rosotacom-dev 2.5.dev55__tar.gz → 2.5.dev56__tar.gz

{rosotacom_dev-2.5.dev55 → rosotacom_dev-2.5.dev56}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: rosotacom-dev
-Version: 2.5.dev55
+Version: 2.5.dev56
 Summary: Host CLI for ROS OTA communication sessions.
 Author: develNor
 License-Expression: BSD-3-Clause

{rosotacom_dev-2.5.dev55 → rosotacom_dev-2.5.dev56}/src/rosotacom/cli.py

@@ -12,6 +12,7 @@ from __future__ import annotations
 import argparse
 import base64
 import binascii
+import contextlib
 import getpass
 import hashlib
 import importlib
@@ -23,12 +24,13 @@ import os
 import re
 import shlex
 import shutil
+import signal
 import subprocess
 import sys
 import tarfile
 import tempfile
 import time
-from collections.abc import Callable, Mapping, Sequence
+from collections.abc import Callable, Iterator, Mapping, Sequence
 from dataclasses import dataclass, replace
 from datetime import datetime
 from pathlib import Path
@@ -2761,12 +2763,49 @@ def _passwordless_watchdog_command(argv: Sequence[str]) -> str:
     return f"nohup sudo -n {shlex.join(argv)} >/dev/null 2>&1 &"
 
 
+@contextlib.contextmanager
+def _terminate_via_exception() -> Iterator[None]:
+    """Make SIGTERM unwind the stack instead of ending the process outright.
+
+    The OTA run already has a `finally` that reverts shaping, collects logs and
+    stops both peers. SIGINT reaches it, because Python raises KeyboardInterrupt.
+    SIGTERM does not: the default disposition ends the process immediately, so
+    `finally` never runs and every peer keeps its containers -- and, worse, any
+    netem qdisc stays armed on the peer's own interface, quietly degrading a
+    machine nobody is looking at.
+
+    SIGTERM is not the exotic case. `timeout` sends it, systemd sends it, and a
+    CI job cancelling a step sends it. On 2026-08-19 a run killed by `timeout`
+    left containers behind on both peers, which then blocked the next run's
+    exclusivity check.
+
+    Raising SystemExit gives the same exit status while letting the cleanup run.
+    The handler is restored on the way out so this does not leak into a caller.
+    """
+
+    def handler(signum: int, _frame: object) -> None:
+        raise SystemExit(128 + signum)
+
+    try:
+        previous = signal.signal(signal.SIGTERM, handler)
+    except ValueError:
+        # Not the main thread; the parent already owns signal disposition.
+        yield
+        return
+    try:
+        yield
+    finally:
+        signal.signal(signal.SIGTERM, previous)
+
+
 def _ota_conflict_check(plan: OtaSmokePlan, *, dry_run: bool) -> None:
     """Fail-safe preflight: OTA runs require exclusive control of the peers.
 
-    Aborts when a peer already runs rosotacom containers or has active network
-    shaping (netem/tbf/htb) on the data interface toward another peer — both
-    signal an OTA run that is still active or was not cleaned up.
+    Aborts when a peer already runs rosotacom containers, has active network
+    shaping (netem/tbf/htb) on the data interface toward another peer, or is
+    carrying so much unrelated load that a measurement taken on it would not
+    mean anything. The first two signal an OTA run that is still active or was
+    not cleaned up; the third signals a machine somebody else is using.
     """
     from .network_shaper import ota_interface_from_route
 
@@ -2820,6 +2859,59 @@ def _ota_conflict_check(plan: OtaSmokePlan, *, dry_run: bool) -> None:
                 )
 
 
+# A run on a machine somebody else is saturating does not fail — it succeeds
+# and reports numbers that describe the contention rather than the link. That
+# is the expensive kind of wrong, because nothing in the artefact says so
+# afterwards.
+#
+# The thresholds are per-CPU, so they read the same on an 8-core laptop and a
+# 32-thread desktop. At 2.0 the run stops: with twice as many runnable threads
+# as CPUs, a thread waits on average as long as it runs, which lands directly
+# in the scheduling latency a transport measurement is trying to resolve.
+_LOAD_WARN_PER_CPU = 0.7
+_LOAD_REFUSE_PER_CPU = 2.0
+
+
+def _ota_load_check(plan: OtaSmokePlan, *, dry_run: bool) -> None:
+    """Refuse to measure on a peer that is busy with someone else's work."""
+    for peer in plan.peers.values():
+        result = _ota_run(
+            peer,
+            "cut -d' ' -f1-3 /proc/loadavg; nproc",
+            label=f"{peer.name}: host load",
+            dry_run=dry_run,
+            batch=True,
+            check=False,
+        )
+        if dry_run or result.returncode != 0:
+            continue
+        lines = (result.stdout or "").split()
+        if len(lines) < 4:
+            continue
+        try:
+            load1, load5, load15 = (float(value) for value in lines[:3])
+            cpus = max(1, int(lines[3]))
+        except ValueError:
+            continue
+        # load5 as well as load1: a run takes minutes, and a machine that was
+        # busy a moment ago is about to be busy again.
+        ratio = max(load1, load5) / cpus
+        # Two decimals to match /proc/loadavg and uptime, so the number in an
+        # artefact can be compared against what the machine reports itself.
+        summary = f"load {load1:.2f}/{load5:.2f}/{load15:.2f} on {cpus} CPUs ({ratio:.2f} per CPU)"
+        if ratio >= _LOAD_REFUSE_PER_CPU:
+            raise RuntimeError(
+                f"Peer {peer.name} is carrying unrelated load: {summary}.\n"
+                "A measurement taken here would describe the contention, not the link.\n"
+                "Wait for the machine to go quiet, measure on another pair, or — if this "
+                "load is yours and expected — pass --skip-conflict-check."
+            )
+        if ratio >= _LOAD_WARN_PER_CPU:
+            print(f"! {peer.name}: {summary} — comparisons survive if runs are interleaved, absolute latencies do not")
+        else:
+            print(f"+ {peer.name}: {summary}")
+
+
 def _ota_preflight(
     plan: OtaSmokePlan,
     *,
@@ -2924,6 +3016,7 @@ def _ota_preflight(
 
     if check_conflicts:
         _ota_conflict_check(plan, dry_run=dry_run)
+        _ota_load_check(plan, dry_run=dry_run)
 
 
 _OTA_STAGE_EXCLUDES = {
@@ -3137,6 +3230,7 @@ def _ota_verify_checkouts(plan: OtaSmokePlan, *, dry_run: bool) -> None:
     distinct one-line failure here, before any container starts, rather than a
     confusing one an hour later.
     """
+    peer_profiles: dict[str, list[str]] = {}
     for peer in plan.peers.values():
         checkout = shlex.quote(_ota_peer_workdir(plan, peer))
         project = shlex.quote(plan.project)
@@ -3174,6 +3268,46 @@ def _ota_verify_checkouts(plan: OtaSmokePlan, *, dry_run: bool) -> None:
             # a real hazard, but it is also a legitimate mid-migration state,
             # and a check that refuses it would be the wrong place to find out.
             print(f"+ {peer.name}: rosotacom {version.stdout.strip() or 'unknown'}")
+        profiles = _ota_run(
+            peer,
+            f"{OTA_CHECKOUT_PATH_PREFIX}"
+            'ls "$(rosotacom resources path ws)"/ota_configs/*.template '
+            "2>/dev/null | xargs -r -n1 basename | sort | tr '\\n' ' '",
+            label=f"{peer.name}: OTA profile templates",
+            dry_run=dry_run,
+            batch=True,
+            check=False,
+        )
+        if not dry_run and profiles.returncode == 0:
+            peer_profiles[peer.name] = profiles.stdout.split()
+
+    # Unlike the version, this one refuses. A peer whose installed stack lacks
+    # the template a session names does not fail loudly: the renderer errors,
+    # the profile file is left unusable, and the DDS stack silently falls back
+    # to its own defaults — multicast discovery, which no tunnelled OTA link
+    # carries. The run then completes and reports zero delivered messages, which
+    # reads as a middleware that does not work. That is exactly what happened on
+    # 2026-08-19, when #299 made fastdds_tuned.xml the Fast DDS default while
+    # the deployed stack predated the template: eight runs, all recorded as
+    # evidence about Fast DDS, none of which had ever configured it.
+    if len(peer_profiles) > 1:
+        names = sorted(peer_profiles)
+        reference = set(peer_profiles[names[0]])
+        for name in names[1:]:
+            missing = reference - set(peer_profiles[name])
+            extra = set(peer_profiles[name]) - reference
+            if missing or extra:
+                detail = []
+                if missing:
+                    detail.append(f"absent on {name}: {', '.join(sorted(missing))}")
+                if extra:
+                    detail.append(f"only on {name}: {', '.join(sorted(extra))}")
+                raise RuntimeError(
+                    f"Peers disagree about their OTA profile templates ({'; '.join(detail)}). "
+                    "A missing template renders nothing and the DDS stack falls back to "
+                    "defaults, so the link would come up and carry no data. Bring the peers "
+                    "to the same rosotacom before measuring."
+                )
 
 
 def _ota_prepare_hosts(args: argparse.Namespace, runtime: RuntimeConfig, plan: OtaSmokePlan) -> None:
@@ -4159,6 +4293,7 @@ def _start_interactive_ota_smoke(args: argparse.Namespace) -> int:
 
     if not getattr(args, "skip_preflight", False) and not getattr(args, "skip_conflict_check", False):
         _ota_conflict_check(plan, dry_run=dry_run)
+        _ota_load_check(plan, dry_run=dry_run)
     _ota_prepare_hosts(args, runtime, plan)
     instance = _resolve_session_instance(
         runtime,
@@ -4355,7 +4490,8 @@ def ota_smoke(args: argparse.Namespace) -> int:
                 "session). Run the non-interactive ota-smoke to shape the link, or apply the profile manually."
             )
         return _start_interactive_ota_smoke(args)
-    return _start_noninteractive_ota_smoke(args)
+    with _terminate_via_exception():
+        return _start_noninteractive_ota_smoke(args)
 
 
 def _scenario_name_completer(

rosotacom_dev-2.5.dev56/src/rosotacom/resources/ws/ota_configs/fastdds_local.xml.template

@@ -0,0 +1,80 @@
+<?xml version="1.0" encoding="UTF-8"?>
+<!--
+  Local-domain Fast DDS configuration: pin the loopback locators, change nothing else.
+
+  This is the Fast DDS counterpart to cyclonedds_local_participants.xml, and it
+  exists because the two stacks fail differently when nothing is configured.
+  CycloneDDS talks between containers on a host out of the box. Fast DDS does
+  not, and the way it fails is quiet: the publisher reports a matched
+  subscriber, the subscriber reports zero messages, and the session's status
+  shows the last stage IDLE with `pub=1` next to it — a pipeline that looks
+  connected and moves nothing.
+
+  Measured on lamborghini, 2026-08-19, two containers on `--network host`,
+  84 B at 10 Hz, ~80 samples per run:
+
+    no profile at all .................................. 0 of 80
+    + shared memory off (useBuiltinTransports=false) ... 0 of 80
+    + a unicast initial peer on 127.0.0.1 .............. 0 of 80
+    + explicit unicast locator lists ................... 79 of 80
+    locator lists ALONE, shared memory left on ......... 79 of 80
+    (control) CycloneDDS, no profile ................... 79 of 80
+
+  So the locator lists are the whole fix, and shared memory is a red herring —
+  worth stating because turning SHM off is the obvious first guess and it
+  changes nothing here.
+
+  Why they are load-bearing: with no locator list Fast DDS binds every interface
+  it can enumerate and announces all of them. A developer machine running Docker
+  has dozens of `br-*` bridges, most of them DOWN, and a peer that walks that
+  list can settle on one that carries nothing. Naming 127.0.0.1 makes the
+  announcement match the one path that is certain to work between processes on
+  a shared network namespace.
+
+  Scope, and its one limit: this pins the LOCAL domain, which is host-local by
+  definition, so loopback is enough — and it is more robust than the host's
+  routable address, which depends on which interface is up. It assumes the
+  session's containers share a network namespace (`--network host`, which the
+  ros2docker profiles set). A container on its own bridge network has its own
+  127.0.0.1 and would need its address named here instead.
+
+  Nothing else is set: no interface allowlist, no transport override, no
+  discovery tuning. Unlike the OTA templates this profile is not trying to cross
+  a tunnel.
+-->
+<dds xmlns="http://www.eprosima.com/XMLSchemas/fastRTPS_Profiles">
+  <profiles>
+    <participant profile_name="rosotacom_fastdds_local" is_default_profile="true">
+      <rtps>
+        <defaultUnicastLocatorList>
+          <locator>
+            <udpv4>
+              <address>127.0.0.1</address>
+              <port>0</port>
+            </udpv4>
+          </locator>
+        </defaultUnicastLocatorList>
+
+        <builtin>
+          <metatrafficUnicastLocatorList>
+            <locator>
+              <udpv4>
+                <address>127.0.0.1</address>
+                <port>0</port>
+              </udpv4>
+            </locator>
+          </metatrafficUnicastLocatorList>
+
+          <initialPeersList>
+            <locator>
+              <udpv4>
+                <address>127.0.0.1</address>
+                <port>0</port>
+              </udpv4>
+            </locator>
+          </initialPeersList>
+        </builtin>
+      </rtps>
+    </participant>
+  </profiles>
+</dds>

{rosotacom_dev-2.5.dev55 → rosotacom_dev-2.5.dev56}/src/rosotacom/resources/ws/ota_configs/get_ota_xml.py

@@ -12,6 +12,8 @@ get_fastdds_easy_mode_xml.py, get_cyclonedds_xml.py).
 import argparse
 import os
 import re
+import sys
+import tempfile
 
 
 KNOWN_PLACEHOLDERS = {"#host_ip", "#peer", "#easy_mode_ip", "#spdp_interval"}
@@ -204,6 +206,16 @@ if __name__ == "__main__":
         "-e", "--easy-mode-ip", dest="easy_mode_ip",
         help="Resolved Fast DDS Easy Mode address (#easy_mode_ip).",
     )
+    parser.add_argument(
+        "-o", "--output", dest="output",
+        help=(
+            "Write the rendered XML to this path, atomically. Prefer this over a shell "
+            "redirect: `> file` truncates the target before this script runs, so a "
+            "failure here leaves a zero-byte profile behind, and a zero-byte profile is "
+            "not an error to a DDS stack — it falls back to its defaults and the link "
+            "comes up carrying nothing."
+        ),
+    )
     parser.add_argument(
         "--spdp-interval",
         dest="spdp_interval",
@@ -211,4 +223,25 @@ if __name__ == "__main__":
         help=f"CycloneDDS SPDP interval duration (#spdp_interval, default: {DEFAULT_SPDP_INTERVAL}).",
     )
     args = parser.parse_args()
-    print(main(**{k: v for k, v in vars(args).items() if v is not None}))
+    output = args.output
+    rendered = main(**{k: v for k, v in vars(args).items()
+                       if v is not None and k != "output"})
+    if not output:
+        print(rendered)
+        raise SystemExit(0)
+    if not rendered.strip():
+        raise SystemExit(f"Refusing to write an empty profile to {output}.")
+    directory = os.path.dirname(os.path.abspath(output)) or "."
+    os.makedirs(directory, exist_ok=True)
+    handle, temporary = tempfile.mkstemp(dir=directory, suffix=".partial")
+    try:
+        with os.fdopen(handle, "w", encoding="utf-8") as stream:
+            stream.write(rendered)
+        os.replace(temporary, output)
+    except BaseException:
+        try:
+            os.unlink(temporary)
+        except OSError:
+            pass
+        raise
+    print(f"Wrote {output} ({len(rendered)} bytes).", file=sys.stderr)

{rosotacom_dev-2.5.dev55 → rosotacom_dev-2.5.dev56}/src/rosotacom/resources/ws/session/content/base/session_plugin_base.yaml

@@ -292,7 +292,7 @@ common:
   before_commands:
     - if [ '${local_domain_id}' != "None" ] && [ -n '${local_domain_id}' ]; then export ROS_DOMAIN_ID='${local_domain_id}'; fi
     - if [ '${rmw_local}' != "None" ]; then case ${rmw_local} in cyclone) export RMW_IMPLEMENTATION=rmw_cyclonedds_cpp;; fastdds) export RMW_IMPLEMENTATION=rmw_fastrtps_cpp; export FASTDDS_BUILTIN_TRANSPORTS=${FASTDDS_BUILTIN_TRANSPORTS:-UDPv4};; zenoh) export RMW_IMPLEMENTATION=rmw_zenoh_cpp;; *) export RMW_IMPLEMENTATION='${rmw_local}';; esac; fi
-    - if [ '${local_config_template}' != "None" ] && [ -n '${local_config_template}' ]; then local_args=(--config "${local_config_template}" --host-ip "${ip_local}" --peer "${ip_remote}"); [ '${local_easy_mode_ip}' != "None" ] && [ -n '${local_easy_mode_ip}' ] && local_args+=(--easy-mode-ip "${local_easy_mode_ip}"); [ '${local_spdp_interval}' != "None" ] && [ -n '${local_spdp_interval}' ] && local_args+=(--spdp-interval "${local_spdp_interval}"); python3 /ws/ota_configs/get_ota_xml.py "${local_args[@]}" > "${local_config_file}"; case ${rmw_local} in cyclone) export CYCLONEDDS_URI="file://${local_config_file}";; fastdds) export FASTDDS_DEFAULT_PROFILES_FILE="${local_config_file}"; export RMW_FASTRTPS_USE_QOS_FROM_XML=1;; esac; fi
+    - if [ '${local_config_template}' != "None" ] && [ -n '${local_config_template}' ]; then local_args=(--config "${local_config_template}" --host-ip "${ip_local}" --peer "${ip_remote}"); [ '${local_easy_mode_ip}' != "None" ] && [ -n '${local_easy_mode_ip}' ] && local_args+=(--easy-mode-ip "${local_easy_mode_ip}"); [ '${local_spdp_interval}' != "None" ] && [ -n '${local_spdp_interval}' ] && local_args+=(--spdp-interval "${local_spdp_interval}"); python3 /ws/ota_configs/get_ota_xml.py "${local_args[@]}" --output "${local_config_file}" || { echo "FATAL - could not render the local DDS profile ${local_config_template}; refusing to start on defaults." >&2; exit 1; }; case ${rmw_local} in cyclone) export CYCLONEDDS_URI="file://${local_config_file}";; fastdds) export FASTDDS_DEFAULT_PROFILES_FILE="${local_config_file}"; export RMW_FASTRTPS_USE_QOS_FROM_XML=1;; esac; fi
     - if [ '${before_command}' != "None" ]; then ${before_command}; fi
 
 windows:
@@ -338,7 +338,7 @@ windows:
     if: use_domain_bridge
     splits:
       - commands:
-        - if [ '${ota_config_template}' != "None" ] && [ -n '${ota_config_template}' ]; then ota_args=(--config "${ota_config_template}" --host-ip "${ip_local}" --peer "${ip_remote}"); [ '${ota_easy_mode_ip}' != "None" ] && [ -n '${ota_easy_mode_ip}' ] && ota_args+=(--easy-mode-ip "${ota_easy_mode_ip}"); [ '${ota_spdp_interval}' != "None" ] && [ -n '${ota_spdp_interval}' ] && ota_args+=(--spdp-interval "${ota_spdp_interval}"); python3 /ws/ota_configs/get_ota_xml.py "${ota_args[@]}" > "${ota_config_file}"; case ${rmw_ota} in cyclone) export CYCLONEDDS_URI="file://${ota_config_file}";; fastdds) export FASTDDS_DEFAULT_PROFILES_FILE="${ota_config_file}"; export RMW_FASTRTPS_USE_QOS_FROM_XML=1;; esac; fi
+        - if [ '${ota_config_template}' != "None" ] && [ -n '${ota_config_template}' ]; then ota_args=(--config "${ota_config_template}" --host-ip "${ip_local}" --peer "${ip_remote}"); [ '${ota_easy_mode_ip}' != "None" ] && [ -n '${ota_easy_mode_ip}' ] && ota_args+=(--easy-mode-ip "${ota_easy_mode_ip}"); [ '${ota_spdp_interval}' != "None" ] && [ -n '${ota_spdp_interval}' ] && ota_args+=(--spdp-interval "${ota_spdp_interval}"); python3 /ws/ota_configs/get_ota_xml.py "${ota_args[@]}" --output "${ota_config_file}" || { echo "FATAL - could not render the OTA DDS profile ${ota_config_template}; refusing to start on defaults." >&2; exit 1; }; case ${rmw_ota} in cyclone) export CYCLONEDDS_URI="file://${ota_config_file}";; fastdds) export FASTDDS_DEFAULT_PROFILES_FILE="${ota_config_file}"; export RMW_FASTRTPS_USE_QOS_FROM_XML=1;; esac; fi
         - ros2 run domain_bridge domain_bridge --wait-for-publisher false ${domain_bridge_config_file}
   - name: IN
     if: in
@@ -346,7 +346,7 @@ windows:
       - commands:
         - if [ '${ota_domain_id}' != "None" ] && [ -n '${ota_domain_id}' ]; then export ROS_DOMAIN_ID='${ota_domain_id}'; fi
         - case ${rmw_ota} in cyclone) export RMW_IMPLEMENTATION=rmw_cyclonedds_cpp;; fastdds) export RMW_IMPLEMENTATION=rmw_fastrtps_cpp; export FASTDDS_BUILTIN_TRANSPORTS=${FASTDDS_BUILTIN_TRANSPORTS:-UDPv4};; zenoh) export RMW_IMPLEMENTATION=rmw_zenoh_cpp;; esac
-        - if [ '${ota_config_template}' != "None" ] && [ -n '${ota_config_template}' ]; then ota_args=(--config "${ota_config_template}" --host-ip "${ip_local}" --peer "${ip_remote}"); [ '${ota_easy_mode_ip}' != "None" ] && [ -n '${ota_easy_mode_ip}' ] && ota_args+=(--easy-mode-ip "${ota_easy_mode_ip}"); [ '${ota_spdp_interval}' != "None" ] && [ -n '${ota_spdp_interval}' ] && ota_args+=(--spdp-interval "${ota_spdp_interval}"); python3 /ws/ota_configs/get_ota_xml.py "${ota_args[@]}" > "${ota_config_file}"; case ${rmw_ota} in cyclone) export CYCLONEDDS_URI="file://${ota_config_file}";; fastdds) export FASTDDS_DEFAULT_PROFILES_FILE="${ota_config_file}"; export RMW_FASTRTPS_USE_QOS_FROM_XML=1;; esac; fi
+        - if [ '${ota_config_template}' != "None" ] && [ -n '${ota_config_template}' ]; then ota_args=(--config "${ota_config_template}" --host-ip "${ip_local}" --peer "${ip_remote}"); [ '${ota_easy_mode_ip}' != "None" ] && [ -n '${ota_easy_mode_ip}' ] && ota_args+=(--easy-mode-ip "${ota_easy_mode_ip}"); [ '${ota_spdp_interval}' != "None" ] && [ -n '${ota_spdp_interval}' ] && ota_args+=(--spdp-interval "${ota_spdp_interval}"); python3 /ws/ota_configs/get_ota_xml.py "${ota_args[@]}" --output "${ota_config_file}" || { echo "FATAL - could not render the OTA DDS profile ${ota_config_template}; refusing to start on defaults." >&2; exit 1; }; case ${rmw_ota} in cyclone) export CYCLONEDDS_URI="file://${ota_config_file}";; fastdds) export FASTDDS_DEFAULT_PROFILES_FILE="${ota_config_file}"; export RMW_FASTRTPS_USE_QOS_FROM_XML=1;; esac; fi
         # For zenoh_ros2dds the OTA domain is loopback-only (zenoh is the inter-host
         # transport), so confine discovery to localhost instead of statically peering
         # the remote host — otherwise the OTA-domain DDS double-delivers alongside zenoh.
@@ -373,7 +373,7 @@ windows:
       - commands:
         - if [ '${ota_domain_id}' != "None" ] && [ -n '${ota_domain_id}' ]; then export ROS_DOMAIN_ID='${ota_domain_id}'; fi
         - case ${rmw_ota} in cyclone) export RMW_IMPLEMENTATION=rmw_cyclonedds_cpp;; fastdds) export RMW_IMPLEMENTATION=rmw_fastrtps_cpp; export FASTDDS_BUILTIN_TRANSPORTS=${FASTDDS_BUILTIN_TRANSPORTS:-UDPv4};; zenoh) export RMW_IMPLEMENTATION=rmw_zenoh_cpp;; esac
-        - if [ '${ota_config_template}' != "None" ] && [ -n '${ota_config_template}' ]; then ota_args=(--config "${ota_config_template}" --host-ip "${ip_local}" --peer "${ip_remote}"); [ '${ota_easy_mode_ip}' != "None" ] && [ -n '${ota_easy_mode_ip}' ] && ota_args+=(--easy-mode-ip "${ota_easy_mode_ip}"); [ '${ota_spdp_interval}' != "None" ] && [ -n '${ota_spdp_interval}' ] && ota_args+=(--spdp-interval "${ota_spdp_interval}"); python3 /ws/ota_configs/get_ota_xml.py "${ota_args[@]}" > "${ota_config_file}"; case ${rmw_ota} in cyclone) export CYCLONEDDS_URI="file://${ota_config_file}";; fastdds) export FASTDDS_DEFAULT_PROFILES_FILE="${ota_config_file}"; export RMW_FASTRTPS_USE_QOS_FROM_XML=1;; esac; fi
+        - if [ '${ota_config_template}' != "None" ] && [ -n '${ota_config_template}' ]; then ota_args=(--config "${ota_config_template}" --host-ip "${ip_local}" --peer "${ip_remote}"); [ '${ota_easy_mode_ip}' != "None" ] && [ -n '${ota_easy_mode_ip}' ] && ota_args+=(--easy-mode-ip "${ota_easy_mode_ip}"); [ '${ota_spdp_interval}' != "None" ] && [ -n '${ota_spdp_interval}' ] && ota_args+=(--spdp-interval "${ota_spdp_interval}"); python3 /ws/ota_configs/get_ota_xml.py "${ota_args[@]}" --output "${ota_config_file}" || { echo "FATAL - could not render the OTA DDS profile ${ota_config_template}; refusing to start on defaults." >&2; exit 1; }; case ${rmw_ota} in cyclone) export CYCLONEDDS_URI="file://${ota_config_file}";; fastdds) export FASTDDS_DEFAULT_PROFILES_FILE="${ota_config_file}"; export RMW_FASTRTPS_USE_QOS_FROM_XML=1;; esac; fi
         # For zenoh_ros2dds the OTA domain is loopback-only (zenoh is the inter-host
         # transport), so confine discovery to localhost instead of statically peering
         # the remote host — otherwise the OTA-domain DDS double-delivers alongside zenoh.

{rosotacom_dev-2.5.dev55 → rosotacom_dev-2.5.dev56}/src/rosotacom/resources/ws/session/creation/generate_session_files.py

@@ -231,6 +231,24 @@ _DDS_OTA_DEFAULT_CONFIG = {
     "fastdds": "fastdds_tuned.xml",
 }
 
+# The *local* domain needs a default only for Fast DDS, and only because the two
+# stacks differ in what they do with no configuration at all. CycloneDDS talks
+# between containers on a host out of the box; Fast DDS announces every
+# interface it can enumerate, a developer machine has dozens of down `br-*`
+# bridges, and the peer can settle on one that carries nothing. Measured
+# 2026-08-19 on two containers sharing a network namespace: 0 of 80 samples
+# without a profile, 79 of 80 once the loopback locators are named. The failure
+# is quiet — the writer reports a matched reader and the pipeline's last stage
+# sits IDLE with pub=1 — so this is a default rather than something a session
+# has to know to ask for.
+#
+# Cyclone deliberately has no entry: cyclonedds_local_participants.xml raises
+# the participant-index ceiling, which is a control-centre concern, not
+# something every session should inherit.
+_DDS_LOCAL_DEFAULT_CONFIG = {
+    "fastdds": "fastdds_local.xml",
+}
+
 
 def _validate_positive_seconds(value: str, ctx: str) -> None:
     text = value.strip().lower()
@@ -2207,8 +2225,9 @@ def func(
         items: List[Tuple[str, Any]] = []
         if effective_impl is not None:
             items.append(("rmw_local", effective_impl))
-        if side.dds_config:
-            items.append(("local_config_template", side.dds_config))
+        local_dds_config = side.dds_config or _DDS_LOCAL_DEFAULT_CONFIG.get(effective_impl or "")
+        if local_dds_config:
+            items.append(("local_config_template", local_dds_config))
             items.append(("local_config_file", "${peer_dir}/local_dds.xml"))
             if side.impl == "fastdds" and side.dds_config == "fastdds_easy_mode.xml":
                 items.append(

{rosotacom_dev-2.5.dev55 → rosotacom_dev-2.5.dev56}/src/rosotacom_dev.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: rosotacom-dev
-Version: 2.5.dev55
+Version: 2.5.dev56
 Summary: Host CLI for ROS OTA communication sessions.
 Author: develNor
 License-Expression: BSD-3-Clause

{rosotacom_dev-2.5.dev55 → rosotacom_dev-2.5.dev56}/src/rosotacom_dev.egg-info/SOURCES.txt

@@ -207,6 +207,7 @@ src/rosotacom/resources/ws/ota_configs/cyclonedds_local_participants.xml.templat
 src/rosotacom/resources/ws/ota_configs/cyclonedds_minimal.xml.template
 src/rosotacom/resources/ws/ota_configs/cyclonedds_tuned.xml.template
 src/rosotacom/resources/ws/ota_configs/fastdds_easy_mode.xml.template
+src/rosotacom/resources/ws/ota_configs/fastdds_local.xml.template
 src/rosotacom/resources/ws/ota_configs/fastdds_tuned.xml.template
 src/rosotacom/resources/ws/ota_configs/fastdds_unicast.xml.template
 src/rosotacom/resources/ws/ota_configs/get_ota_xml.py

{rosotacom_dev-2.5.dev55 → rosotacom_dev-2.5.dev56}/src/rosotacom_dev.egg-info/scm_file_list.json

@@ -210,6 +210,7 @@
     "src/rosotacom/resources/ws/ota_configs/cyclonedds_minimal.xml.template",
     "src/rosotacom/resources/ws/ota_configs/cyclonedds_tuned.xml.template",
     "src/rosotacom/resources/ws/ota_configs/fastdds_easy_mode.xml.template",
+    "src/rosotacom/resources/ws/ota_configs/fastdds_local.xml.template",
     "src/rosotacom/resources/ws/ota_configs/fastdds_tuned.xml.template",
     "src/rosotacom/resources/ws/ota_configs/fastdds_unicast.xml.template",
     "src/rosotacom/resources/ws/ota_configs/get_ota_xml.py",

{rosotacom_dev-2.5.dev55 → rosotacom_dev-2.5.dev56}/src/rosotacom_dev.egg-info/scm_version.json

@@ -1,7 +1,7 @@
 {
   "tag": "2.4",
-  "distance": 55,
-  "node": "gd772be38947b2d3b6d4fc5d45d9fa57da1e1723a",
+  "distance": 56,
+  "node": "g18a62c4bee965c3d64b7eaa70544f6bb7f078a98",
   "dirty": false,
   "branch": "HEAD",
   "node_date": "2026-08-19"

{rosotacom_dev-2.5.dev55 → rosotacom_dev-2.5.dev56}/tests/contract/test_packaged_resources.py

@@ -437,3 +437,42 @@ def test_no_credential_path_is_tracked_and_no_symlink_escapes_the_repository() -
         f"tracked symlinks point outside the repository: {escaping}. Checking one out overwrites "
         "whatever is at that path, and it is shipped in the sdist."
     )
+
+
+def test_dds_profiles_are_rendered_atomically_and_never_left_empty() -> None:
+    """A profile that fails to render must not leave a usable-looking file.
+
+    `get_ota_xml.py ... > file` truncates the target before the renderer runs, so
+    a missing template leaves a zero-byte profile behind — and a zero-byte
+    profile is not an error to either DDS stack. Fast DDS logs a parse error and
+    falls back to multicast discovery, which a tunnelled OTA link cannot carry,
+    so the session comes up healthy-looking and delivers nothing. That is how
+    eight runs on 2026-08-19 were recorded as "Fast DDS delivers 0 messages"
+    when the middleware had simply never been configured.
+    """
+    plugin_path = cli.WS_DIR / "session" / "content" / "base" / "session_plugin_base.yaml"
+    text = plugin_path.read_text(encoding="utf-8")
+
+    # Not [^;]* — the failure branch itself contains a ';', and stopping there
+    # would silently check only the half of the statement that cannot fail.
+    calls = [text[m.start() : m.start() + 400] for m in re.finditer(r"get_ota_xml\.py", text)]
+    assert calls, "no get_ota_xml.py call sites found — did the plugin move?"
+    for call in calls:
+        assert "--output" in call, f"renders through a shell redirect: {call[:120]}"
+        assert not re.search(r">\s*\"\$\{\w*config_file\}\"", call), f"still truncates its target: {call[:120]}"
+        assert "exit 1" in call, f"a failed render does not stop the start: {call[:120]}"
+
+
+def test_every_default_ota_profile_template_is_packaged() -> None:
+    """The default a session falls back to must exist in the installed package.
+
+    #299 pointed the Fast DDS default at fastdds_tuned.xml while the stack
+    deployed on the bench machines predated that template, so every Fast DDS run
+    rendered nothing.
+    """
+    from rosotacom.resources.ws.session.creation import generate_session_files as gsf
+
+    ota_configs = cli.WS_DIR / "ota_configs"
+    defaults = list(gsf._DDS_OTA_DEFAULT_CONFIG.values()) + list(gsf._DDS_LOCAL_DEFAULT_CONFIG.values())
+    missing = [name for name in defaults if name and not (ota_configs / f"{name}.template").exists()]
+    assert not missing, f"default profile templates absent from the package: {missing}"

{rosotacom_dev-2.5.dev55 → rosotacom_dev-2.5.dev56}/tests/unit/test_concurrency.py

@@ -8,6 +8,8 @@ parallel/conflict end-to-end half lives in tests/e2e/test_parallel_smoke.py.
 from __future__ import annotations
 
 import argparse
+import os
+import signal
 import subprocess
 from pathlib import Path
 
@@ -604,3 +606,99 @@ def test_local_benchmark_aborts_on_any_active_rosotacom_container(
 def test_local_benchmark_passes_on_quiet_host(monkeypatch: pytest.MonkeyPatch) -> None:
     monkeypatch.setattr(rosotacom, "_list_docker_containers", lambda all_states=False: [("onedrive", ["bridge"])])
     cli_benchmark._abort_on_local_benchmark_conflicts(argparse.Namespace())
+
+
+def _load_plan(tmp_path: Path) -> rosotacom.OtaSmokePlan:
+    return rosotacom.OtaSmokePlan(
+        state_path=tmp_path / "ota-deployment.yaml",
+        workdir="/tmp/rosotacom_ota",
+        rosotacom="rosotacom",
+        project="rosotacom.yaml",
+        peers={"a": rosotacom.OtaSmokePeer("a", None, "10.0.0.10")},
+    )
+
+
+def test_ota_load_check_refuses_a_peer_someone_else_is_saturating(
+    tmp_path: Path,
+    monkeypatch: pytest.MonkeyPatch,
+) -> None:
+    """A busy peer must stop the run, not quietly distort it.
+
+    Measured on majestic on 2026-08-19: load 78 on 32 CPUs while a colleague
+    trained models. Runs there did not fail — they completed and reported
+    numbers that described the contention.
+    """
+    monkeypatch.setattr(
+        rosotacom,
+        "_ota_run",
+        _fake_ota_run_factory({"/proc/loadavg": "78.24 81.86 81.28\n32\n"}),
+    )
+
+    with pytest.raises(RuntimeError) as excinfo:
+        rosotacom._ota_load_check(_load_plan(tmp_path), dry_run=False)
+
+    message = str(excinfo.value)
+    assert "unrelated load" in message
+    assert "--skip-conflict-check" in message
+
+
+def test_ota_load_check_warns_but_proceeds_on_moderate_load(
+    tmp_path: Path,
+    monkeypatch: pytest.MonkeyPatch,
+    capsys: pytest.CaptureFixture[str],
+) -> None:
+    monkeypatch.setattr(
+        rosotacom,
+        "_ota_run",
+        _fake_ota_run_factory({"/proc/loadavg": "9.0 8.0 7.0\n8\n"}),
+    )
+    rosotacom._ota_load_check(_load_plan(tmp_path), dry_run=False)
+    out = capsys.readouterr().out
+    assert "interleaved" in out
+
+
+def test_ota_load_check_is_quiet_on_an_idle_peer(
+    tmp_path: Path,
+    monkeypatch: pytest.MonkeyPatch,
+    capsys: pytest.CaptureFixture[str],
+) -> None:
+    """The reading is printed even when it passes: a result should carry the
+    conditions it was measured under, not only the ones that stopped it."""
+    monkeypatch.setattr(
+        rosotacom,
+        "_ota_run",
+        _fake_ota_run_factory({"/proc/loadavg": "0.21 0.08 0.02\n8\n"}),
+    )
+    rosotacom._ota_load_check(_load_plan(tmp_path), dry_run=False)
+    out = capsys.readouterr().out
+    assert "0.21" in out and "8 CPUs" in out
+
+
+def test_sigterm_unwinds_so_cleanup_runs() -> None:
+    """SIGTERM must reach the run's `finally`, not end the process outright.
+
+    `timeout`, systemd, and a cancelled CI step all send SIGTERM. Under the
+    default disposition Python exits immediately, so the OTA run's cleanup —
+    revert shaping, collect logs, stop both peers — never happens. On
+    2026-08-19 a run killed by `timeout` left containers on both peers, which
+    then blocked the next run's exclusivity check.
+    """
+    cleaned: list[str] = []
+
+    with pytest.raises(SystemExit) as excinfo:
+        with rosotacom._terminate_via_exception():
+            try:
+                os.kill(os.getpid(), signal.SIGTERM)
+            finally:
+                cleaned.append("teardown ran")
+
+    assert cleaned == ["teardown ran"]
+    assert excinfo.value.code == 128 + signal.SIGTERM
+
+
+def test_sigterm_disposition_is_restored() -> None:
+    """The handler must not leak into whatever called us."""
+    before = signal.getsignal(signal.SIGTERM)
+    with rosotacom._terminate_via_exception():
+        assert signal.getsignal(signal.SIGTERM) is not before
+    assert signal.getsignal(signal.SIGTERM) is before