Files
DECNET/decnet/web/router/stream/api_stream_events.py
anti 32340bea0d perf: migrate hot-path JSON serialization to orjson
stdlib json was FastAPI's default. Every response body, every SSE frame,
and every add_log/state/payload write paid the stdlib encode cost.

- pyproject.toml: add orjson>=3.10 as a core dep.
- decnet/web/api.py: default_response_class=ORJSONResponse on the
  FastAPI app, so every endpoint return goes through orjson without
  touching call sites. Explicit JSONResponse sites in the validation
  exception handlers migrated to ORJSONResponse for consistency.
- health endpoint's explicit JSONResponse → ORJSONResponse.
- SSE stream (api_stream_events.py): 6 json.dumps call sites →
  orjson.dumps(...).decode() — the per-event frames that fire on every
  sse tick.
- sqlmodel_repo.py: encode sites on the log-insert path switched to
  orjson (fields, payload, state value). Parser sites (json.loads)
  left as-is for now — not on the measured hot path.
2026-04-17 15:07:28 -04:00

148 lines
5.8 KiB
Python

import asyncio
import orjson
from typing import AsyncGenerator, Optional
from fastapi import APIRouter, Depends, Query, Request
from fastapi.responses import StreamingResponse
from decnet.env import DECNET_DEVELOPER
from decnet.logging import get_logger
from decnet.telemetry import traced as _traced, get_tracer as _get_tracer
from decnet.web.dependencies import require_stream_viewer, repo
log = get_logger("api")
router = APIRouter()
def _build_trace_links(logs: list[dict]) -> list:
"""Build OTEL span links from persisted trace_id/span_id in log rows.
Returns an empty list when tracing is disabled (no OTEL imports).
"""
try:
from opentelemetry.trace import Link, SpanContext, TraceFlags
except ImportError:
return []
links: list[Link] = []
for entry in logs:
tid = entry.get("trace_id")
sid = entry.get("span_id")
if not tid or not sid or tid == "0":
continue
try:
ctx = SpanContext(
trace_id=int(tid, 16),
span_id=int(sid, 16),
is_remote=True,
trace_flags=TraceFlags(TraceFlags.SAMPLED),
)
links.append(Link(ctx))
except (ValueError, TypeError):
continue
return links
@router.get("/stream", tags=["Observability"],
responses={
200: {
"content": {"text/event-stream": {}},
"description": "Real-time Server-Sent Events (SSE) stream"
},
401: {"description": "Could not validate credentials"},
403: {"description": "Insufficient permissions"},
422: {"description": "Validation error"}
},
)
@_traced("api.stream_events")
async def stream_events(
request: Request,
last_event_id: int = Query(0, alias="lastEventId"),
search: Optional[str] = None,
start_time: Optional[str] = None,
end_time: Optional[str] = None,
max_output: Optional[int] = Query(None, alias="maxOutput"),
user: dict = Depends(require_stream_viewer)
) -> StreamingResponse:
# Prefetch the initial snapshot before entering the streaming generator.
# With asyncmy (pure async TCP I/O), the first DB await inside the generator
# fires immediately after the ASGI layer sends the keepalive chunk — the HTTP
# write and the MySQL read compete for asyncio I/O callbacks and the MySQL
# callback can stall. Running these here (normal async context, no streaming)
# avoids that race entirely. aiosqlite is immune because it runs SQLite in a
# thread, decoupled from the event loop's I/O scheduler.
_start_id = last_event_id if last_event_id != 0 else await repo.get_max_log_id()
_initial_stats = await repo.get_stats_summary()
_initial_histogram = await repo.get_log_histogram(
search=search, start_time=start_time, end_time=end_time, interval_minutes=15,
)
async def event_generator() -> AsyncGenerator[str, None]:
last_id = _start_id
stats_interval_sec = 10
loops_since_stats = 0
emitted_chunks = 0
try:
yield ": keepalive\n\n" # flush headers immediately
# Emit pre-fetched initial snapshot — no DB calls in generator until the loop
yield f"event: message\ndata: {orjson.dumps({'type': 'stats', 'data': _initial_stats}).decode()}\n\n"
yield f"event: message\ndata: {orjson.dumps({'type': 'histogram', 'data': _initial_histogram}).decode()}\n\n"
while True:
if DECNET_DEVELOPER and max_output is not None:
emitted_chunks += 1
if emitted_chunks > max_output:
log.debug("Developer mode: max_output reached (%d), closing stream", max_output)
break
if await request.is_disconnected():
break
new_logs = await repo.get_logs_after_id(
last_id, limit=50, search=search,
start_time=start_time, end_time=end_time,
)
if new_logs:
last_id = max(entry["id"] for entry in new_logs)
# Create a span linking back to the ingestion traces
# stored in each log row, closing the pipeline gap.
_links = _build_trace_links(new_logs)
_tracer = _get_tracer("sse")
with _tracer.start_as_current_span(
"sse.emit_logs", links=_links,
attributes={"log_count": len(new_logs)},
):
yield f"event: message\ndata: {orjson.dumps({'type': 'logs', 'data': new_logs}).decode()}\n\n"
loops_since_stats = stats_interval_sec
if loops_since_stats >= stats_interval_sec:
stats = await repo.get_stats_summary()
yield f"event: message\ndata: {orjson.dumps({'type': 'stats', 'data': stats}).decode()}\n\n"
histogram = await repo.get_log_histogram(
search=search, start_time=start_time,
end_time=end_time, interval_minutes=15,
)
yield f"event: message\ndata: {orjson.dumps({'type': 'histogram', 'data': histogram}).decode()}\n\n"
loops_since_stats = 0
loops_since_stats += 1
await asyncio.sleep(1)
except asyncio.CancelledError:
pass
except Exception:
log.exception("SSE stream error for user %s", last_event_id)
yield f"event: error\ndata: {orjson.dumps({'type': 'error', 'message': 'Stream interrupted'}).decode()}\n\n"
return StreamingResponse(
event_generator(),
media_type="text/event-stream",
headers={
"Cache-Control": "no-cache",
"X-Accel-Buffering": "no",
},
)