Refactor exporter into modules

This commit is contained in:
2026-07-09 10:21:37 +00:00
parent dc8aa26893
commit 43be027b5d
9 changed files with 821 additions and 720 deletions
+3
View File
@@ -0,0 +1,3 @@
from .api import app
__all__ = ["app"]
+13
View File
@@ -0,0 +1,13 @@
from __future__ import annotations
from fastapi import FastAPI
from fastapi.responses import PlainTextResponse
from .metrics import render_metrics
app = FastAPI(title="openclaw-usage-exporter")
@app.get("/metrics", response_class=PlainTextResponse)
def metrics() -> str:
return render_metrics()
+171
View File
@@ -0,0 +1,171 @@
from __future__ import annotations
import json
from datetime import datetime, timedelta
from pathlib import Path
from typing import Iterable
from .config import HISTORY_DAYS, HISTORY_HOURS, SESSIONS_DIR, SESSIONS_INDEX_PATH, TIMEZONE, USAGE_COST_CACHE_PATH
from .models import SessionIndexReport, UsageCostCacheReport, UsageReport
from .utils import as_int, open_session_file, sanitize_label, now_in_tz, session_id_from_path
def iter_session_paths() -> Iterable[Path]:
for path in sorted(SESSIONS_DIR.glob("*.jsonl*")):
if ".trajectory." in path.name:
continue
yield path
def collect_usage() -> UsageReport:
report = UsageReport()
now = now_in_tz()
today = now.date()
day_cutoff = today - timedelta(days=HISTORY_DAYS - 1)
hour_cutoff = now.replace(minute=0, second=0, microsecond=0) - timedelta(hours=HISTORY_HOURS - 1)
for path in iter_session_paths():
report.files_scanned += 1
session_id = session_id_from_path(path)
try:
with open_session_file(path) as handle:
for line in handle:
report.lines_scanned += 1
if not line.strip():
continue
try:
record = json.loads(line)
except json.JSONDecodeError:
report.parse_errors += 1
continue
message = record.get("message") or {}
timestamp_ms = message.get("timestamp", record.get("timestamp"))
if not isinstance(timestamp_ms, (int, float)):
continue
dt = datetime.fromtimestamp(timestamp_ms / 1000, TIMEZONE)
usage_date = dt.date()
is_today = usage_date == today
if record.get("type") != "message":
report.non_message_records_all_time += 1
if is_today:
report.non_message_records_today += 1
continue
report.message_records_all_time += 1
if is_today:
report.message_records_today += 1
role = sanitize_label(message.get("role"))
model = sanitize_label(message.get("model"))
provider = sanitize_label(message.get("provider"))
stop_reason = sanitize_label(message.get("stopReason"))
tool_name = sanitize_label(message.get("toolName"))
error_type = sanitize_label(message.get("errorType"))
is_error = bool(message.get("isError"))
usage = message.get("usage")
if is_today:
report.role_counts_today[role] += 1
if model != "unknown":
report.unique_models_today.add(model)
if provider != "unknown":
report.unique_providers_today.add(provider)
if session_id:
report.unique_session_ids_today.add(session_id)
report.session_message_counts_today[(session_id, role)] += 1
if stop_reason != "unknown":
report.stop_reason_counts_today[(stop_reason, model, provider)] += 1
if tool_name != "unknown":
report.tool_call_counts_today[tool_name] += 1
if session_id:
report.session_tool_call_counts_today[(session_id, tool_name)] += 1
if is_error:
report.error_type_counts_today[error_type] += 1
if session_id:
report.session_error_counts_today[session_id] += 1
if not isinstance(usage, dict):
continue
total_tokens, input_tokens, output_tokens, _, _, reasoning_tokens, skipped_negative_costs = report.all_time.add_usage(usage)
if skipped_negative_costs:
report.negative_cost_entries_skipped += 1
for cost_field, raw_value in skipped_negative_costs.items():
report.negative_cost_fields_skipped[cost_field] += 1
report.negative_cost_amounts_skipped[cost_field] += abs(raw_value)
report.tokens_by_model_all_time[(model, provider)] += total_tokens
report.input_tokens_by_model_all_time[(model, provider)] += input_tokens
report.output_tokens_by_model_all_time[(model, provider)] += output_tokens
if session_id:
report.session_tokens_all_time[session_id] += total_tokens
if usage_date >= day_cutoff:
report.day_buckets[usage_date.isoformat()].add_usage(usage)
hour_bucket = dt.replace(minute=0, second=0, microsecond=0)
if hour_bucket >= hour_cutoff:
report.hour_buckets[hour_bucket.isoformat()].add_usage(usage)
if is_today:
report.today.add_usage(usage)
report.tokens_by_model_today[(model, provider)] += total_tokens
report.input_tokens_by_model_today[(model, provider)] += input_tokens
report.output_tokens_by_model_today[(model, provider)] += output_tokens
report.reasoning_tokens_by_model_today[(model, provider)] += reasoning_tokens
if session_id:
report.session_tokens_today[session_id] += total_tokens
report.session_input_tokens_today[session_id] += input_tokens
report.session_output_tokens_today[session_id] += output_tokens
except OSError:
report.parse_errors += 1
continue
return report
def collect_session_index() -> SessionIndexReport:
report = SessionIndexReport()
try:
data = json.loads(SESSIONS_INDEX_PATH.read_text(encoding="utf-8"))
except (OSError, json.JSONDecodeError):
return report
if not isinstance(data, dict):
return report
if isinstance(data.get("sessions"), list):
sessions = [session for session in data.get("sessions", []) if isinstance(session, dict)]
report.visible_count = as_int(data.get("count")) or len(sessions)
report.total_count = as_int(data.get("totalCount")) or len(sessions)
report.has_more = bool(data.get("hasMore"))
report.sessions = sessions
return report
sessions = [session for session in data.values() if isinstance(session, dict)]
report.visible_count = len(sessions)
report.total_count = len(sessions)
report.has_more = False
report.sessions = sessions
return report
def collect_usage_cost_cache() -> UsageCostCacheReport:
report = UsageCostCacheReport()
try:
data = json.loads(USAGE_COST_CACHE_PATH.read_text(encoding="utf-8"))
except (OSError, json.JSONDecodeError):
return report
if not isinstance(data, dict):
return report
files = data.get("files")
report.version = as_int(data.get("version")) if data.get("version") is not None else None
report.updated_at_ms = as_int(data.get("updatedAt")) if data.get("updatedAt") is not None else None
report.files_tracked = len(files) if isinstance(files, dict) else 0
return report
+12
View File
@@ -0,0 +1,12 @@
from __future__ import annotations
import os
from pathlib import Path
from zoneinfo import ZoneInfo
SESSIONS_DIR = Path(os.environ.get("OPENCLAW_SESSIONS_DIR", "/root/.openclaw/agents/main/sessions"))
TIMEZONE = ZoneInfo(os.environ.get("OPENCLAW_EXPORTER_TZ", "Europe/Berlin"))
SESSIONS_INDEX_PATH = Path(os.environ.get("OPENCLAW_SESSIONS_INDEX", str(SESSIONS_DIR / "sessions.json")))
USAGE_COST_CACHE_PATH = Path(os.environ.get("OPENCLAW_USAGE_COST_CACHE", str(SESSIONS_DIR / ".usage-cost-cache.json")))
HISTORY_DAYS = max(1, int(os.environ.get("OPENCLAW_HISTORY_DAYS", "30")))
HISTORY_HOURS = max(1, int(os.environ.get("OPENCLAW_HISTORY_HOURS", "168")))
+403
View File
@@ -0,0 +1,403 @@
from __future__ import annotations
from .collectors import collect_session_index, collect_usage, collect_usage_cost_cache
from .config import HISTORY_DAYS, HISTORY_HOURS, TIMEZONE
from .models import COST_FIELDS, SessionIndexReport, UsageCostCacheReport, UsageReport
from .utils import as_float, as_int, metric_line, sanitize_label, today_in_tz
def append_current_day_metrics(lines: list[str], usage: UsageReport, today: str, tz_name: str) -> None:
lines.extend([
"# HELP openclaw_daily_total_tokens Total OpenClaw tokens recorded today.",
"# TYPE openclaw_daily_total_tokens gauge",
metric_line("openclaw_daily_total_tokens", usage.today.total, date=today, tz=tz_name),
"# HELP openclaw_daily_input_tokens OpenClaw input tokens recorded today.",
"# TYPE openclaw_daily_input_tokens gauge",
metric_line("openclaw_daily_input_tokens", usage.today.input, date=today, tz=tz_name),
"# HELP openclaw_daily_output_tokens OpenClaw output tokens recorded today.",
"# TYPE openclaw_daily_output_tokens gauge",
metric_line("openclaw_daily_output_tokens", usage.today.output, date=today, tz=tz_name),
"# HELP openclaw_daily_cache_read_tokens OpenClaw cache-read tokens recorded today.",
"# TYPE openclaw_daily_cache_read_tokens gauge",
metric_line("openclaw_daily_cache_read_tokens", usage.today.cache_read, date=today, tz=tz_name),
"# HELP openclaw_daily_cache_write_tokens OpenClaw cache-write tokens recorded today.",
"# TYPE openclaw_daily_cache_write_tokens gauge",
metric_line("openclaw_daily_cache_write_tokens", usage.today.cache_write, date=today, tz=tz_name),
"# HELP openclaw_daily_reasoning_tokens OpenClaw reasoning tokens recorded today.",
"# TYPE openclaw_daily_reasoning_tokens gauge",
metric_line("openclaw_daily_reasoning_tokens", usage.today.reasoning, date=today, tz=tz_name),
"# HELP openclaw_daily_usage_entries OpenClaw usage entries recorded today.",
"# TYPE openclaw_daily_usage_entries gauge",
metric_line("openclaw_daily_usage_entries", usage.today.entries, date=today, tz=tz_name),
"# HELP openclaw_daily_cost_input OpenClaw input cost recorded today.",
"# TYPE openclaw_daily_cost_input gauge",
metric_line("openclaw_daily_cost_input", f"{usage.today.cost_input:.12g}", date=today, tz=tz_name),
"# HELP openclaw_daily_cost_output OpenClaw output cost recorded today.",
"# TYPE openclaw_daily_cost_output gauge",
metric_line("openclaw_daily_cost_output", f"{usage.today.cost_output:.12g}", date=today, tz=tz_name),
"# HELP openclaw_daily_cost_cache_read OpenClaw cache-read cost recorded today.",
"# TYPE openclaw_daily_cost_cache_read gauge",
metric_line("openclaw_daily_cost_cache_read", f"{usage.today.cost_cache_read:.12g}", date=today, tz=tz_name),
"# HELP openclaw_daily_cost_cache_write OpenClaw cache-write cost recorded today.",
"# TYPE openclaw_daily_cost_cache_write gauge",
metric_line("openclaw_daily_cost_cache_write", f"{usage.today.cost_cache_write:.12g}", date=today, tz=tz_name),
"# HELP openclaw_daily_cost_total OpenClaw total cost recorded today.",
"# TYPE openclaw_daily_cost_total gauge",
metric_line("openclaw_daily_cost_total", f"{usage.today.cost_total:.12g}", date=today, tz=tz_name),
"# HELP openclaw_daily_message_records Total message records seen today.",
"# TYPE openclaw_daily_message_records gauge",
metric_line("openclaw_daily_message_records", usage.message_records_today, date=today, tz=tz_name),
"# HELP openclaw_daily_non_message_records Total non-message records seen today.",
"# TYPE openclaw_daily_non_message_records gauge",
metric_line("openclaw_daily_non_message_records", usage.non_message_records_today, date=today, tz=tz_name),
"# HELP openclaw_daily_unique_sessions Number of unique session ids seen today.",
"# TYPE openclaw_daily_unique_sessions gauge",
metric_line("openclaw_daily_unique_sessions", len(usage.unique_session_ids_today), date=today, tz=tz_name),
"# HELP openclaw_daily_unique_models Number of unique models seen today.",
"# TYPE openclaw_daily_unique_models gauge",
metric_line("openclaw_daily_unique_models", len(usage.unique_models_today), date=today, tz=tz_name),
"# HELP openclaw_daily_unique_providers Number of unique providers seen today.",
"# TYPE openclaw_daily_unique_providers gauge",
metric_line("openclaw_daily_unique_providers", len(usage.unique_providers_today), date=today, tz=tz_name),
])
def append_counter_metrics(lines: list[str], usage: UsageReport) -> None:
lines.extend([
"# HELP openclaw_total_tokens_total All-time OpenClaw tokens from session usage logs.",
"# TYPE openclaw_total_tokens_total counter",
metric_line("openclaw_total_tokens_total", usage.all_time.total),
"# HELP openclaw_input_tokens_total All-time OpenClaw input tokens from session usage logs.",
"# TYPE openclaw_input_tokens_total counter",
metric_line("openclaw_input_tokens_total", usage.all_time.input),
"# HELP openclaw_output_tokens_total All-time OpenClaw output tokens from session usage logs.",
"# TYPE openclaw_output_tokens_total counter",
metric_line("openclaw_output_tokens_total", usage.all_time.output),
"# HELP openclaw_cache_read_tokens_total All-time OpenClaw cache-read tokens from session usage logs.",
"# TYPE openclaw_cache_read_tokens_total counter",
metric_line("openclaw_cache_read_tokens_total", usage.all_time.cache_read),
"# HELP openclaw_cache_write_tokens_total All-time OpenClaw cache-write tokens from session usage logs.",
"# TYPE openclaw_cache_write_tokens_total counter",
metric_line("openclaw_cache_write_tokens_total", usage.all_time.cache_write),
"# HELP openclaw_reasoning_tokens_total All-time OpenClaw reasoning tokens from session usage logs.",
"# TYPE openclaw_reasoning_tokens_total counter",
metric_line("openclaw_reasoning_tokens_total", usage.all_time.reasoning),
"# HELP openclaw_usage_entries_total All-time OpenClaw usage entries from session usage logs.",
"# TYPE openclaw_usage_entries_total counter",
metric_line("openclaw_usage_entries_total", usage.all_time.entries),
"# HELP openclaw_cost_input_usd_total All-time OpenClaw input cost in USD from session usage logs.",
"# TYPE openclaw_cost_input_usd_total counter",
metric_line("openclaw_cost_input_usd_total", f"{usage.all_time.cost_input:.12g}"),
"# HELP openclaw_cost_output_usd_total All-time OpenClaw output cost in USD from session usage logs.",
"# TYPE openclaw_cost_output_usd_total counter",
metric_line("openclaw_cost_output_usd_total", f"{usage.all_time.cost_output:.12g}"),
"# HELP openclaw_cost_cache_read_usd_total All-time OpenClaw cache-read cost in USD from session usage logs.",
"# TYPE openclaw_cost_cache_read_usd_total counter",
metric_line("openclaw_cost_cache_read_usd_total", f"{usage.all_time.cost_cache_read:.12g}"),
"# HELP openclaw_cost_cache_write_usd_total All-time OpenClaw cache-write cost in USD from session usage logs.",
"# TYPE openclaw_cost_cache_write_usd_total counter",
metric_line("openclaw_cost_cache_write_usd_total", f"{usage.all_time.cost_cache_write:.12g}"),
"# HELP openclaw_cost_total_usd_total All-time OpenClaw total cost in USD from session usage logs.",
"# TYPE openclaw_cost_total_usd_total counter",
metric_line("openclaw_cost_total_usd_total", f"{usage.all_time.cost_total:.12g}"),
"# HELP openclaw_message_records_total All-time message records seen in session logs.",
"# TYPE openclaw_message_records_total counter",
metric_line("openclaw_message_records_total", usage.message_records_all_time),
"# HELP openclaw_non_message_records_total All-time non-message records seen in session logs.",
"# TYPE openclaw_non_message_records_total counter",
metric_line("openclaw_non_message_records_total", usage.non_message_records_all_time),
])
def append_bucket_metrics(lines: list[str], usage: UsageReport, tz_name: str) -> None:
lines.extend([
"# HELP openclaw_day_total_tokens Daily token totals for recent retained days.",
"# TYPE openclaw_day_total_tokens gauge",
"# HELP openclaw_day_cost_total Daily cost totals for recent retained days.",
"# TYPE openclaw_day_cost_total gauge",
"# HELP openclaw_day_usage_entries Daily usage-entry totals for recent retained days.",
"# TYPE openclaw_day_usage_entries gauge",
])
for bucket_date, totals in sorted(usage.day_buckets.items()):
labels = {"date": bucket_date, "tz": tz_name}
lines.append(metric_line("openclaw_day_total_tokens", totals.total, **labels))
lines.append(metric_line("openclaw_day_cost_total", f"{totals.cost_total:.12g}", **labels))
lines.append(metric_line("openclaw_day_usage_entries", totals.entries, **labels))
lines.extend([
"# HELP openclaw_hour_total_tokens Hourly token totals for recent retained hours.",
"# TYPE openclaw_hour_total_tokens gauge",
"# HELP openclaw_hour_cost_total Hourly cost totals for recent retained hours.",
"# TYPE openclaw_hour_cost_total gauge",
"# HELP openclaw_hour_usage_entries Hourly usage-entry totals for recent retained hours.",
"# TYPE openclaw_hour_usage_entries gauge",
])
for bucket_hour, totals in sorted(usage.hour_buckets.items()):
labels = {"hour": bucket_hour, "tz": tz_name}
lines.append(metric_line("openclaw_hour_total_tokens", totals.total, **labels))
lines.append(metric_line("openclaw_hour_cost_total", f"{totals.cost_total:.12g}", **labels))
lines.append(metric_line("openclaw_hour_usage_entries", totals.entries, **labels))
def append_breakdown_metrics(lines: list[str], usage: UsageReport, today: str, tz_name: str) -> None:
lines.extend([
"# HELP openclaw_daily_messages_total Daily message count by role.",
"# TYPE openclaw_daily_messages_total gauge",
])
for role, count in sorted(usage.role_counts_today.items()):
lines.append(metric_line("openclaw_daily_messages_total", count, date=today, tz=tz_name, role=role))
lines.extend([
"# HELP openclaw_daily_stop_reasons_total Daily assistant stop reasons by model/provider.",
"# TYPE openclaw_daily_stop_reasons_total gauge",
])
for (stop_reason, model, provider), count in sorted(usage.stop_reason_counts_today.items()):
lines.append(metric_line("openclaw_daily_stop_reasons_total", count, date=today, tz=tz_name, stop_reason=stop_reason, model=model, provider=provider))
lines.extend([
"# HELP openclaw_daily_tool_calls_total Daily tool call count by tool name.",
"# TYPE openclaw_daily_tool_calls_total gauge",
])
for tool_name, count in sorted(usage.tool_call_counts_today.items()):
lines.append(metric_line("openclaw_daily_tool_calls_total", count, date=today, tz=tz_name, tool_name=tool_name))
lines.extend([
"# HELP openclaw_daily_errors_total Daily message error count by error type.",
"# TYPE openclaw_daily_errors_total gauge",
])
for error_type, count in sorted(usage.error_type_counts_today.items()):
lines.append(metric_line("openclaw_daily_errors_total", count, date=today, tz=tz_name, error_type=error_type))
lines.extend([
"# HELP openclaw_daily_model_total_tokens Daily total tokens by model/provider.",
"# TYPE openclaw_daily_model_total_tokens gauge",
])
for (model, provider), count in sorted(usage.tokens_by_model_today.items()):
lines.append(metric_line("openclaw_daily_model_total_tokens", count, date=today, tz=tz_name, model=model, provider=provider))
lines.extend([
"# HELP openclaw_daily_model_input_tokens Daily input tokens by model/provider.",
"# TYPE openclaw_daily_model_input_tokens gauge",
])
for (model, provider), count in sorted(usage.input_tokens_by_model_today.items()):
lines.append(metric_line("openclaw_daily_model_input_tokens", count, date=today, tz=tz_name, model=model, provider=provider))
lines.extend([
"# HELP openclaw_daily_model_output_tokens Daily output tokens by model/provider.",
"# TYPE openclaw_daily_model_output_tokens gauge",
])
for (model, provider), count in sorted(usage.output_tokens_by_model_today.items()):
lines.append(metric_line("openclaw_daily_model_output_tokens", count, date=today, tz=tz_name, model=model, provider=provider))
lines.extend([
"# HELP openclaw_daily_model_reasoning_tokens Daily reasoning tokens by model/provider.",
"# TYPE openclaw_daily_model_reasoning_tokens gauge",
])
for (model, provider), count in sorted(usage.reasoning_tokens_by_model_today.items()):
lines.append(metric_line("openclaw_daily_model_reasoning_tokens", count, date=today, tz=tz_name, model=model, provider=provider))
lines.extend([
"# HELP openclaw_model_total_tokens_total All-time total tokens by model/provider.",
"# TYPE openclaw_model_total_tokens_total counter",
])
for (model, provider), count in sorted(usage.tokens_by_model_all_time.items()):
lines.append(metric_line("openclaw_model_total_tokens_total", count, model=model, provider=provider))
lines.extend([
"# HELP openclaw_model_input_tokens_total All-time input tokens by model/provider.",
"# TYPE openclaw_model_input_tokens_total counter",
])
for (model, provider), count in sorted(usage.input_tokens_by_model_all_time.items()):
lines.append(metric_line("openclaw_model_input_tokens_total", count, model=model, provider=provider))
lines.extend([
"# HELP openclaw_model_output_tokens_total All-time output tokens by model/provider.",
"# TYPE openclaw_model_output_tokens_total counter",
])
for (model, provider), count in sorted(usage.output_tokens_by_model_all_time.items()):
lines.append(metric_line("openclaw_model_output_tokens_total", count, model=model, provider=provider))
lines.extend([
"# HELP openclaw_daily_session_total_tokens Daily total tokens by session id.",
"# TYPE openclaw_daily_session_total_tokens gauge",
])
for session_id, count in sorted(usage.session_tokens_today.items()):
lines.append(metric_line("openclaw_daily_session_total_tokens", count, date=today, tz=tz_name, session_id=session_id))
lines.extend([
"# HELP openclaw_daily_session_input_tokens Daily input tokens by session id.",
"# TYPE openclaw_daily_session_input_tokens gauge",
])
for session_id, count in sorted(usage.session_input_tokens_today.items()):
lines.append(metric_line("openclaw_daily_session_input_tokens", count, date=today, tz=tz_name, session_id=session_id))
lines.extend([
"# HELP openclaw_daily_session_output_tokens Daily output tokens by session id.",
"# TYPE openclaw_daily_session_output_tokens gauge",
])
for session_id, count in sorted(usage.session_output_tokens_today.items()):
lines.append(metric_line("openclaw_daily_session_output_tokens", count, date=today, tz=tz_name, session_id=session_id))
lines.extend([
"# HELP openclaw_session_total_tokens_total All-time total tokens by session id.",
"# TYPE openclaw_session_total_tokens_total counter",
])
for session_id, count in sorted(usage.session_tokens_all_time.items()):
lines.append(metric_line("openclaw_session_total_tokens_total", count, session_id=session_id))
lines.extend([
"# HELP openclaw_daily_session_messages_total Daily message count by session id and role.",
"# TYPE openclaw_daily_session_messages_total gauge",
])
for (session_id, role), count in sorted(usage.session_message_counts_today.items()):
lines.append(metric_line("openclaw_daily_session_messages_total", count, date=today, tz=tz_name, session_id=session_id, role=role))
lines.extend([
"# HELP openclaw_daily_session_errors_total Daily error count by session id.",
"# TYPE openclaw_daily_session_errors_total gauge",
])
for session_id, count in sorted(usage.session_error_counts_today.items()):
lines.append(metric_line("openclaw_daily_session_errors_total", count, date=today, tz=tz_name, session_id=session_id))
lines.extend([
"# HELP openclaw_daily_session_tool_calls_total Daily tool call count by session id and tool name.",
"# TYPE openclaw_daily_session_tool_calls_total gauge",
])
for (session_id, tool_name), count in sorted(usage.session_tool_call_counts_today.items()):
lines.append(metric_line("openclaw_daily_session_tool_calls_total", count, date=today, tz=tz_name, session_id=session_id, tool_name=tool_name))
def append_exporter_metrics(lines: list[str], usage: UsageReport, session_index: SessionIndexReport, usage_cost_cache: UsageCostCacheReport, today: str, tz_name: str) -> None:
lines.extend([
"# HELP openclaw_exporter_files_scanned Session files scanned for metrics generation.",
"# TYPE openclaw_exporter_files_scanned gauge",
metric_line("openclaw_exporter_files_scanned", usage.files_scanned, date=today, tz=tz_name),
"# HELP openclaw_exporter_lines_scanned Session lines scanned for metrics generation.",
"# TYPE openclaw_exporter_lines_scanned gauge",
metric_line("openclaw_exporter_lines_scanned", usage.lines_scanned, date=today, tz=tz_name),
"# HELP openclaw_exporter_parse_errors Parse or read errors while generating metrics.",
"# TYPE openclaw_exporter_parse_errors gauge",
metric_line("openclaw_exporter_parse_errors", usage.parse_errors, date=today, tz=tz_name),
"# HELP openclaw_exporter_negative_cost_entries_skipped Usage entries whose negative cost values were ignored as invalid.",
"# TYPE openclaw_exporter_negative_cost_entries_skipped gauge",
metric_line("openclaw_exporter_negative_cost_entries_skipped", usage.negative_cost_entries_skipped),
"# HELP openclaw_exporter_history_days_retained Retained day-bucket history window.",
"# TYPE openclaw_exporter_history_days_retained gauge",
metric_line("openclaw_exporter_history_days_retained", HISTORY_DAYS),
"# HELP openclaw_exporter_history_hours_retained Retained hour-bucket history window.",
"# TYPE openclaw_exporter_history_hours_retained gauge",
metric_line("openclaw_exporter_history_hours_retained", HISTORY_HOURS),
"# HELP openclaw_sessions_visible_count Session entries currently present in sessions.json output.",
"# TYPE openclaw_sessions_visible_count gauge",
metric_line("openclaw_sessions_visible_count", session_index.visible_count),
"# HELP openclaw_sessions_total_count Total stored sessions reported by sessions.json.",
"# TYPE openclaw_sessions_total_count gauge",
metric_line("openclaw_sessions_total_count", session_index.total_count),
"# HELP openclaw_sessions_has_more Whether sessions.json was truncated when written.",
"# TYPE openclaw_sessions_has_more gauge",
metric_line("openclaw_sessions_has_more", int(session_index.has_more)),
"# HELP openclaw_usage_cost_cache_files_tracked Number of files tracked in .usage-cost-cache.json.",
"# TYPE openclaw_usage_cost_cache_files_tracked gauge",
metric_line("openclaw_usage_cost_cache_files_tracked", usage_cost_cache.files_tracked),
])
lines.extend([
"# HELP openclaw_exporter_negative_cost_fields_skipped Invalid negative cost fields ignored by field name.",
"# TYPE openclaw_exporter_negative_cost_fields_skipped gauge",
])
for cost_field in COST_FIELDS:
if usage.negative_cost_fields_skipped.get(cost_field):
lines.append(metric_line("openclaw_exporter_negative_cost_fields_skipped", usage.negative_cost_fields_skipped[cost_field], cost_field=cost_field))
lines.extend([
"# HELP openclaw_exporter_negative_cost_amount_skipped Total absolute invalid negative cost amount ignored by field name.",
"# TYPE openclaw_exporter_negative_cost_amount_skipped gauge",
])
for cost_field in COST_FIELDS:
if usage.negative_cost_amounts_skipped.get(cost_field):
lines.append(metric_line("openclaw_exporter_negative_cost_amount_skipped", f"{usage.negative_cost_amounts_skipped[cost_field]:.12g}", cost_field=cost_field))
if usage_cost_cache.version is not None:
lines.extend([
"# HELP openclaw_usage_cost_cache_version Version field from .usage-cost-cache.json.",
"# TYPE openclaw_usage_cost_cache_version gauge",
metric_line("openclaw_usage_cost_cache_version", usage_cost_cache.version),
])
if usage_cost_cache.updated_at_ms is not None:
lines.extend([
"# HELP openclaw_usage_cost_cache_updated_timestamp_seconds Last update time of .usage-cost-cache.json.",
"# TYPE openclaw_usage_cost_cache_updated_timestamp_seconds gauge",
metric_line("openclaw_usage_cost_cache_updated_timestamp_seconds", f"{usage_cost_cache.updated_at_ms / 1000:.3f}"),
])
def append_session_inventory_metrics(lines: list[str], session_index: SessionIndexReport) -> None:
lines.extend([
"# HELP openclaw_session_info Non-sensitive session metadata from sessions.json. Session key is intentionally omitted.",
"# TYPE openclaw_session_info gauge",
"# HELP openclaw_session_age_seconds Current age of a stored session from sessions.json.",
"# TYPE openclaw_session_age_seconds gauge",
"# HELP openclaw_session_updated_timestamp_seconds Last update time of a stored session from sessions.json.",
"# TYPE openclaw_session_updated_timestamp_seconds gauge",
"# HELP openclaw_session_started_timestamp_seconds Session start time from sessions.json.",
"# TYPE openclaw_session_started_timestamp_seconds gauge",
"# HELP openclaw_session_last_interaction_timestamp_seconds Last interaction time from sessions.json.",
"# TYPE openclaw_session_last_interaction_timestamp_seconds gauge",
"# HELP openclaw_session_total_tokens Stored session totalTokens from sessions.json.",
"# TYPE openclaw_session_total_tokens gauge",
"# HELP openclaw_session_context_tokens Stored session contextTokens from sessions.json.",
"# TYPE openclaw_session_context_tokens gauge",
"# HELP openclaw_session_runtime_seconds Stored runtimeMs from sessions.json.",
"# TYPE openclaw_session_runtime_seconds gauge",
"# HELP openclaw_session_estimated_cost_usd Stored estimatedCostUsd from sessions.json.",
"# TYPE openclaw_session_estimated_cost_usd gauge",
"# HELP openclaw_session_compaction_count Stored compactionCount from sessions.json.",
"# TYPE openclaw_session_compaction_count gauge",
"# HELP openclaw_session_usage_family_session_count Stored usageFamilySessionIds count from sessions.json.",
"# TYPE openclaw_session_usage_family_session_count gauge",
])
for session in sorted(session_index.sessions, key=lambda item: sanitize_label(item.get("sessionId"))):
session_id = sanitize_label(session.get("sessionId"))
usage_family_session_ids = session.get("usageFamilySessionIds")
labels = {
"session_id": session_id,
"status": sanitize_label(session.get("status")),
"chat_type": sanitize_label(session.get("chatType")),
"last_channel": sanitize_label(session.get("lastChannel")),
"model": sanitize_label(session.get("model")),
"provider": sanitize_label(session.get("modelProvider")),
"auth_profile_override": sanitize_label(session.get("authProfileOverride")),
"system_sent": str(bool(session.get("systemSent"))).lower(),
"aborted_last_run": str(bool(session.get("abortedLastRun"))).lower(),
"total_tokens_fresh": str(bool(session.get("totalTokensFresh"))).lower(),
}
lines.append(metric_line("openclaw_session_info", 1, **labels))
lines.append(metric_line("openclaw_session_age_seconds", f"{as_int(session.get('ageMs')) / 1000:.3f}", session_id=session_id))
lines.append(metric_line("openclaw_session_updated_timestamp_seconds", f"{as_int(session.get('updatedAt')) / 1000:.3f}", session_id=session_id))
lines.append(metric_line("openclaw_session_started_timestamp_seconds", f"{as_int(session.get('sessionStartedAt')) / 1000:.3f}", session_id=session_id))
lines.append(metric_line("openclaw_session_last_interaction_timestamp_seconds", f"{as_int(session.get('lastInteractionAt')) / 1000:.3f}", session_id=session_id))
lines.append(metric_line("openclaw_session_total_tokens", as_int(session.get("totalTokens")), session_id=session_id))
lines.append(metric_line("openclaw_session_context_tokens", as_int(session.get("contextTokens")), session_id=session_id))
lines.append(metric_line("openclaw_session_runtime_seconds", f"{as_int(session.get('runtimeMs')) / 1000:.3f}", session_id=session_id))
lines.append(metric_line("openclaw_session_estimated_cost_usd", f"{as_float(session.get('estimatedCostUsd')):.12g}", session_id=session_id))
lines.append(metric_line("openclaw_session_compaction_count", as_int(session.get("compactionCount")), session_id=session_id))
lines.append(metric_line("openclaw_session_usage_family_session_count", len(usage_family_session_ids) if isinstance(usage_family_session_ids, list) else 0, session_id=session_id))
def render_metrics() -> str:
usage = collect_usage()
session_index = collect_session_index()
usage_cost_cache = collect_usage_cost_cache()
today = today_in_tz().isoformat()
tz_name = TIMEZONE.key
lines: list[str] = []
append_current_day_metrics(lines, usage, today, tz_name)
append_counter_metrics(lines, usage)
append_bucket_metrics(lines, usage, tz_name)
append_breakdown_metrics(lines, usage, today, tz_name)
append_exporter_metrics(lines, usage, session_index, usage_cost_cache, today, tz_name)
append_session_inventory_metrics(lines, session_index)
lines.append("")
return "\n".join(lines)
+135
View File
@@ -0,0 +1,135 @@
from __future__ import annotations
from collections import Counter, defaultdict
from dataclasses import dataclass, field
from .utils import as_float, as_int
COST_FIELDS = ("input", "output", "cacheRead", "cacheWrite", "total")
@dataclass
class Totals:
total: int = 0
input: int = 0
output: int = 0
cache_read: int = 0
cache_write: int = 0
reasoning: int = 0
entries: int = 0
cost_input: float = 0.0
cost_output: float = 0.0
cost_cache_read: float = 0.0
cost_cache_write: float = 0.0
cost_total: float = 0.0
def add_usage(self, usage: dict) -> tuple[int, int, int, int, int, int, dict[str, float]]:
total_tokens = as_int(usage.get("totalTokens"))
input_tokens = as_int(usage.get("input"))
output_tokens = as_int(usage.get("output"))
cache_read_tokens = as_int(usage.get("cacheRead"))
cache_write_tokens = as_int(usage.get("cacheWrite"))
reasoning_tokens = as_int(usage.get("reasoningTokens"))
self.entries += 1
self.total += total_tokens
self.input += input_tokens
self.output += output_tokens
self.cache_read += cache_read_tokens
self.cache_write += cache_write_tokens
self.reasoning += reasoning_tokens
skipped_negative_costs: dict[str, float] = {}
cost = usage.get("cost") or {}
if isinstance(cost, dict):
input_cost = as_float(cost.get("input"))
output_cost = as_float(cost.get("output"))
cache_read_cost = as_float(cost.get("cacheRead"))
cache_write_cost = as_float(cost.get("cacheWrite"))
total_cost = as_float(cost.get("total"))
if input_cost < 0:
skipped_negative_costs["input"] = input_cost
input_cost = 0.0
if output_cost < 0:
skipped_negative_costs["output"] = output_cost
output_cost = 0.0
if cache_read_cost < 0:
skipped_negative_costs["cacheRead"] = cache_read_cost
cache_read_cost = 0.0
if cache_write_cost < 0:
skipped_negative_costs["cacheWrite"] = cache_write_cost
cache_write_cost = 0.0
if total_cost < 0:
skipped_negative_costs["total"] = total_cost
total_cost = 0.0
self.cost_input += input_cost
self.cost_output += output_cost
self.cost_cache_read += cache_read_cost
self.cost_cache_write += cache_write_cost
self.cost_total += total_cost
return (
total_tokens,
input_tokens,
output_tokens,
cache_read_tokens,
cache_write_tokens,
reasoning_tokens,
skipped_negative_costs,
)
@dataclass
class UsageReport:
today: Totals = field(default_factory=Totals)
all_time: Totals = field(default_factory=Totals)
files_scanned: int = 0
lines_scanned: int = 0
parse_errors: int = 0
negative_cost_entries_skipped: int = 0
negative_cost_fields_skipped: Counter[str] = field(default_factory=Counter)
negative_cost_amounts_skipped: defaultdict[str, float] = field(default_factory=lambda: defaultdict(float))
message_records_today: int = 0
message_records_all_time: int = 0
non_message_records_today: int = 0
non_message_records_all_time: int = 0
unique_session_ids_today: set[str] = field(default_factory=set)
unique_models_today: set[str] = field(default_factory=set)
unique_providers_today: set[str] = field(default_factory=set)
role_counts_today: Counter[str] = field(default_factory=Counter)
stop_reason_counts_today: Counter[tuple[str, str, str]] = field(default_factory=Counter)
tool_call_counts_today: Counter[str] = field(default_factory=Counter)
error_type_counts_today: Counter[str] = field(default_factory=Counter)
session_message_counts_today: Counter[tuple[str, str]] = field(default_factory=Counter)
session_error_counts_today: Counter[str] = field(default_factory=Counter)
session_tool_call_counts_today: Counter[tuple[str, str]] = field(default_factory=Counter)
tokens_by_model_today: defaultdict[tuple[str, str], int] = field(default_factory=lambda: defaultdict(int))
input_tokens_by_model_today: defaultdict[tuple[str, str], int] = field(default_factory=lambda: defaultdict(int))
output_tokens_by_model_today: defaultdict[tuple[str, str], int] = field(default_factory=lambda: defaultdict(int))
reasoning_tokens_by_model_today: defaultdict[tuple[str, str], int] = field(default_factory=lambda: defaultdict(int))
session_tokens_today: defaultdict[str, int] = field(default_factory=lambda: defaultdict(int))
session_input_tokens_today: defaultdict[str, int] = field(default_factory=lambda: defaultdict(int))
session_output_tokens_today: defaultdict[str, int] = field(default_factory=lambda: defaultdict(int))
tokens_by_model_all_time: defaultdict[tuple[str, str], int] = field(default_factory=lambda: defaultdict(int))
input_tokens_by_model_all_time: defaultdict[tuple[str, str], int] = field(default_factory=lambda: defaultdict(int))
output_tokens_by_model_all_time: defaultdict[tuple[str, str], int] = field(default_factory=lambda: defaultdict(int))
session_tokens_all_time: defaultdict[str, int] = field(default_factory=lambda: defaultdict(int))
day_buckets: defaultdict[str, Totals] = field(default_factory=lambda: defaultdict(Totals))
hour_buckets: defaultdict[str, Totals] = field(default_factory=lambda: defaultdict(Totals))
@dataclass
class SessionIndexReport:
visible_count: int = 0
total_count: int = 0
has_more: bool = False
sessions: list[dict] = field(default_factory=list)
@dataclass
class UsageCostCacheReport:
version: int | None = None
updated_at_ms: int | None = None
files_tracked: int = 0
+69
View File
@@ -0,0 +1,69 @@
from __future__ import annotations
import gzip
import re
from datetime import date, datetime
from pathlib import Path
from .config import TIMEZONE
UUID_RE = re.compile(r"[0-9a-fA-F]{8}-[0-9a-fA-F]{4}-[0-9a-fA-F]{4}-[0-9a-fA-F]{4}-[0-9a-fA-F]{12}")
def as_int(value: object) -> int:
if value is None:
return 0
if isinstance(value, bool):
return int(value)
if isinstance(value, (int, float)):
return int(value)
return 0
def as_float(value: object) -> float:
if value is None:
return 0.0
if isinstance(value, bool):
return float(value)
if isinstance(value, (int, float)):
return float(value)
return 0.0
def today_in_tz() -> date:
return datetime.now(TIMEZONE).date()
def now_in_tz() -> datetime:
return datetime.now(TIMEZONE)
def sanitize_label(value: object, default: str = "unknown") -> str:
if value is None:
return default
text = str(value).strip()
return text or default
def prom_value(value: str) -> str:
return value.replace("\\", r"\\").replace("\n", r"\n").replace('"', r'\"')
def metric_line(name: str, value: object, **labels: object) -> str:
if labels:
rendered = ",".join(f'{key}="{prom_value(sanitize_label(val))}"' for key, val in sorted(labels.items()))
return f"{name}{{{rendered}}} {value}"
return f"{name} {value}"
def open_session_file(path: Path):
if path.suffix == ".gz":
return gzip.open(path, "rt", encoding="utf-8")
return path.open("r", encoding="utf-8")
def session_id_from_path(path: Path) -> str | None:
match = UUID_RE.search(path.name)
if match:
return match.group(0).lower()
return None