# Description of Changes
Closes
[#2686](https://github.com/clockworklabs/SpacetimeDB/issues/2686).
Add support for listening using the [PG wire
protocol](https://www.postgresql.org/docs/current/protocol.html) so `pg`
clients could be used against the database.
# API and ABI breaking changes
The output of `duration` is changed to `rfc3339`, instead of the way is
made with `sats` because is what is done in `pg`, see note below.
# Expected complexity level and risk
2
~~There is open questions that are in the [ticket
#2686](https://github.com/clockworklabs/SpacetimeDB/issues/2686). Also
the crate used here require `RustTls`, so it could be good idea to
decide if~~:
* ~~Rewrite a big chunk of code to use `OpenSSL`~~
* ~~Move to `RustTls`
https://github.com/clockworklabs/SpacetimeDB/pull/1700~~
* ~~Pay for the extra compilation cost~~.
I open another port(`5433`) to listen for `pg` connections using `ssl`.
Need to be decided if this is the way or instead try to multi-plex the
current port for both protocols.
# Testing
Only manual testing so far. Solving the above questions allow me to
implement some unit tests. Also, not yet integrated into cloud for the
same reasons.
- [x] Adding some test for the binary encoding of special and primitive
types
- [x] Smoke test using `psql` that connect to the db instance and run
some queries
- [x] Manually inspect using a UI database explorer how infer the types,
some of this tools generate special widgets when displaying `json,
duration, etc`
---------
Co-authored-by: Noa <coolreader18@gmail.com>
# Description of Changes
Implements [subscribing to durable
commits](https://github.com/clockworklabs/SpacetimeDBPrivate/issues/1594).
The setting works on a per-connection level, and essentially just delays
sending transaction updates until the transaction is reported as durable
by the database.
For connectionless SQL operations, the setting works per-request. No SQL
syntax is provided by this patch to toggle the configuration.
After some deliberation, I opted to obtain the offset when a transaction
commits (as opposed to when it starts). This creates some mild
inconvenience, because we prevent the transaction from committing until
the corresponding subscription updates are enqueued.
The strategy is, however, more correct should we ever support weaker
isolation levels, and it is easier to document.
Follow-ups include:
- Provide SQL syntax (`SET synchronous_commit = ON` or something)
- C# and TypeScript SDKs
- Reference docs?
# API and ABI breaking changes
Not breaking, but adds a parameter to the subscribe and sql endpoints.
# Expected complexity level and risk
4
To the author's understanding, ordering of outbound messages is not
changed by this patch, even if there are messages that don't have a
transaction offset (such as error messages). I.e. while waiting for the
transaction offset of a message to become durable, no message enqueued
after that message will be delivered. This may not be desirable in some
cases.
The patch may contain concurrency bugs, e.g. awaiting futures that may
never resolve.
# Testing
- [x] Implemented a new test in the `module_subscription_actor` module
- [x] Added unit tests for the core logic in `ClientConnectionReceiver`
It would be desirable to also have integration-level tests, but I'm
currently unsure how to write those without being able to control if and
when the database reports an offset as durable.
---------
Signed-off-by: Kim Altintop <kim@eagain.io>
Co-authored-by: Phoebe Goldman <phoebe@clockworklabs.io>
# Description of Changes
Please note, much of the code changed in this PR is generated code.
This change updates the TypeScript SDK to use a new `spacetimedb`
TypeScript library which lives under the `/crates/bindings-typescript`
folder alongside `/crates/bindings-csharp`. Just like with the C#
bindings library, the types for `AlgebraicType` and `RawModuleDef` are
now code generated with a script in `/crates/codegen/examples`.
Pulling this out into a library allows us to use the same types and
serialization code on both the server and the client for TypeScript
modules.
In the process of making this change I also found and fixed several
issues with the TypeScript code generation. Namely an issue with
recursive types and an issue with the `never` type. I also removed any
use of `namespace`s since those are a TypeScript only feature, and we
want to have JavaScript + types so that we can use the generated code
with ESBuild without TSC.
I have also improved the npm/pnpm scripts to be able to generate
TypeScript code for us automatically by running `pnpm generate` for any
place that we have to generate typescript code. Namely:
- Quickstart module bindings
- AlgebraicType/ModuleDef for TypeScript module library
- Client API messages for the TypeScript API
- TestApp module bindings
# API and ABI breaking changes
IMPORTANT! This is an API breaking change for clients, as such it should
be a major version change. However, I am going to see if I can shim in
the old API as best as possible to make it compatible.
Notably, we were previously exporting APIs that end users do not need,
and I don't think it would ever occur to them to use, namely the whole
`AlgebraicValue` API and also the `AlgebraicType` API. In principle, no
one should have a need for these, although it was technically possible
for them to use it.
Indeed, we could potentially even just remove AlgebraicType completely
from the API by directly code generating the serialization code.
Listed below are all of the **BREAKING** changes to the API and their
effect:
- `AlgebraicType` is now a structural union literal type instead of a
class (nominal type), this is a consequence of generating the type with
our code gen. Users did not have a reason to use `AlgebraicType`
directly.
- The `AlgebraicValue` type has been removed entirely. This was
previously a class that was exported from the SDK, but very unlikely to
be used by users.
- The `ProductValue` type has been removed.
- The `ReducerArgsAdapter` and `ValueAdapter` types, which were used
with AlgebraicValues have been removed.
- Generated code has changed incompatibly so users will have to
regenerate code when upgrading to this version. Technically a breaking
change, but pretty easy to fix.
Listed below are the non-breaking API changes:
- I am now exporting generated product types as the default export in
addition to how I was exporting them before
- For generated sum type `T`, I am now exporting a `TVariants` namespace
which has the types of all the variants for `T`. This was previously
exposed on the namespace `T`, but was inaccessible in modules other than
the one it was defined in because I also export a type `T` in addition
to namespace `T` and in the type position `T` was being interpreted as a
type rather than a namespace. It's unclear why TypeScript resolved it as
the `T` namespace within the module in which is was defined previously.
Anyway, since the old types were apparently unobservable to users, I've
replaced them with the types in `TVariants`. (Open to other names for
this namespace).
- I fixed a bug where never types (sum types with no variants) were not
correctly generated.
# Expected complexity level and risk
3. The most complex thing about the PR are the potential impacts to the
API. I am reasonably certain, but not 100% certain that I have accounted
for everything above.
# Testing
- [x] I ran `pnpm test` which runs all the tests in the repo including
an integration test which tests the connection to SpacetimeDB. I also
fixed broken tests.
---------
Co-authored-by: Zeke Foppa <bfops@users.noreply.github.com>
# Description of Changes
<!-- Please describe your change, mention any related tickets, and so on
here. -->
- Simple patch version bump. Because we forgot to bump the version
number for 1.3.1 this is bumping from 1.3.0 => 1.3.2
# API and ABI breaking changes
0
# Testing
- Verified that the version number in the license has been updated
- Verified the output of `spacetime --version`:
```
$ spacetime --version
spacetime Path: C:\Users\boppy\AppData\Local\SpacetimeDB\bin\current\spacetimedb-cli.exe
Commit: 0027d094e1
spacetimedb tool version 1.3.2; spacetimedb-lib version 1.3.2;
```
---------
Signed-off-by: John Detter <4099508+jdetter@users.noreply.github.com>
# Description of Changes
Bumped all versions to 1.3.0 in preparation for an upcoming minor
release.
# API and ABI breaking changes
No breaking changes.
# Expected complexity level and risk
1
# Testing
None
---------
Co-authored-by: Zeke Foppa <bfops@users.noreply.github.com>
The `subscribe` command would drop the connection without sending a
close frame. Doing so creates a warning on the server, which can be
distracting when debugging other connections issues.
# Expected complexity level and risk
1
# Testing
Use the `subscribe` command with a local database and compare server
logs before
and after.
Mario Montoya