases for MIN_PAGESIZE/MAX_PAGESIZE.	2019-03-05 17:55:33 +03:00
Leo Yuriev	3535e7a6d6	mdbx: returns as-is (i.e. negative) the estimation results for an inverted ranges.	2019-03-05 15:50:45 +03:00
Leonid Yuriev	8ddfd1f34a	mdbx: adds functions for distance/move/range estimation (initial). Change-Id: If59eccf7311123ab6384c4b93f9b1fed5a0a10d1	2019-03-05 02:57:15 +03:00
Leo Yuriev	7d383350e8	mdbx: workaround for musl-libc __assert_fail() prototype bug.	2019-03-04 14:41:50 +03:00
Leo Yuriev	9ffd17d58b	mdbx: refine mdbx_filesync() to avoid hide fdatasync() error.	2019-03-04 13:53:05 +03:00
Leo Yuriev	9f410597df	mdbx: checking only _POSIX_SYNCHRONIZED_IO for fdatasync (musl).	2019-03-04 13:39:33 +03:00
Leo Yuriev	cca2c91058	mdbx: don't check __GLIBC_PREREQ/_BSD_SOURCE/_XOPEN_SOURCE for use fsync (musl).	2019-03-04 13:34:38 +03:00
Leo Yuriev	46b551e386	mdbx: add in-source definition for _POSIX_C_SOURCE and _XOPEN_SOURCE (musl).	2019-03-04 13:33:27 +03:00
Leonid Yuriev	14ae9fb2a1	mdbx: env_set_geometry() treat zero-values also as defaults. Change-Id: If8c6f7d7bbeffe71ae4e28f27184103dd1da257b	2019-03-02 14:26:09 +03:00
Leonid Yuriev	01797cf1bc	mdbx: env_open() consider zero mode_t as open-existing flag. Change-Id: I6f9dbf2059822afaba4c3de8f4ce380613a7dc36	2019-03-02 13:31:08 +03:00
Leonid Yuriev	7617cce0c6	mdbx-tools: fix Coverity warning (paranoia). Change-Id: I3ff33a9eb2c58fe601566fd4101f9c95d76d29de	2019-02-27 23:26:15 +03:00
Leo Yuriev	ac6d423451	mdbx: fix one more comment typo (minor).	2019-02-13 20:23:43 +03:00
Leo Yuriev	44a067283a	mdbx: fix comment typo (minor).	2019-02-12 13:39:16 +03:00
Leonid Yuriev	49fa9b9c35	mdbx: fix handling MDBX_APPENDDUP mode. Change-Id: I36de2a8dcab5126dab3857a7840ab3904a1d19c8	2019-02-04 01:41:11 +03:00
Leonid Yuriev	0639f54280	mdbx-test: add 'append' testcase. Change-Id: I71620ea1a019e16b8e3d84a81dcc042961eae5b5	2019-02-04 01:41:11 +03:00
Leonid Yuriev	73bef80347	mdbx-check: add checking for complete duplicates. Change-Id: I8308b725418ef69188eeadfc656dead4ce9cee27	2019-02-04 01:32:15 +03:00
Leonid Yuriev	460751bc01	mdbx-chk: add '-i' option for custom comparators. Change-Id: Ie3d7fdb3c3a881a484d351ca9a3160eb467b43b9	2019-02-04 01:32:15 +03:00
Leo Yuriev	309955be75	mdbx-load: add '-a' option for loading dumps of custom-sorted DBs. Based on http://www.openldap.org/devel/gitweb.cgi?p=openldap.git;a=commitdiff;h=aa77c832b8e6fc696078017f550d119cdfc0f232 Change-Id: If7de71c8f6ffc29d4316c6074995fab38f2c1b4b +load Change-Id: Iff6cbca2514840ee290f801e3b273edf160913b4	2019-02-04 00:06:39 +03:00
Leo Yuriev	9ba8434c1d	mdbx: fix __ANDROID__ typo. Thank to Howard Chu <hyc@openldap.org>. Change-Id: Ibcbe2e4790a5df5758d9fd6c621793ea42a94682	2019-02-03 17:53:00 +03:00
Howard Chu	bfffaa66b8	mdbx: import - tweak mdb_page_split (ITS#8969). Bump up number of keys for which we use fine-grained splitpoint search Change-Id: Icca2e1953cbcd6898b790f657636c2195b397790	2019-02-03 13:07:17 +03:00