Add trace tool. Closes #130.

This commit is contained in:
Jack Kingsman
2026-03-30 19:26:12 -07:00
parent eb1f7ae638
commit 134e8d0d29
19 changed files with 1625 additions and 21 deletions
+53
View File
@@ -628,6 +628,59 @@ class TraceResponse(BaseModel):
path_len: int = Field(description="Number of hops in the trace path")
class RadioTraceHopRequest(BaseModel):
"""One requested hop in a radio trace path."""
public_key: str | None = Field(
default=None,
description="Full repeater public key when this hop maps to a known repeater",
)
hop_hex: str | None = Field(
default=None,
description="Raw hop hash hex when using a custom repeater prefix",
)
class RadioTraceRequest(BaseModel):
"""Ordered trace path for a radio trace loop."""
hop_hash_bytes: Literal[1, 2, 4] = Field(
default=4,
description="Hash width in bytes for every hop in this trace path",
)
hops: list[RadioTraceHopRequest] = Field(
min_length=1,
description="Ordered repeater hops, using either known repeater keys or custom hop hex",
)
class RadioTraceNode(BaseModel):
"""One resolved node in a radio trace result."""
role: Literal["repeater", "custom", "local"] = Field(description="Node role in the trace")
public_key: str | None = Field(
default=None,
description="Resolved full public key for this node when known",
)
name: str | None = Field(default=None, description="Display name for this node when known")
observed_hash: str | None = Field(
default=None,
description="Observed 4-byte trace hash for this node as hex",
)
snr: float | None = Field(default=None, description="Reported SNR for this node in dB")
class RadioTraceResponse(BaseModel):
"""Resolved multi-hop radio trace result."""
path_len: int = Field(description="Number of hashed nodes returned by the trace response")
timeout_seconds: float = Field(description="Timeout window used while waiting for the trace")
nodes: list[RadioTraceNode] = Field(
default_factory=list,
description="Ordered trace nodes: repeater hops followed by the terminal local radio",
)
class PathDiscoveryRoute(BaseModel):
"""One resolved route returned by contact path discovery."""
+206
View File
@@ -2,6 +2,7 @@ import asyncio
import logging
import random
import time
from contextlib import suppress
from typing import Literal, TypeAlias
from fastapi import APIRouter, HTTPException
@@ -10,10 +11,15 @@ from pydantic import BaseModel, Field
from app.dependencies import require_connected
from app.models import (
CONTACT_TYPE_REPEATER,
ContactUpsert,
RadioDiscoveryRequest,
RadioDiscoveryResponse,
RadioDiscoveryResult,
RadioTraceHopRequest,
RadioTraceNode,
RadioTraceRequest,
RadioTraceResponse,
)
from app.radio_sync import send_advertisement as do_send_advertisement
from app.radio_sync import sync_radio_time
@@ -45,6 +51,12 @@ _DISCOVERY_NODE_TYPES: dict[int, DiscoveryNodeType] = {
2: "repeater",
4: "sensor",
}
TRACE_WAIT_TIMEOUT_SECONDS = 45.0
TRACE_DEFAULT_TIMEOUT_SECONDS = 15.0
TRACE_TIMEOUT_MIN_SECONDS = 5.0
TRACE_TIMEOUT_MAX_SECONDS = 30.0
TRACE_TIMEOUT_MARGIN = 1.2
TRACE_HASH_FLAGS = {1: 0, 2: 1, 4: 2}
async def _prepare_connected(*, broadcast_on_success: bool) -> bool:
@@ -217,6 +229,101 @@ async def _attach_known_names(results: list[RadioDiscoveryResult]) -> None:
result.name = contact.name
def _trace_hash_for_key(public_key: str, hop_hash_bytes: int) -> str:
return public_key[: hop_hash_bytes * 2].lower()
def _trace_timeout_seconds(send_result: object) -> float:
payload = getattr(send_result, "payload", None) or {}
suggested_timeout = payload.get("suggested_timeout")
try:
if suggested_timeout is None:
raise TypeError
timeout_seconds = float(suggested_timeout) / 1000.0 * TRACE_TIMEOUT_MARGIN
except (TypeError, ValueError):
timeout_seconds = TRACE_DEFAULT_TIMEOUT_SECONDS
return max(TRACE_TIMEOUT_MIN_SECONDS, min(TRACE_TIMEOUT_MAX_SECONDS, timeout_seconds))
async def _resolve_trace_hops(
hops: list[RadioTraceHopRequest], hop_hash_bytes: int
) -> tuple[list[RadioTraceNode], list[str]]:
trace_nodes: list[RadioTraceNode] = []
requested_hashes: list[str] = []
expected_hex_len = hop_hash_bytes * 2
for hop in hops:
public_key = hop.public_key.strip().lower() if isinstance(hop.public_key, str) else None
hop_hex = hop.hop_hex.strip().lower() if isinstance(hop.hop_hex, str) else None
if bool(public_key) == bool(hop_hex):
raise HTTPException(
status_code=400,
detail="Each trace hop must provide exactly one of public_key or hop_hex",
)
if public_key:
if len(public_key) != 64:
raise HTTPException(
status_code=400,
detail="Trace repeater keys must be full 64-character public keys",
)
try:
bytes.fromhex(public_key)
except ValueError as exc:
raise HTTPException(
status_code=400,
detail="Trace repeater keys must be valid hex public keys",
) from exc
contact = await ContactRepository.get_by_key(public_key)
if contact is None:
raise HTTPException(
status_code=404, detail=f"Trace repeater not found: {public_key}"
)
if contact.type != CONTACT_TYPE_REPEATER:
raise HTTPException(
status_code=400,
detail=f"Trace node is not a repeater: {public_key[:12]}",
)
requested_hashes.append(_trace_hash_for_key(contact.public_key, hop_hash_bytes))
trace_nodes.append(
RadioTraceNode(
role="repeater",
public_key=contact.public_key,
name=contact.name,
observed_hash=None,
snr=None,
)
)
continue
assert hop_hex is not None
if len(hop_hex) != expected_hex_len:
raise HTTPException(
status_code=400,
detail=f"Custom trace hops must be exactly {expected_hex_len} hex characters",
)
try:
bytes.fromhex(hop_hex)
except ValueError as exc:
raise HTTPException(
status_code=400,
detail="Custom trace hops must be valid hex",
) from exc
requested_hashes.append(hop_hex)
trace_nodes.append(
RadioTraceNode(
role="custom",
public_key=None,
name=None,
observed_hash=hop_hex,
snr=None,
)
)
return trace_nodes, requested_hashes
@router.get("/config", response_model=RadioConfigResponse)
async def get_radio_config() -> RadioConfigResponse:
"""Get the current radio configuration."""
@@ -388,6 +495,105 @@ async def discover_mesh(request: RadioDiscoveryRequest) -> RadioDiscoveryRespons
)
@router.post("/trace", response_model=RadioTraceResponse)
async def trace_path(request: RadioTraceRequest) -> RadioTraceResponse:
"""Send a multi-hop trace loop through known repeaters and back to the local radio."""
require_connected()
trace_nodes, requested_hashes = await _resolve_trace_hops(request.hops, request.hop_hash_bytes)
tag = random.randint(1, 0xFFFFFFFF)
trace_flags = TRACE_HASH_FLAGS[request.hop_hash_bytes]
async with radio_manager.radio_operation("radio_trace", pause_polling=True) as mc:
local_public_key = str((mc.self_info or {}).get("public_key") or "").lower()
if len(local_public_key) != 64:
raise HTTPException(status_code=503, detail="Local radio public key is unavailable")
local_name = (mc.self_info or {}).get("name")
response_task = asyncio.create_task(
mc.wait_for_event(
EventType.TRACE_DATA,
attribute_filters={"tag": tag},
timeout=TRACE_WAIT_TIMEOUT_SECONDS,
)
)
try:
send_result = await mc.commands.send_trace(
path=",".join(requested_hashes),
tag=tag,
flags=trace_flags,
)
if send_result is None or send_result.type == EventType.ERROR:
raise HTTPException(status_code=500, detail="Failed to send trace")
timeout_seconds = _trace_timeout_seconds(send_result)
try:
event = await asyncio.wait_for(response_task, timeout=timeout_seconds)
except asyncio.TimeoutError as exc:
raise HTTPException(status_code=504, detail="No trace response heard") from exc
finally:
if not response_task.done():
response_task.cancel()
with suppress(asyncio.CancelledError):
await response_task
if event is None:
raise HTTPException(status_code=504, detail="No trace response heard")
payload = event.payload if isinstance(event.payload, dict) else {}
path_len = payload.get("path_len")
if not isinstance(path_len, int):
raise HTTPException(status_code=500, detail="Trace response was malformed")
raw_path = payload.get("path")
path_nodes = raw_path if isinstance(raw_path, list) else []
final_local_node = (
path_nodes[-1]
if path_nodes
and isinstance(path_nodes[-1], dict)
and not isinstance(path_nodes[-1].get("hash"), str)
else None
)
hashed_nodes = path_nodes[:-1] if final_local_node is not None else path_nodes
if len(hashed_nodes) < len(trace_nodes):
raise HTTPException(status_code=500, detail="Trace response was incomplete")
nodes: list[RadioTraceNode] = []
for index, trace_node in enumerate(trace_nodes):
observed = hashed_nodes[index] if index < len(hashed_nodes) else {}
observed_hash = observed.get("hash") if isinstance(observed, dict) else None
observed_snr = observed.get("snr") if isinstance(observed, dict) else None
nodes.append(
RadioTraceNode(
role=trace_node.role,
public_key=trace_node.public_key,
name=trace_node.name,
observed_hash=(
observed_hash if isinstance(observed_hash, str) else trace_node.observed_hash
),
snr=float(observed_snr) if isinstance(observed_snr, (int, float)) else None,
)
)
terminal_snr_value = final_local_node.get("snr") if isinstance(final_local_node, dict) else None
nodes.append(
RadioTraceNode(
role="local",
public_key=local_public_key,
name=local_name if isinstance(local_name, str) and local_name else None,
observed_hash=None,
snr=float(terminal_snr_value) if isinstance(terminal_snr_value, (int, float)) else None,
)
)
return RadioTraceResponse(
path_len=path_len,
timeout_seconds=timeout_seconds,
nodes=nodes,
)
async def _attempt_reconnect() -> dict:
"""Shared reconnection logic for reboot and reconnect endpoints."""
radio_manager.resume_connection()