There are modules that are implemented as two parts: a Lua/C module for
internal use and a public module written on Lua. There is a practice to
name both parts the same: just capture the internal part within the
public part and rewrite `package.loaded` then.

This name overlap is confusing at reading the sources and complicates
debugging. And it conflicts with a built-in module loading logic that
will be implemented for #7774.

Let's use `foo.lib` for the internal part and `foo` for the public one.
This approach is already used in some built-in modules.

src/box/lua/upgrade.lua requires src/box/lua/xlog.lua, so changed the
loading order.

Eliminated extra `internal` field in `uri.lib`, because the whole module
is internal.

Part of #7774

NO_DOC=user visible behavior is unchanged, pure refactoring change
NO_TEST=see NO_DOC
NO_CHANGELOG=see NO_DOC

The idea of the refactoring is described in a previous commit.

The change is straightforward:

* Use `luaL_findtable()` to find/create a nested table in `_G`.
* Use `luaL_setfuncs()` to assign functions to a table.
* Use `luaT_newmodule()` to register a module for `require()`.

`require('box.internal.<...>')` calls are replaced with
`box.internal.<...>`, because `box` submodules are not registered as
modules anymore (IOW, are not placed into `package.loaded`).

However `_G.box == require('box')`, it remains unchanged.

One may wonder why fourth argument of `luaL_findtable()` (size hint for
the hash part of the table) is always zero in the added calls. It is an
effect of the following points:

* Those calls are not on hot paths and, moreover, there is no sense to
  use `luaL_findtable()` with a string literal as the second argument on
  a hot path (no sense to parse the string literal in runtime if a
  performance is the goal).
* Using `lengthof()` macro would add more code to read for a human (the
  argument itself, a line carry to fit 80 symbols, an inclusion of
  trivia/util.h) at no gain.
* A hardcoded array length is a developer's stress factor: it looks like
  it requires an update at a change of the array. So zero is better.

Part of #7774

NO_DOC=user visible behavior is unchanged, pure refactoring change
NO_TEST=see NO_DOC
NO_CHANGELOG=see NO_DOC

See the previous commit regarding the idea behind this refactoring.

The src/box/lua part is updated in a next commit, only `src/lua` is
updated here. It is split to ease reading.

The patch is mostly straightforward, I'll highlight several interesting
places.

* `compat`

  `_G.internal.compat` was created by `luaL_register()`. Removing of
  this call drops `_G.internal`: it is not created anymore.

* `fiber.channel` and `fiber.cond`

  The initialization code add fields into fiber module. So we call
  `require('fiber')` here.

  The alternative is to access `package.loaded.fiber` directly, however
  I plan to eliminate such accesses in further patches. (Built-in
  modules registration process will use its own table instead of
  `package.loaded`.)

* `http.client`

  Here we also eliminate a direct access to `package.loaded`, which is
  important for implementing #7774.

  Dropping of the `package.loaded['http.client']` field in unnecessary,
  because it is anyway replaced by the resulting module table in
  `tarantool_lua_init()`.

* `utf8`

  This module is also set to `_G.utf8` for compatibility with Lua 5.3.
  It remains unchanged.

Part of #7774

NO_DOC=user visible behavior is unchanged, pure refactoring change
NO_TEST=see NO_DOC
NO_CHANGELOG=see NO_DOC

This commit starts a series, which refactors built-in module
registration mechanisms. The series aims several goals:

* Offer a standard tarantool specific way to register a built-in module.
  It allows to change the registration process for all built-in modules
  at once.
* Make the API of such functions simple and clean.
* Eliminate known problems of the existing approach to register built-in
  modules.

This particular commit offers the new `luaT_newmodule()` function. It
replaces `luaL_register()` and `luaL_register_module()` to some extent.

Here I also start a practice to use `luaL_setfuncs()` instead of
`luaL_register(L, NULL, funcs)` for readability purposes. The function
is part of Lua 5.2 ([documentation][1]), but it is implemented in
LuaJIT.

Let's look on examples below.

## How it works now

First of all, consider the [documentation][2] about the
`luaL_register()` function. Then look on the examples.

```c
luaL_register(L, "foo.bar", funcs);
```

Creates a table `package.loaded['foo.bar']` and `_G.foo.bar` (if neither
of them exists), fill it with the functions and pushes it to the stack.

What is not ok:

* Pollutes `_G`.
* `package.loaded` is written flat, but `_G` is written deeply.
* No control over (un)intended module rewritting.

It also worth to note that we can't change this function, because it is
part of the Lua API. So we need another function for our registration
process in order to implement #7774.

Another usage of the function makes it even more confusing:

```c
luaL_register(L, NULL, funcs);
```

Does it create a table, fill it with the functions and push it to the
stack? No.

Unlike other usages, it fills a table on the top of the stack with the
functions and leaves it on the stack. And it actually has no relation to
module registration.

There is tarantool specific function `luaL_register_module()`, which
looks as an attempt to solve some of the problems.

```c
luaL_register_module(L, "foo.bar", funcs);
```

It doesn't touch `_G` and, surprisingly, changes how a module is written
to `package.loaded`.

The call creates a table (or finds it) in `package.loaded.foo.bar`,
fills it with the functions and pushes to the stack. *This* function
writes `package.loaded` deeply.

It leaves us with `luaL_register()` for assigning `_G.foo.bar` (with
pollution of `package.loaded`). It leaves us with `luaL_register()` for
setting functions into a table on the stack (with confusing name in this
context and confusing contract). And we has no a function to just assign
`package.loaded[modname]` without the deep tables creation and with
appropriate checks.

## How it will work

```c
luaT_newmodule(L, "foo.bar", funcs);
```

Create a table, set it to `package.loaded['foo.bar']` and push into the
stack. Assert that the table doesn't exist.

```c
luaL_setfuncs(L, funcs, 0);
```

Add functions into a table on top of the stack.

```c
luaL_findtable(L, LUA_GLOBALSINDEX, "foo.bar", 0);
luaL_setfuncs(L, funcs, 0);
```

Find/create a table `_G.foo.bar` and add functions to it.

Next commits will adopt built-in modules to follow this practice.

[1]: https://www.lua.org/manual/5.2/manual.html#luaL_setfuncs
[2]: https://www.lua.org/manual/5.1/manual.html#luaL_register

Part of #7774

NO_DOC=user visible behavior is unchanged, pure refactoring change
NO_TEST=see NO_DOC
NO_CHANGELOG=see NO_DOC

Generate a test build key when the ALPINE_BUILD_KEY secret is empty.
It is needed for fork PRs where secrets are unavailable.

NO_DOC=ci
NO_TEST=ci
NO_CHANGELOG=ci

Since applied transactions are committed asynchronously, they are
inherently 'read committed', hence their isolation level should be adjusted
accordingly.

Closes #8121

NO_DOC=bugfix

XLOG_READ_AHEAD defines the number of bytes added to the `count' argument
of the `pread' syscall, currently it equals to 16 KB. However xlog and snap
files are written by 128 KB chunks of data, which turn into ~80 KB chunks
after compression (in average, depending on the data), so the 16 KB read-
ahead doesn't make any sense.

According to performance experiments, 8 MB readahead gives the best results
for large files. However, 8 MB read buffers would increase memory
consumption of the replication relays, which usually read small portions of
data and does not need such a big buffers. For this reason, dynamically-
sized read buffer is implemented by this patch. The minimal readahead is
now 128 KB, and the maximal is 8 MB. As a result, the recovery time of a
900 MB snapshot decreased from 25 sec to 13 sec (when reading from HDD).
Performance of a recovery from SSD is not affected.

Closes #8108

NO_DOC=performance improvement
NO_TEST=performance improvement

Recently, ffi select was broken on M1 in commit ec1a71ff
("box: introduce pagination to memtx_tree and tuple position methods").
It turned out that ffi on M1 poorly supports a big quantity of arguments.
Fortunately, there is a workaround - we can pass only 64-bit integer
arguments beyond the 8th argument. Let's do it.

Closes #7946

NO_TEST=reflected in existing tests
NO_CHANGELOG=bugfix for unreleased feature
NO_DOC=bugfix

The transaction isolation level used by Tarantool by default is
'best-effort' so adding support of transaction isolation levels to Vinyl
effectively switched the isolation level used by Vinyl transactions from
'read-committed' to 'best-effort', which potentially may cause more
conflicts. Let's mention it in the changelog and mark this change as
breaking.

Follow-up commit 588170a7 ("vinyl: implement transaction isolation
levels")
Follow-up #5522

NO_DOC=changelog
NO_TEST=changelog

Changed the work with hooks in the part of tests.

Legacy approach:

```
g.before_all = function() ... end
g.after_all = function() ... end
```

The logic of executing hooks will be expected and
more controlled if use following approach:

```
g.before_all(function() ... end)
g.after_all(function() ... end)
```

Resolve #8066

NO_DOC=test fix
NO_TEST=test fix
NO_CHANGELOG=test fix

This reverts commit 38f88795.

That checking is too tight and may break the code, that is successfully
working on the non-upgraded schema. Let's postpone the fix to 3.0.

Follow-up #7149

NO_DOC=revert of the unreleased bugfix

The change can only breaks users that were not follow the documented usage
pattern. The documented pattern is to allocate and then truncate you
allocations.

Follow-up #5665

NO_DOC=changelog fix
NO_TEST=changelog fix

Add building Alpine 3.16 packages for amd64 and aarch64.

Part of tarantool/tarantool-qa#266

NO_DOC=ci
NO_TEST=ci

* Update build-time and run-time dependencies.

* Support build with gc64 enabled.

* Disable sysprof due to the following error while building:

NO_WRAP
  ```
  <...>/third_party/luajit/src/lj_sysprof.c:29:10: fatal error: execinfo.h: No such file or directory
    29 | #include <execinfo.h>
       |
  ```
NO_WRAP

* Update .pack.mk to set package version in the case of a git tag to the
  following format:

    2.11.0[_<release type>]

  Where `<release type>` is `alpha1`, `beta1`, `rc1`, etc.
  Note, we have an extra logic when the git tag is `x.x.x-entrypoint`.
  Alpine build system doesn't support package name with such kind of
  version and that's why we use `alpha0` instead of `entrypoint`.
  See for details [1].

* Update .pack.mk to set package version in the other cases to the
  following format:

    2.11.0_<release type>_p<N>

  Where `<release type>` is `alpha1`, `beta1`, `rc1`, etc and `<N>` is
  the patch number after the latest release. Unfortunately, we cannot
  use the similar version pattern that we use for dev DEB/RPM packages
  (`2.11.0_<release type>.<N>.dev`) because Alpine build system doesn't
  support it. Plus, we have the same logic for `entrypoint` here.
  See for details [1].

[1] https://wiki.alpinelinux.org/wiki/APKBUILD_Reference#pkgver

Part of tarantool/tarantool-qa#266

NO_DOC=ci
NO_TEST=ci
NO_CHANGELOG=ci

test_promote_no_quorum testcase in election_pre_vote test started flaky
hanging after commit 5765fdc4 ("raft: fix 'manual' nodes bumping the
term excessively").

Disable it until the issue (#8217) is resolved.

In-scope-of #8217

NO_DOC=tests
NO_CHANGELOG=tests

Remove issues released in 2.10.* from the master's unreleased list.
This change leaves only issues that form the diff between the latest
released version 2.10.4 and the upcoming 2.11-RC (which is the correct
2.11-RC changelog by definition).

NO_CHANGELOG=changelog
NO_DOC=changelog
NO_TEST=changelog

Promotion of a leader in the test requires previous leader state
arrived to all replicas. Inconsistency may lead to promotion failure,
which is correct from the raft point of view, but not expected by the
test.

Closes #7785

NO_DOC=test
NO_CHANGELOG=test

See #4691 for the problem description. In brief: before the fix net.box
expected _vcollation presence if tarantool's runtime on the server is
new enough disregarding a server's schema version. The old schema
version has no this view, so net.box was unable to connect. After #4691
net.box tolerates lack of the view.

The fix was pushed without a test (see [1]) and now I want to pay this
debt.

[1]: https://lists.tarantool.org/pipermail/tarantool-patches/2020-April/016207.html

Follows up #4307
Follows up #4691
Fixes #5482

NO_DOC=just a new test
NO_CHANGELOG=see NO_DOC

It was removed in the commit 390311bb ("fiber: get rid of cpu_misses
in fiber.top()"). Return it back in order to avoid potential breaking
changes.

Follow-up #5869

NO_DOC=tarantool/doc#2989

Remove the "breaking change" mark, as it's highly unlikely that it will
break any log parser.

Follow-up #3211

NO_DOC=changelog
NO_TEST=changelog

Remove the "breaking change" mark, as it's highly unlikely that it will
break anything.

Follow-up #7166

NO_DOC=tarantool/doc#3132
NO_TEST=changelog

Flight recorder uses mmapped file for performance reasons. Unfortunately
if for some reason mapping is not possible we got SIGBUS on accessing
mmapped memory.

We already handle this issue on flight recorder start in [1]. But if we
got SIGBUS in the meanwhile we currently got a crash. Let's just panic.

[1] https://github.com/tarantool/tarantool-ee/commit/e68b84768f3ae76aa9f03d36dd6f0587b884e1bb

CE Part of https://github.com/tarantool/tarantool-ee/issues/198

NO_DOC=stub for EE part
NO_CHANGELOG=stub for EE part
NO_TEST=tested in EE part

Currently crash has code to report crashes to feedback URL. It does
not belong to core. Also it brings dependencies from box to core with
INSTANCE_UUID and REPLICASET_UUID variables. So let's move this part
back to box.

NO_DOC=refactoring
NO_TEST=refactoring
NO_CHANGELOG=refactoring

Currently we keep nanoseconds but do not use them and as a result we have
unneeded scaling. Let's use second precision which is on par with our
need.

Also drop ns_to_localtime. It can be written much shorter and it feels like
we don't need distinct function here. Like when we formatting time in
say.c.

NO_TEST=refactoring
NO_DOC=refactoring
NO_CHANGELOG=refactoring

1. Add pretty-printer for 'rlist' type
Script holds table of possible rlists in tarantool. It recognizes
automatically actual type of entries and either expression refers to a
single items or to entire list. Then either single entry is displayed
(along with its index in the list) or the entire list (entry-by-entry).

2. Add command 'tt-list' that use the mentioned pretty-printer and
   provide additional functionality, namely:
- walking over the list (both directions)
- filter entries with predicate
- allow to explicitly specify list head and actual entry type if
  automatic recognition doesn't work
See 'help tt-list' for details.

Close #7731

NO_DOC=gdb extension
NO_CHANGELOG=gdb extension
NO_TEST=gdb extension

Currently all ballot_watcher fibers for different appliers have
identical names: "applier_ballot_watcher". It's hard to distinguish
them in logs and to know which ballot_watcher belongs to which applier,
and that complicates debugging quite a bit.

Let's fix this and name the fibers properly. With master's uri at the
end.

NO_DOC=no behaviour changed
NO_CHANGELOG=no behaviour changed
NO_TEST=doesn't need one

Add is_joinable parameter to applier_fiber_new. For now all usages of
the function pass true, but the next commit will create a non-joinable
applier fiber.

NO_DOC=refactoring
NO_TEST=refactoring
NO_CHANGELOG=refactoring

Some of the SQL modules have not been used for a long time. This patch
drops these modules.

NO_DOC=refactoring
NO_TEST=refactoring
NO_CHANGELOG=refactoring

A node configured in 'manual' mode and promoted by `box.ctl.promote()`
stays in is_candidate state for the whole term, even though it is not
is_cfg_candidate.

If such a node is the first one to notice leader death or to hit the
election timeout, it bumps the term excessively, then immediately
becomes a mere follower, because its is_candidate is reset with
is_cfg_candidate.

This extra term bump (one term after the node was actually promoted) is
unnecessary and might lead to strange errors:

tarantool> box.ctl.promote()
---
- error: 'The term is outdated: old - 3, new - 4'
...

Fix this by checking if a node is configured as a candidate before
trying to start new elections.

Closes #8168

NO_DOC=bugfix

There is a false assertion in raft_stop_candidate(): it assumes that the
node must always have a running timer whenever it sees the leader.
This is not true when the node is busy writing the new term on disk.

Cover the mentioned case in the assertion.

Closes #8169

NO_DOC=bugfix

Co-authored-by: Sergey Ostanevich <sergos@tarantool.org>

Commit 2be74a65 ("test/cmake: add a function for generating unit
test targets") added a function for generating unit test targets in
CMake. This function makes code simpler and less error-prone.

Proposed patch adds a similar function for generating performance test
targets in CMake.

NO_CHANGELOG=build infrastructure updated
NO_DOC=build infrastructure updated
NO_TEST=build infrastructure updated

Commit 2be74a65 ("test/cmake: add a function for generating unit
test targets") added a function for generating unit test targets in
CMake. This function makes code simpler and less error-prone.

Proposed patch adds a similar function for generating fuzzing test
targets in CMake.

NO_CHANGELOG=build infrastructure updated
NO_DOC=build infrastructure updated
NO_TEST=build infrastructure updated

Commit 2be74a65 ("test/cmake: add a function for generating unit
test targets") added a function for generating unit test targets in
CMake. However, due to a wrong scope UNIT_TEST_TARGETS remains empty
after generating unit tests. Patch fixes that.

NO_CHANGELOG=build infrastructure
NO_DOC=build infrastructure
NO_TEST=build infrastructure

We define flight_recorder_cfg struct, box_get_flightrec_cfg function,
and box.internal.cfg_configure_flightrec Lua function in the CE
repository although they are actually needed only in the EE repository.
Let's drop them all from the CE repository and instead define stub
functions box_check_flightrec and box_set_flightrec that would
check/apply box.cfg flight recorder parameters.

While we are at it, add missing comments to flightrec function stubs.

NO_DOC=refactoring
NO_TEST=refactoring
NO_CHANGELOG=refactoring

The approach

  - name: Set output
    run: echo "::set-output name={name}::{value}"

is deprecated [1].

Switching to the new approach:

  - name: Set output
    run: echo "{name}={value}" >> $GITHUB_OUTPUT

[1] https://github.blog/changelog/2022-10-11-github-actions-deprecating-save-state-and-set-output-commands/

NO_DOC=ci
NO_TEST=ci
NO_CHANGELOG=ci

The title is pretty self-explanatory. That's all this commit does. Now a
couple of words on why this is needed.

Commit 2a0c4f2b ("replication: make replica subscribe to master's
ballot") changed replica connect behaviour: instead of holding a single
connection to the master, replica may have two: master's ballot retrieval
is now performed in a separate connection owned by a separate fiber
called ballot_watcher.

First connection to master is initialized as always and then applier
fiber creates the ballot_watcher, which connects to the same address on
its own.

This lead to some unexpected consequences: random cartridge integration
tests started failing with the following error:
tarantool/tarantool/cartridge/test-helpers/cluster.lua:209:
"localhost:13303": Replication setup failed, instance orphaned

Here's what happened. Cartridge has a module named remote
control. The module mimics a tarantool server and "listens" on the same
socket the tarantool is intended to listen before box.cfg{listen=...} is
called.

For example one can see such output in tarantool logs with cartridge:
NO_WRAP
13:07:43.210 [10265] main/132/applier/admin@localhost:13301 I>
remote master 46a71a25-4328-4a41-985d-d93d6ed7fb7f at 127.0.0.1:13301  running Tarantool 2.11.0
13:07:43.210 [10265] main/133/applier/admin@localhost:13302 I>
remote master 00000000-0000-0000-0000-000000000000 at 127.0.0.1:13302  running Tarantool 1.10.0
13:07:43.210 [10265] main/134/applier/admin@localhost:13303 I>
remote master bcce45ad-38b7-4d8a-936a-133614a7775f at 127.0.0.1:13303  running Tarantool 2.11.0
NO_WRAP

The second "Tarantool" in the output (with zero instance uuid and
running Tarantool 1.10.0) is the remote control on an unconfigured
tarantool instance.

Before splitting applier connection in two, this was no problem: applier
would try to get the instance's ballot from a remote control listener and
fail (remote control doesn't answer to replication requests). Applier would
retry connecting to the same address until it got a reply, meaning that
remote control is stopped and real tarantool became listening on the
socket.

Now applier has two connections, and the following situation became
possible: when applier connection is initialized, remote control is
still working, and applier is connected to the remote control instance.
Applier performs ballot receipt in a separate fiber, which's not yet
initialized, so no errors are raised.

As soon as applier creates the ballot watcher, remote control is stopped
and the real tarantool starts listening on the socket. This means that
no error happens in the ballot watcher as well (normal tarantool answers
to replication requests, of course). And we get to an unhandled
situation when applier itself is connected to (already dead) remote
control instance, while its ballot watcher is connected to the real
tarantool.

As soon as applier sees the ballot is fetched, it continues connection
process to the already dead remote control instance and gets an error:
NO_WRAP
13:07:44.214 [10265] main/133/applier/admin@localhost:13302 I>
failed to authenticate
13:07:44.214 [10265] main/133/applier/admin@localhost:13302 coio.c:326 E>
SocketError: unexpected EOF when reading from socket,       called on fd 1620, aka 127.0.0.1:54150: Broken pipe
13:07:44.214 [10265] main/133/applier/admin@localhost:13302 I>
will retry every 1.00 second
13:07:44.214 [10265] main/115/remote_control/127.0.0.1:50242 C>
failed to synchronize with 1 out of 3 replicas
13:07:44.214 [10265] main/115/remote_control/127.0.0.1:50242 I>
entering orphan mode
NO_WRAP

Follow-up #5272
Closes #8185

NO_CHANGELOG=not user-visible
NO_DOC=not user-visible (can't create Tarantool with zero uuid)

When we rollback a prepared statement that deletes an MVCC story, we need
to reset the deleted story's PSN.

Closes #7930

NO_DOC=bugfix

During transaction rollback, we unconditionally assign a PSN to it: we
should do this only when necessary, i.e., a transaction is RW and is not
already prepared.

Needed for #7930

NO_CHANGELOG=refactoring
NO_DOC=refactoring
NO_TEST=refactoring

Currently, if transaction preparation fails, the transaction is left in an
inconsistent state: it has a PSN assigned to it, but its status is not
'prepared' — fix this by resetting its PSN.

Needed for #7930

NO_CHANGELOG=refactoring
NO_DOC=refactoring
NO_TEST=refactoring

During preparation of insert statements in MVCC, we define an old story and
abort all transactions that delete this story.

If there exists an older
story in the history chain, but the story is deleted by a prepared (not
necessarily committed) transaction, we consider that it de-facto does not
exist anymore — this logic is consistent, since during preparation of the
transaction deleting this story, the conflict resolution described above
was already done.

In this manner, there can be no more than one prepared statement deleting
a story at any point in time.

Closes #8104

NO_DOC=bugfix