Implement compression of objects before they are saved into the storage
provider. Each container type can specify which algorithm to use
(currently `none` or `zstd`) and the compression level.
Switch out from using uvicorn to hypercorn for the pypeline daemon, make
the facilities create hypercorn commands in stand-alone functions, as
they will be re-used for other servers later on.
Expand the `ProjectMetadata` dataclass to include additional timestamps
and the hash of the pipeline description. Change all the
`StorageProvider`s so that all the new data is saved and returned.
Consolidate the saving of object dependencies and saving the actual
objects' data into a single function called `add_objects`. The previous
methods `add_dependencies`, `add_custom_invalidation_data` and `put`
have been removed.
Due to the drifting differences, split `ScheduledTask` into
`PipeScheduledTask` and `SavepointScheduledTask`, which allows each to
specialize on the task it has to accomplish.
Make the issuing of notification completely out of band with respect to
the `Pipeline`. Now the `StorageProvider` is responsible for providing
notifications to clients. Since some providers are local-only, there is
a `LOCAL_QUEUE` which allows sensing notifications through a local
queue, re-using the local revng daemon.
Split the `invalidate` method in two phases: in `invalidateCheck` it
just checks if the model diff warrants the execution of the actual
`invalidate` method, which requires fetching the custom invalidation
data from storage. The actual `invalidation` method remains the same.
This should make invalidation faster for storage providers that store
the invalidation remotely.
Overhaul the pipeline configuration logic by collapsing all dynamic
configuration options, both for analyses and pipes into a single
dictionary. Change all the interfaces so that there is no longer
distinction between the configuration of an analysis and of pipes.
Expose these options to the command line via `--{name}-configuration`
options for each pipe/analysis that is applicable to the command-line
invocation.
Rework the artifact endpoint, making it similar to the CLI invocation.
It now works on a single artifact and can be made to return both `json`
and `tar` as formats.
Add categories in the pipeline, allow an artifact to have a category
specified which allows it to be shown or hidden by default. Adapt the
CLI tools to hide artifacts of the category that don't
`show_by_default=True`.
Rework which information is transmitted in the pipeline metadata,
avoiding redundancy and moving some information there instead of
returning it every time a request is made.
Change the `get_model` and `set_model` interface of `StorageProvider` so
that it is responsibility of the `StorageProvider` to
serialize/deserialize the model before returning to the caller.
This is in preparation to the model migration being implemented, since
it's now a responsibility of the storage provider to deserialize it it
can trivially re-save it to disk if it is migrated.
Instead of executing `chdir` when using the `-C` option, which is
fragile when other paths are involved, store the option in the click
context and propagate it to the required code paths that require knowing
what the base directory is.
Add the functionality the the pipeline infrastructure and CLI to run
individual pipe and analyses as subcommands instead of in-process. This
allow better debuggability of individual pipes.
Improve the handling of signals by leveraging the `Py_AtExit`
functionality to trigger cleanup when the interpreter exits.
When receiving SIGINT handle it specially because exceptions in python
are only thrown when the interpreter is running and not when C code is.
Add infrastructure to pypeline that allows containers to be notified
when they are being used last, this allows two things:
* `Pipe`s eagerly clearing those containers once they are done reading
their contents
* `ScheduledTask`s clearing those out at the end of their execution in
case the pipe did not do it
This overall should improve memory usage as container no longer take up
memory if they are no longer used as part of a `Schedule`.
Inline the body of `PipelineNode.run` into `ScheduledTask.run`, making
`PipelineNode` a pure data structure. This also allows eliding all the
arguments since all of them as properties of `ScheduledTask`.
In equivalence with `LLVMFunctionMixin`, add a mixin that allows
writing a pipe that instead of defining the `runOnFunction` method it
defines the `runOnCliftFunction` which receives both the
`module::Function` and the `FunctionOp`.
Add an enum to explicitly specify the `TaskArgumentAccess` of a given
`Pipe`. This is needed because in some cases it's required to manually
specify the `READ` or `WRITE` access.
Move the epoch inside the storage_provider, this allows us to make the
storage_provider_factory to return an async context manager.
This way we can simplify the locking for the daemon, but forces the
CLI project commands to deal with async code.
Giacomo Vercesi