# Description of Changes
This reapplies the patch from #3704, and fixes the issues that were
causing it to deadlock.
The reason it was deadlocking was that it allowed for the following
sequence of events:
* `SchedulerActor::handle_queued()` begins mutable tx
* `ModuleHost::disconnect_client()` submits call to `call_reducer(tx:
None)`
* scheduler submits call to `call_reducer(tx: Some)`
* `WasmModuleInstance::disconnect_client` now has to try to take tx
lock, but the scheduler's call_reducer already holds it and is behind it
in the queue
So, I moved most of the logic from `handle_queued` back to being
executed in the module worker thread, but kept the code in
`scheduler.rs` so that it can all be reasoned about locally.
Fixes#3645. Should I uncomment the implementation of
`ExportFunctionForScheduledTable for F: Procedure` now?
# Expected complexity level and risk
2 - there's a chance that this patch hasn't fully fixed the deadlock
issue from #3704, but I'm quite confident.
# Testing
- [x] Manually verified that deadlock no longer occurs - previously,
`while true; do python -m smoketests schedule_reducer -k
test_scheduled_table_subscription; done` would freeze up in only 2 or 3
iterations, but now it can run for 10 minutes without issues.
This reverts commit b2e37e8008.
# Description of Changes
<!-- Please describe your change, mention any related tickets, and so on
here. -->
Reverts #3704 which I'm pretty sure contains some sort of bug which is
causing the smoketests to hang.
# API and ABI breaking changes
None
<!-- If this is an API or ABI breaking change, please apply the
corresponding GitHub label. -->
# Expected complexity level and risk
1
<!--
How complicated do you think these changes are? Grade on a scale from 1
to 5,
where 1 is a trivial change, and 5 is a deep-reaching and complex
change.
This complexity rating applies not only to the complexity apparent in
the diff,
but also to its interactions with existing and future code.
If you answered more than a 2, explain what is complex about the PR,
and what other components it interacts with in potentially concerning
ways. -->
# Testing
<!-- Describe any testing you've done, and any testing you'd like your
reviewers to do,
so that you're confident that all the changes work as expected! -->
- [x] CI passing again
# Description of Changes
Reworks how `SchedulerActor::handle_queued` works so that it first
determines the parameters of the call to a reducer or the parameters of
the call to the procedure. This also enables the removal of the special
case `call_scheduled_reducer`.
Fixes#3645.
# API and ABI breaking changes
None
# Expected complexity level and risk
2
# Testing
A test `schedule_procedure` is added.
---------
Co-authored-by: Noa <coolreader18@gmail.com>
Co-authored-by: Phoebe Goldman <phoebe@goldman-tribe.org>
Co-authored-by: rekhoff <r.ekhoff@clockworklabs.io>
# Description of Changes
Closes#3517 .
With this PR, procedures (at least, those defined in Rust modules) can
perform HTTP requests! This is performed through a new field on the
`ProcedureContext`, `http: HttpClient`, which has a method `send` for
sending an `http::Request`, as well as a convenience wrapper `get`.
Internally, these methods hit the `procedure_http_request` ABI call /
host function, which uses reqwest to perform an HTTP request. The
request is run with a user-configurable timeout which defaults and is
clamped to 500 ms.
Rather than exposing the HTTP stream to modules, we download the entire
response body immediately, within the same timeout.
I've added an example usage of `get` to `module-test` which performs a
request against `localhost:3000` to read its own schema/moduledef.
This PR also makes all procedure-related definitions in the Rust module
bindings library `#[cfg(feature = "unstable")]`, as per #3644 . The
rename of the `/v1/database/:name/procedure/:name` route is not included
in this PR, so this does not close#3644 .
Left as TODOs are:
- Metrics for recording request and response size.
- Improving performance by stashing a long-lived `reqwest::Client`
someplace.
Currently we build a new `Client` for each request.
- Improving performance (possibly) by passing the request-future to the
global tokio executor
rather than running it on the single-threaded database executor.
# API and ABI breaking changes
Adds new APIs, which are marked as unstable. Adds a new ABI, which is
not unstable in any meaningful way (we can't really do that). Marks
unreleased APIs as unstable. Does not affect any pre-existing
already-released APIs or ABIs.
# Expected complexity level and risk
3 or so: networking is scary, and even though we impose a timeout which
prevents these connections from being truly long-lived, they're still
potentially long-lived on the scale of Tokio futures. It's possible that
running them on the database core is problematic in some way, and so
what I've left as a performance TODO could actually be a
concurrency-correctness issue.
# Testing
- [x] Manually wrote and executed some procedures which make HTTP
requests.
- [x] Added two automated tests to the `sdk-test` suite,
`procedure::http_ok` and `procedure::http_err`, which make successful
and failing requests respectively, then return its result. A client then
makes some assertions about the result.
---------
Co-authored-by: Noa <coolreader18@gmail.com>
# Description of Changes
Adds `ProcedureContext::{with_tx, try_with_tx}`.
Fixes https://github.com/clockworklabs/SpacetimeDB/issues/3515.
# API and ABI breaking changes
None
# Expected complexity level and risk
2
# Testing
An integration test `test_calling_with_tx` is added.
# Description of Changes
This commit adds support to the Rust client SDK for calling procedures.
Similar to reducers, each `DbContext` implementor has a `pub procedures:
RemoteProcedures` field, with methods provided by extension traits for
each procedure.
Unlike reducers, the provided methods are invoke and invoke-then.
Invoke-then takes a `FnOnce` callback to run
when the SDK is notified of the procedure's termination status, while
invoke ignores that notification.
No mechanism is provided for observing procedures invoked by other
clients.
Procedure callbacks are implemented by storing a map from `request_id`
to `ProcedureCallback`, with the callback closure internally knowing how
to deserialize the return value. It's mildly unfortunate to deserialize
within the callback instead of on the preprocess background task, but it
saves significant complexity.
This commit also adds a new sdk-test module, `sdk-test-procedures`, and
a new Rust test client, `procedure-client`.
Together, these are used in two tests of invoking and observing
procedures. I've left TODOs for other tests that we should write as we
implement additional procedure features.
I also had to fix a few minor bugs in the Rust codegen which were not
strictly related to procedures: we previously assumed that the sets of
reducers and of tables were non-empty, which led to wonky invalid
codegen on modules which did not define any reducers or which did not
define any tables.
I'm sneaking a change to the Nix flake into this PR as well - when
initially writing it I had included `cargoArtifacts` (the pre-built and
cached dependencies of our actual builds) in its `packages`, but that
was neither necessary or useful, and just made building the shell take a
long time after dependency changes.
# API and ABI breaking changes
Breaks the internal interface between the Rust client SDK and codegen,
so users will have to re-run `spacetime generate`.
# Expected complexity level and risk
2-ish? Pretty simple change to the Rust SDK overall.
# Testing
- [x] Added new automated integration tests exercising the new
functionality.
---------
Co-authored-by: Zeke Foppa <bfops@users.noreply.github.com>
Noa