It is now possible to write in migration `@_plugin_config.foo` where foo is
variable from plugin configuration. Note that configuration typically
belongs to individual services, in catalog this field is called entity. For
migration special values there is special entity called `migration_context`.
So corresponding alter looks this way:

ALTER PLUGIN p 0.1.0 SET migration_context.foo = '"nondefault"'

(extra quotes are temporary, and should be dealt with separately)

Closes https://git.picodata.io/picodata/picodata/picodata/-/issues/659

Previouslly we've blocked on scope termination before reading from
channel, so blocking still occured.

With this patch there is new `blocking` wrapper function that executes
provided closure on separate thread and yields until result is ready

an intermediate variable is needed to avoid passing GIT_DESCRIBE to dependent pipelines

Виталий Шунков authored 4 months ago and Виталий Шунков committed 4 months ago

When ignoring the .git directory, the image build in the dependent pipeline breaks, since git describe is executed not only in picodata, but also in the tarantool-sys, tarantool-sys/third_party/luajit submodules

Previously, due to the default option --always, git_version::git_version
returned only a raw git commit instead of failing and fallbacking version env.
This commit disables this option.

Needed since introduction of plugin version compatibility check in
86a08ef9. For release candidate versions
we need to properly bump all versions. We could've bumped picodata
version instead but for simplicity just revert to currently unreleased
24.6.0. Ideally we should describe release candidates in our versioning
policy.

The patch prepares picodata-plugin and picodata-plugin-proc-macro
for publishing to crates.io.

This includes Makefile wrapper over cargo commands and manual CI job

For now minimum of required fields have been filled. We should come back
to fill proper readmes, categories, home page etc.

picoplugin -> picodata-plugin
picoplugin_gen -> picodata-plugin-proc-macro

This patch includes several useful changes:
* Tarantool now dumps box include flags at build time (useful for FFI).
* ASan-related fixes (bespoke & from upstream).

Georgy Moshkin authored 5 months ago and Dmitry Rodionov committed 5 months ago

Previous implementation would call functions over the ffi boundary which
weren't defined with the required `extern` specifiers, which is not allowed.

fix: mark test_dml_on_global_tbls flaky test with xfail until we move on new vshard version (which should fix problem with inconsistent reads)

This commit embeds in plugins a new variable, containing picoplugin
version used to build it. If this version differs from the picodata version
when the plugin gets loaded, the loading fails with an error reporting
the expected and the current versions. This check can be disabled via env
PICODATA_UNSAFE_DISABLE_PLUGIN_COMPATIBILITY_CHECK.

Dmitry Ivanov authored 5 months ago and Dmitry Ivanov committed 5 months ago

Log:

```
i1  | 2024-10-09 02:40:07.850 [135492] main/133/main C> panicked at library/core/src/panicking.rs:221:5:
i1  | panic in a function that cannot unwind
i1  | 2024-10-09 02:40:07.850 [135492] main/133/main C> backtrace:
i1  | disabled backtrace
i1  | 2024-10-09 02:40:07.850 [135492] main/133/main C> aborting due to panic
i1  | [supervisor:135482] no ipc message from child
i1  | [supervisor:135482] subprocess 135492 was signaled with SIGABRT
i1  | [supervisor:135482] core dumped
```

Arseniy Volynets authored 5 months ago and Denis Smirnov committed 5 months ago

Now SQL explain shows the user what buckets the query would be
executed on. Some results are exact, others are not (we return
`buckets: unknown` or `buckets: <= [..]` according to the query).