- 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>

Msgpuck functions mp_snprint() and mp_fprint() now support
customizable MP_EXT serializer. This patch adds such for
MP_UUID. It is not activated yet, but will be later, when last
extension - MP_ERROR - also gets its serializer.

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>

Msgpuck functions mp_snprint() and mp_fprint() now support
customizable MP_EXT serializer. This patch adds such for
MP_DECIMAL. It is not activated yet, but will be later, when other
extensions also get their serializers.

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>

In the new version an API is added to be able to customize MP_EXT
serializer for mp_snprint() and mp_fprint() functions.

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>

It is hotfix for the previous test-run update, where 'core = app' test
suites starts to support 'use_unix_sockets_iproto' option. We found a
case when unix socket names may clash and attempt to listen on the
socket will report 'Address already in use' error. It is fixed in this
test-run update.

See more in [1].

[1]: https://github.com/tarantool/test-run/pull/211

Support 'use_unix_sockets_iproto' suite.ini option in 'core = app' test
suites. It'll allow to workaround TCP port choosing race (see more in
[1]) by using Unix sockets instead of TCP ones: we did the same for
'core = tarantool' test suites in [2].

[1]: https://github.com/tarantool/test-run/issues/141
[2]: 60f84cbf ('test: use unix sockets
     for iproto connections')

Part of #4459.

vy_read_iterator_cmp_stmt() uses tuple_compare_with_key() under the hood
which follows next convention:
(a) it returns 0 if key_fields of tuple equal to given key;
(b) < 0 if key_fields less than given key;
(c) > 0 otherwise.
However, comment to vy_read_iterator_cmp_stmt() says that:
(a) it returns -1 if first statement precedes second one;
(b) 0 if they at the same position;
(c) +1 if first statement supersedes second one.
so, since there's no "normalization" of return code from
tuple_compare_with_key() it would be correct to say that
vy_read_iterator_cmp_stmt() returns:
(a) it returns arbitrary integer value < 0 if first statement precedes
second one;
(b) 0 if they at the same position;
(c) arbitrary integer value > 0 if first statement supersedes second one.

Add MAP_STACK to mmap() flags.

Changed number of tests to packaging testing jobs from
suites: "unit/ app/ app-tap/ box/ box-tap/ engine/ vinyl/",
to all suites except 'replication/' suite which has special
issue for it's enabling:
  https://github.com/tarantool/tarantool/issues/4798
Also added testing to the all available packing jobs except
'CentOS 6'.

Removed test-run option that run testing inline to be
able to run it in parallel. Also changed 'test-run' to
'make test-force' command.

Closes #4599

Found that issues #1227 and #1322 were closed, returned
the tests blocked by it into the testing.

Part of #4599

Tarantool packages build scenario at rpm/tarantool.spec
installs Python packages for Tarantool testing. In some
latest OS like Fedora 32 old Python packages based on
Python 2 version were deprecated and changed to Python 3.
Currently Tarantool testing based on Python 2 which
packages stored at local backport repository and it
should be used from there calling it as python2-* packages.

Issue:
...
No matching package to install: 'python-yaml >= 3.0.9'
...
Not all dependencies satisfied
Error: Some packages could not be found.

Needed for #4599
Closes #4965

Found that modules may have only sources packages w/o binaries
packages. Script updated to be able to work with only binaries
either sources RPM packages. The same fix was already done for
DEB packages at commit 4527a4da.

Close #4839

Fragiled flaky tests from parallel runs to avoid
of flaky fails in regular testing:

  app/fiber.test.lua                        ; gh-4987
  app/fiber_channel.test.lua                ; gh-4961
  app/socket.test.lua                       ; gh-4978
  box/alter_limits.test.lua                 ; gh-4926
  box/misc.test.lua                         ; gh-4982
  box/role.test.lua                         ; gh-4998
  box/rtree_rect.test.lua                   ; gh-4994
  box/sequence.test.lua                     ; gh-4996
  box/tuple.test.lua                        ; gh-4988
  engine/ddl.test.lua                       ; gh-4353
  replication/box_set_replication_stress    ; gh-4992
  replication/recover_missing_xlog.test.lua ; gh-4989
  replication/replica_rejoin.test.lua       ; gh-4985
  replication/wal_rw_stress.test.lua        ; gh-4977
  replication-py/conflict.test.py           ; gh-4980
  vinyl/errinj_ddl.test.lua                 ; gh-4993
  vinyl/misc.test.lua                       ; gh-4979
  vinyl/snapshot.test.lua                   ; gh-4984
  vinyl/write_iterator.test.lua             ; gh-4572
  xlog/panic_on_broken_lsn.test.lua         ; gh-4991

Part of #4953

Part of #4967

Part of #4967

Part of #4967

- define macros LOG_MAKEPRI() on OpenBSD as it is absent
- replace sigtimedwait() by sigwait() as latter is unsupported on OpenBSD

Part of #4967

Part of #4967

Part of #4967

Part of #4967

OpenBSD includes DL library in a base system

Part of #4967

Part of #4967

Part of #4967

killpg() on Mac OS may don't deliver a signal to a process: it seems
that there is a race when a process is just forked. It means that
popen_handle:close() may leave a process alive, when `opts.setsid` and
`opts.group_signal` are set.

There is simple reproducer, which does not leave alive `sleep 120`
processes on Linux, but does it on Mac OS (within three-four runs in
rows):

 | #include <signal.h>
 | #include <unistd.h>
 | #include <fcntl.h>
 |
 | int
 | main()
 | {
 | 	char *child_argv[] = {
 | 		"/bin/sh",
 | 		"-c",
 | 		"sleep 120",
 | 		NULL,
 | 	};
 | 	pid_t pid;
 | 	int fd[2];
 | 	pipe(fd);
 | 	fcntl(fd[0], F_SETFD, FD_CLOEXEC);
 | 	fcntl(fd[1], F_SETFD, FD_CLOEXEC);
 |
 | 	if ((pid = fork()) == 0) {
 | 		/* Child. */
 | 		close(fd[0]);
 | 		setpgrp();
 | 		for (int i = 0; i < 10; ++i) {
 | 			/* Proceed with killpg. */
 | 			if (i == 5)
 | 				close(fd[1]);
 | 			if (fork() == 0) {
 | 				/* Child. */
 | 				execve("/bin/sh", child_argv, NULL);
 | 			}
 | 		}
 | 	} else {
 | 		/* Parent. */
 | 		close(fd[1]);
 | 		char c;
 | 		read(fd[0], &c, 1);
 | 		killpg(pid, SIGKILL);
 | 	}
 | 	return 0;
 | }

Compile it (`cc test.c -o test`) and run several times:

$ for i in $(seq 1 1000); do                     \
    echo $i;                                     \
    ./test                                       \
    && ps -o pid,pgid,command -ax | grep [s]leep \
    && break;                                    \
done

This is the reason why `sleep 120` process may be alive even when the
whole test passes.

test-run captures stdout and stderr of a 'core = app' test and waits EOF
on them. If a child process inherit one of them or both, the fd is still
open for writing and so EOF situation will not appear until `sleep 120`
will exit.

This commit doesn't try to overcome the root of the problem, but close
stdout and stderr for the child process that may not be killed / exited
in time.

Aside of this, updated found Mac OS peculiars in API comments of C and
Lua popen modules.

Fixes #4938

@TarantoolBot document
Title: popen: add note re group signaling on Mac OS

Copyed from the popen_handle:signal() updated description:

> Note: Mac OS may don't deliver a signal to a process in a group when
> opts.setsid and opts.group_signal are set. It seems there is a race
> here: when a process is just forked it may be not signaled.

Copyed from the popen_handle:close() updated description:

> Details about signaling:
>
> <...>
> - There are peculiars in group signaling on Mac OS,
>   @see popen_handle:signal() for details.

Follows up https://github.com/tarantool/doc/issues/1248