Skip to content

rebase onto latest upstream#1

Open
rmburg wants to merge 17 commits into
pin-protocol-versionfrom
pin-protocol-version-rebased
Open

rebase onto latest upstream#1
rmburg wants to merge 17 commits into
pin-protocol-versionfrom
pin-protocol-version-rebased

Conversation

@rmburg

@rmburg rmburg commented Jun 23, 2026

Copy link
Copy Markdown
Member

Might fix HULKs/hulk#2404

alvgaona and others added 17 commits February 19, 2026 16:24
* fix: store enriched endpoint data with participant default locators in discovery_db

When a remote DDS implementation (e.g. CycloneDDS) omits per-endpoint
locators in SEDP announcements and relies on the participant's default
locators from SPDP, update_subscription() and update_publication()
would store the raw data with empty locator lists. The enriched copy
(with default locators filled in) was only returned for the immediate
notification but never stored back.

This caused endpoints created after initial SEDP discovery to get
reader/writer proxies with empty locator lists, silently dropping all
data. Service calls (single request/reply) failed 100% of the time.

The fix builds the enriched data first, then stores the enriched
version in external_topic_readers/external_topic_writers so that later
lookups via readers_on_topic()/writers_on_topic() return usable locator
information.

* feat: re-export SampleIdentity from crate root

SampleIdentity is required to call WriteOptionsBuilder::related_sample_identity()
but was not publicly accessible. It lived in structure::rpc which is pub(crate),
making the public API method impossible to call from downstream crates. This is
needed for ROS 2 service interop with FastDDS, which uses RTPS Inline QoS
PID_RELATED_SAMPLE_IDENTITY for service request/reply correlation.
…tostek#401)

The DomainParticipantStatusEvent channel had a capacity of 16 and
silently dropped events when full, causing downstream consumers to miss
endpoint discoveries entirely with no indication of data loss.

A single participant can expose many endpoints (e.g. ~16 in a typical
ROS 2 node), so even two participants overwhelm a 16-slot channel
during the initial SEDP burst.

Changes:
- Increase status channel capacity from 16 to 2048
- Upgrade log level from trace! to warn! on channel overflow
Extract PID_USER_DATA from discovered reader and writer parameter lists
and expose it on EndpointDescription. This allows applications to access
DDS user data set by remote participants, such as ROS 2 type hashes.
RustDDS used OS-default UDP receive buffers (~786 KB on macOS, ~208 KB
on Linux), which caused silent packet loss when receiving large
fragmented DDS messages (e.g., 1080p images at ~6 MB). The buffer
overflowed during fragment bursts, and with BestEffort QoS, lost
fragments are never retransmitted.

Set SO_RCVBUF to 8 MB by default (configurable via
DomainParticipantBuilder::socket_receive_buffer_size). On Linux,
users may need to raise net.core.rmem_max for larger payloads.
This matches the approach used by CycloneDDS and FastDDS.
Validate readerSNState.base against last_change_sequence_number + 1 so a
crafted ACKNACK cannot permanently corrupt all_acked_before (issue Atostek#405).

Includes unit tests.

Made-with: Cursor
@github-project-automation github-project-automation Bot moved this to In Progress in Development Jun 23, 2026
@rmburg rmburg moved this from In Progress to Review in Development Jun 23, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

Status: Review

Development

Successfully merging this pull request may close these issues.

5 participants