Region is used for temporary allocations, usually bound to a
transaction, but not always. Keyword - temporary. It is usually
much faster to allocate something on the region than on the heap,
and much much faster to free the region, since it frees data in
whole slabs. Lots of objects at once.

Region is used both for byte arrays (MessagePack, strings, blobs),
and for objects like standard types, structs. Almost always for
allocations was used region_alloc() function, which returns a not
aligned address. It can be anything, even not multiple of 2.

That led to alignment violation for standard types and structs.
Usually it is harmless in terms of errors, but can be slower than
aligned access, and on some platforms may even crash. Also the
crash can be forced using clang undefined behaviour sanitizer,
which revealed all the not aligned accesses fixed in this patch.

Part of #4609

Vdbe at runtime allocates VdbeCursor structure using
allocateCursor() function. Inside there is a pointer at BtCursor
structure. To make the allocation faster and improve cache
locality, both cursors are allocated in one memory block + some
extra memory for uint32_t array, where BtCursor followed
VdbeCursor and the array without any padding:

   VdbeCursor + uint32_t * N + BtCursor

The problem is that BtCursor needs 8 byte alignment. When it
followed VdbeCursor (aligned by 8) + some uint32_t values, its
actual alignment could become 4 bytes. That led to a crash when
alignment sanitizer is enabled in clang.

The patch makes BtCursor offset aligned by 8 bytes.

Part of #4609

There is some assembly working with a byte array like with an
array of unsigned long values. That is incorrect, because the byte
array may be not aligned by 'unsigned long'.

The patch makes crc32 calculate the hash on the prefix
byte-by-byte, then word-by-word on aligned addresses, and again
byte-by-byte on a tail which is less than word.

Assuming that the word-by-word part is the longest, this should
reduce number of memory/cache loads in ~x2 times. Because in case
of a not aligned word load it was necessary to load 2 words, and
then merge them into one. When addresses are aligned, this is only
1 load.

Part of #4609

Clang has a built-in sanitizer for undefined behaviour. Such as
wrong memory alignment, array boundaries violation, 0 division,
bool values with non standard content, etc.

The sanitizer emits runtime checks which lead to either crash, or
a trap, or a warning print, depending on what is chosen.

The patch makes it possible to turn the sanitizer on and catch
UBs. The only supported UB so far is alignment check. Other types
can be added gradually, along with fixing bugs which they find.

The UB sanitizer is activated for ASAN builds in CI.

Part of #4609

Warning about invalid offsetof() (used on non-POD types) was set
for g++, but wasn't for clang++.

Warning about invalid alignof() (when expression is passed to it
instead of a type) wasn't ignored, but is going to be very
useful in upcoming unaligned memory access patches. That allows
to write something like:

    struct some_long_type *object = region_aligned_alloc(
            region, size, alignof(*object));

This will work even if type of 'object' will change in future,
and so it is safer. And shorter.

Part of #4609

Rlist used a hack to implement offsetof() leading to crash under
undefined behaviour clang sanitizer. It was fixed in this update.

Additionally, region_alloc_object() is changed to return the used
size and a new macro region_alloc_array() is added. This small
API change is supposed to simplify switching lots of region
allocations to aligned versions in scope of #4609.

And finally - there was a bug in lsregion, when allocation was
exactly of slab size. It is fixed in small's master.

Part of #4609

GCC 10 produces the following error:
cc1: warning: function may return address of local variable [-Wreturn-local-addr]

Fix it.

Part-of #4966

Add tests for setting up logging via log module.

Part-of #689

Signed-off-by: Cyrill Gorcunov <gorcunov@gmail.com>

In the patch we introduce an ability to configure logging
engine early without calling box.cfg{} at all. For this sake
we have to modify the load_cfg.lua (main cfg loader) together
with log.lua built in module.

In particular the configuration entries which are supposed to
be handled by modules are grouped in module_cfg table, where we
list box.cfg{} keys and methods to fetch/set the values. Same time
the template_cfg{} starts to carry 'module' keyword telling
prepare_cfg routine to use modules.

@TarantoolBot documnent
Title: Module log

To configure log module eary without initializing box module
at all call the `log.cfg({})` method, where the following
arguments are acceptable:

 - `log` to specify file, pipe or syslog (same as
   `box.cfg{log}` option); the default value is nil
   and stderr is used;

 - `level` to specify logging level; can be numbers (1-7) or
   strings (fatal,syserror,error,crit,warn,info,verbose,debug);
   the default value is 5 (i.e. 'info');

   While for symbolic representations of levels we allow only
   predefined keywords to be used, to keep backward compatibility
   the numbers can be any;

 - `format` to specify output encoding ('plain' or 'json');
   the default value is 'plain';

 - `nonblock` to open logging output in nonblocking mode
   (`true` or `false`); the default value is `false`.

Fixes #689

Signed-off-by: Cyrill Gorcunov <gorcunov@gmail.com>

In upgrade testing we need an ability to control running of upgrade.lua
script execution. When constant ERRINJ_AUTO_UPGRADE is set to true
tarantool will skip upgrade.lua script execution.

Part of #4801

During investigation of the issues found that error message had
not user friendly format:
 local pwgr_errstr = "get%s failed [errno %d]: %s"
produced:
 [001] LuajitError: builtin/pwd.lua:101: getgetgrall failed [errno 2]: No such file or directory
while it had to be:
 [001] LuajitError: builtin/pwd.lua:101: getgrall failed [errno 2]: No such file or directory
Fixed the error message.

During creating patch:

  e3d9d8c9 ("build: add CentOS 8 into CI / CD")

was found the issue:

  [001] app-tap/pwd.test.lua                                            [ fail ]
  [001] Test failed! Output from reject file app-tap/pwd.reject:
  [001] TAP version 13
  [001] 1..6
  [001] ok - checking user by id
  [001] ok - checking user by name
  [001] ok - checking group by id
  [001] ok - checking group by name
  [001]
  [001] Last 15 lines of Tarantool Log file [Instance "app_server"][/mnt/test/var/001_app-tap/pwd.test.lua.tarantool.log]:
  [001] tarantool: /lib64/libcurl.so.4: no version information available (required by tarantool)
  [001] builtin/pwd.lua:169: getpwall failed [errno 2]: No such file or directory

Currently it was tried to reproduce the issue, but all the attempts
didn't got the needed fail. After discussion of the issue decided to
close the current one as not the issue for the test and continue the
fixes for getpwall/getgrall functions within issue #5034.

Closes #4592

For some reason we've missed that say_set_log_format
doesn't support json format not only for syslog but
for boottime logging as well.

Part-of #689

Reviewed-by: Oleg Babin <olegrok@tarantool.org>
Signed-off-by: Cyrill Gorcunov <gorcunov@gmail.com>

This allows us to reuse them instead of copying
opencoded constants. They are highly bound to
ones compiled into the C code.

Part-of #689

Reviewed-by: Oleg Babin <olegrok@tarantool.org>
Reviewed-by: Leonid Vasiliev <lvasiliev@tarantool.org>
Signed-off-by: Cyrill Gorcunov <gorcunov@gmail.com>

We gonna use it to provide a way to initialize
logger subsystem early.

Part-of #689

Reviewed-by: Oleg Babin <olegrok@tarantool.org>
Reviewed-by: Leonid Vasiliev <lvasiliev@tarantool.org>
Signed-off-by: Cyrill Gorcunov <gorcunov@gmail.com>

To eliminate opencoded comparision.

Reviewed-by: Leonid Vasiliev <lvasiliev@tarantool.org>
Reviewed-by: Oleg Babin <olegrok@tarantool.org>
Signed-off-by: Cyrill Gorcunov <gorcunov@gmail.com>

We gonna support logger configuration on its own
without requirement to call `box.cfg{}`. Thus lets
say_logger_init() to skip processing if we already
did.

Note it is preparatory for next patches. Currently
the init called once.

Part-of #689

Reviewed-by: Oleg Babin <olegrok@tarantool.org>
Reviewed-by: Leonid Vasiliev <lvasiliev@tarantool.org>
Signed-off-by: Cyrill Gorcunov <gorcunov@gmail.com>

The tcp server starts in a separate fiber.
When server's socket is closed from another fiber,
an exception will be thrown in server's loop from
check_socket function.
A "socket is close" check has been added at server's
fiber and now server's fiber terminates correctly.

Fixes: #4087

For the CustomError the ability to create a message
using a format string was added.

Closes #4903

@TarantoolBot document
Title: Add format string usage to compose a CustomError message
When an error created using the separate members mode
(box.error(code, errtext[, errtext ...])) in the case of
ClientError error creation, a pre-defined format is used
(corresponding to the error code) (nothing has changed), in
the case of CustomError error creation, a format string can
be used to compose a message.

ClientError(nothing has changed):
```Lua
box.error(code, reason args)
```

Example for ER_CREATE_SPACE:
```Lua
box.error(9, "my_space", "reason")
```
Result:
```Lua
error: 'Failed to create space ''my_space'': reason'
```

CustomError:
```Lua
box.error(type, reason format string, reason args)
```

Example:
```Lua
box.error("MyCustomType", "The error reason: %s", "some error reason")
```
Result:
```Lua
error: 'The error reason: some error reason'
```

Make it possible to set temporary directory where
module will be copied before load.

@TarantoolBot document
Title: Justify module (re-)loading semantics

It is now possible to set directory where temporary
copies of modules to be loaded will be created.
It is done by setting $(TMPDIR) variable. It
will be "/tmp" if variable was not set.

Follow up #4945

Tarantool module reload mechanism allows to reload a module even
if there're functions running. It is implemented by refcounting
each invocation of module's routines.
If reload is called, then refcounter is checked:
 - If it is 0, then no routines are in flight and module is
   reloaded by simple pair of dlclose()/dlopen().
 - If it is non-zero, then there're routines in flight. To allow
   to load multiple versions of modules it is loaded not from the
   DSO specified. Symlink to tempdir is created and dlopen() is
   invoked against it (w/RTLD_LOCAL flag to avoid conflicts).

This trick was implemented in order to fool a dynamic linker: one
cannot invoke dlopen() against same file, so let's pretend
there're to independent DSOs.

The problem is that dynamic linker is smart enough. It tracks not
filenames, but i-nodes. Moreover it is smart enough to do stat -L
against DSO to follow symlinks! [1][2] So, any attempts to create
a symlinks to fool dynamic linker fail and instead of doing actual
load it just increments internal refcounter in map w/
corresponding inode, which in turn leads to not-reloading.

This wasn't caught by test since old module was always unlinked
before new one is copied in place.

The patch always copies DSO instead of creating a symlink. Also
it fixes the test so in SEGFAULTs without the change.

Closes #4945

[1] - https://sourceware.org/git/?p=glibc.git;a=blob;f=elf/dl-load\
.c;h=a5318f9c8d1d42745a254479cf6bb1cd2acd516f;hb=58557c229319a3b8d\
2eefdb62e7df95089eabe37#l898

[2] - https://sourceware.org/git/?p=glibc.git;a=blob;f=sysdeps/pos\
ix/dl-fileid.h;hb=58557c229319a3b8d2eefdb62e7df95089eabe37#l33

OSX testing completely implemented on local MAC hosts with OSX 14
and 15 versions. So no more need to test and support issues with
testing OSX with 14 version on travis-ci.

Part of #4893

Test modules are shared libraries. They link with Tarantool at
runtime and resolve missing symbols then. However in case they
already contain some of the same symbols used by tarantool, that
leads to a conflict in ASAN build. Nothing serious, there are
strict rules how a linker should behave in this case, but there
is still a warning. This is how it looked (the text is shortened):

    ==25624==ERROR: AddressSanitizer: odr-violation (0x000001123b60):
      [1] size=1024 'mp_type_hint' /builds/nZUxDh2c/0/tarantool/tarantool/src/lib/msgpuck/hints.c:39:20
      [2] size=1024 'mp_type_hint' /builds/nZUxDh2c/0/tarantool/tarantool/src/lib/msgpuck/hints.c:39:20
    These globals were registered at these points:
      [1]:
        #0 0x478b8e in __asan_register_globals (/builds/nZUxDh2c/0/tarantool/tarantool/src/tarantool+0x478b8e)
        #1 0x7ff7a9bc9d0b in asan.module_ctor (/tmp/tntz2FLhA/function1.so+0x6d0b)

      [2]:
        #0 0x478b8e in __asan_register_globals (/builds/nZUxDh2c/0/tarantool/tarantool/src/tarantool+0x478b8e)
        #1 0xab990b in asan.module_ctor (/builds/nZUxDh2c/0/tarantool/tarantool/src/tarantool+0xab990b)

The symbol mp_type_hint is defined both in tarantool executable,
and in function1.so dynamic library. This is because both build
with msgpuck static library.

However the modules don't need to be built with msgpuck. They can
link with it at runtime from tarantool executable.

This patch removes msgpuck from test dynamic modules, and exports
all msgpuck non-inlined symbols explicitly, so they couldn't be
removed by the linker in future.

Follow-up #2971
Follow-up #5001
Closes #5023

- Adding 2 servers (sh1, sh2) for performance testing.
- Rebalance performance testing between servers:
  - sh1 (tag: docker_sh1_perf) used for benches:
    nosqlbench_hash, nosqlbench_tree
  - sh2 (tag: docker_sh2_perf) used for benches:
    ycsb_hash, ycsb_tree, cbench
  - sh3 (tag: docker_sh3_perf) used for benches:
    sysbench, tpcc
  - sh9 (tag: docker_perf_ssd) used for benches:
    linkbench_ssd
- Changing gitlab-ci tag for performance docker images
  from 'perf' to 'deploy'.

Closes #4868

Each temporary docker perf image is about 5 GB
and we don’t have space to keep them localy on servers.

- add cleanup after perf docker
  image preparation (perf_bootstrap job).
- cleanup perf docker images on execution hosts
  after perf testing (cleanup stage).

Closes #5003

Set hard-coded unix sockets for iproto connections at core = app.
Added its option in *-tap/suites.ini files and fixed test for it.
Fix helped to handle the problem with 'Address already in use' error.
Check the previous commit that set the use of sockets:

60f84cbf ('test: use unix sockets for iproto connections')

Test sql-tap/gh-4077-iproto-execute-no-bind.test.lua removed from
fragiled list.

Closes #4459

The function returns a number of bytes needed to store
an fk_constraint_def object with its name and links.
However it used sizeof(struct fk_constraint) instead
of sizeof(struct fk_constraint_def) to calculate
base object size. This worked only because
fk_constraint is bigger than fk_constraint_def.

Part-of #689

Signed-off-by: Cyrill Gorcunov <gorcunov@gmail.com>

The implementation has been removed in
commit ef2c171d

Part-of #689

Signed-off-by: Cyrill Gorcunov <gorcunov@gmail.com>

Vclock depends on bit library, but wasn't linked to it.

Before this patch box.snapshot() bails out immediately if it sees
that the scheduler is throttled due to errors. For instance:

box.error.injection.set('ERRINJ_VY_RUN_WRITE', true)
snapshot() -- fails due to ERRINJ_VY_RUN_WRITE
box.error.injection.set('ERRINJ_VY_RUN_WRITE', false)
snapshot() -- still fails despite the fact that injected error is unset

As a result, one has to wait up to a minute to make a snapshot. The
reason why throttling was introduced was to avoid flooding the log
in case of repeating disk errors.
What is more, to deal with schedule throttling in tests, we had to
introduce a new error injection (ERRINJ_VY_SCHED_TIMEOUT). It reduces
time duration during which the scheduler remains throttled, which is
ugly and race prone.  So, let's unthrottle scheduler when checkpoint
process is launched via manual box.snapshot() invocation.

Closes #3519

In some cases it may turn out to be useful to know whether checkpoint
process was launched manually (explicitly calling box.snapshot()) or
automatically via checkpoint daemon. In particular, to unthrottle vinyl
scheduler when it comes for manual checkpoints. So let's extend engine's
vtab method begin_checkpoint() with corresponding argument.

Needed for #3519

Assert in vy_tx_write() validates that upsert applied on the top of
cached statement is always replace. In fact, it is not always so. If
vy_apply_upsert() fails due to the fact that PK is modified, it returns
old statement (i.e. statement which upsert is applied on). In this
regard, if tuple cache contains insert and invalid upsert is executed,
vy_apply_upsert() will return insert. As a result, assert will fire.
Let's fix it and take into account that vy_apply_upsert() is able to
return inserts as well.

Closes #5005

vinyl_engine_create_space() didn't set an error in the
diagnostics area when region_alloc() failed. This
could lead to a crash, although was almost impossible
to happen.

There were lots of errors of kind:

/builds/M4RrgQZ3/0/tarantool/tarantool/src/exports.h:395:1: error: variable ‘uuid_nil’ redeclared as function
  EXPORT(uuid_nil)
  ^
 /builds/M4RrgQZ3/0/tarantool/tarantool/src/lib/uuid/tt_uuid.c:39:22: note: previously declared here
  const struct tt_uuid uuid_nil;

when LTO was enabled. That happened because exports.c file, to
take symbol addresses, declared lots of functions and variables
from all the code base as

    extern void <symbol>(void);

This is crazy, but it worked for normal builds. Because symbol is
symbol. The compilers couldn't find conflicting declarations,
because they never met in one compilation unit.

However the lie was revealed by linker with LTO enabled. It could
see, that actual symbol definitions didn't match their exports in
exports.c. It could live with mismatching function-function or
variable-variable cases, but couldn't withstand function-variable
mismatches. When a symbol was declared as a variable in one place
and as a function in another.

This was the case for variables:
- uuid_nil
- tarantool_lua_ibuf
- log_pid
- log_format
- crc32_calc
- _say
- log_level

The errors were false positive, because the symbols were never
used for anything except taking their addresses. To calm the
linker down exports.c now does not participate in LTO.

Closes #5001

Due to inaccuracy FreeBSD skip conditions were overwritten by OpenBSD
skip conditions.

Closes #5004

Functions tt_uuid_is_nil() and tt_uuid_is_equal() used struct
tt_uuid as an array of two uint64. But it is wrong, because struct
tt_uuid has 4 byte alignment, while uint64 requires 8. That led to
unaligned memory access in these 2 functions.

The patch makes these functions use memcmp() - it appers to be
faster (disassembly shows much smaller code and without jumps),
and does not depend on alignment.

Part of #4609

Reviewed-by: Aleksandr Lyapunov <alyapunov@tarantool.org>
Reviewed-by: Timur Safin <tsafin@tarantool.org>

Lib/bit library in some functions dereferenced a byte array as a
pointer at uint64 or uint32. It led to unaligned memory access,
when the array didn't have the needed alignment. That made clang
'alignment' sanitizer crazy.

Also there were places, where bitwise shift was done for the type
size. For example, a uint32 number was allowed to make 32 bit
shift in both sides. This is undefined behaviour.

The patch fixes these UBs. Bitwise shift of too long size is
banned. Unaligned access is fixed in some places by changing
uint64 to uint8, when it is possible without perf loss. And in the
others by using explicitly specified unaligned load. So it works
the same, but the sanitizer is aware of that.

Part of #4609

Reviewed-by: Aleksandr Lyapunov <alyapunov@tarantool.org>
Reviewed-by: Timur Safin <tsafin@tarantool.org>

Msgpuck functions mp_snprint() and mp_fprint() now support
customizable MP_EXT serializer. This patch makes them able to
correctly serialize all tarantool's MP_EXT extensions: MP_DECIMAL,
MP_UUID, MP_ERROR.

Closes #4719

Reviewed-by: Cyrill Gorcunov <gorcunov@gmail.com>
Reviewed-by: Serge Petrenko <sergepetrenko@tarantool.org>
Reviewed-by: Nikita Pettik <korablev@tarantool.org>
Reviewed-by: Alexander Turenko <alexander.turenko@tarantool.org>

Msgpuck functions mp_snprint() and mp_fprint() now support
customizable MP_EXT serializer. This patch adds such for MP_ERROR.
All extension serializers will be activated in a separate commit.

Part of #4719

Reviewed-by: Cyrill Gorcunov <gorcunov@gmail.com>
Reviewed-by: Serge Petrenko <sergepetrenko@tarantool.org>
Reviewed-by: Nikita Pettik <korablev@tarantool.org>
Reviewed-by: Alexander Turenko <alexander.turenko@tarantool.org>