Websockets asyncio#152
Conversation
gonzalocasas
requested review from
EzraBrooks,
MatthijsBurgh,
chenkasirer and
jihoonl
|
@RobotWebTools/roslibpy if anyone has comments on this, I would appreciate it! |
|
chenkasirer
left a comment
chenkasirer
left a comment
There was a problem hiding this comment.
Discussed with @gonzalocasas, seems that authobahn already has an asyncio transport option that can be used instead of twisted, so perhaps that's a better approach.
The asyncio transport now runs the same Autobahn WebSocket stack as the default transport, on an asyncio event loop instead of the Twisted reactor, rather than depending on the standalone websockets library. Because both transports build on Autobahn, whose txaio layer binds a single async framework per process, the twisted and asyncio transports are mutually exclusive within one process. To accommodate that: - The back-compat RosBridgeClientFactory binding resolves lazily (PEP 562), so importing roslibpy no longer eagerly locks txaio to twisted before a transport is chosen. - select_factory translates the deep txaio RuntimeError into a clear, actionable conflict message. - Tests select a single transport per process via ROSLIBPY_TRANSPORT, and the ROS 1/2 CI workflows run the suite once per transport in separate processes. The asyncio extra is dropped since no extra dependency is required beyond Autobahn. Verified against live ROS 1 and ROS 2 rosbridge: asyncio matches the twisted baseline test-for-test. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
|
Besides the benchmarks that CI runs, I did several runs locally with different settings, including a multi-iteration run against ROS2: Transport benchmark — ROS 2, 6× run (high sample count)Ran
Values are
Conclusions
No regression on any axis. Consistent with the ROS 1 benchmark and with the |
| version: 2 | ||
|
|
||
| build: | ||
| os: ubuntu-22.04 |
| build: | ||
| os: ubuntu-22.04 | ||
| tools: | ||
| python: "3.11" |
3 participants
Trying out an alternative, opt-in transport/event loop based on the asyncio implementation of Autobahn.
The new transport can be be manually selected via a
transportkwargs onRos, or set viaset_default_transport, or with an environment variable.I added some benchmarks as well. The benchmarks run on CI, the results are uploaded to the summary page of the CI task, here are the results of one of the latest runs. The conclusion is that the new transport has marginally higher topic throughput.
roslibpy transport benchmark
These numbers are sampled on shared CI infrastructure. Topic p95 and max
latencies can be noisy because of runner scheduling, Docker networking,
and rosbridge timing. Prefer medians and throughput for quick comparisons;
interpret tail latency across several runs.
What type of change is this?
Checklist
Put an
xin the boxes that apply. You can also fill these out after creating the PR. If you're unsure about any of them, don't hesitate to ask. We're here to help! This is simply a reminder of what we are going to look for before merging your code.CHANGELOG.rstfile in theUnreleasedsection under the most fitting heading (e.g.Added,Changed,Removed).invoke test).invoke check).