OSVauco/ml/billing_agent.py

402 lines
18 KiB
Python

import os
import datetime
from google.cloud import bigquery
# ── Kreditt-konfig: tildelte totaler per type (NOK) ──────────────────────────
# Sett disse i Cloud Run env-vars eller cloudbuild.yaml --set-env-vars
# GOOGLE_CREDIT_INFRA_NOK = Free Trial (Cloud Run, BQ, Compute, Artifact Registry)
# GOOGLE_CREDIT_VERTEX_NOK = Trial for Gen App Builder (Vertex AI tokens)
# GOOGLE_CREDIT_DIALOGFLOW_NOK = Dialogflow CX Trial
CREDIT_CONFIG = [
{
"env_key": "GOOGLE_CREDIT_INFRA_NOK",
"label": "Free Trial (Infrastruktur)",
"bq_types": ["PROMOTION"], # BQ credit type
"skus": None, # alle SKU-er (generell)
"expires_days": 27, # kjent utløp fra Console
"currency": "NOK",
},
{
"env_key": "GOOGLE_CREDIT_VERTEX_NOK",
"label": "Trial for Gen App Builder (Vertex AI)",
"bq_types": ["PROMOTION"],
"skus": ["Vertex AI", "Generative AI", "Cloud AI"],
"expires_days": None,
"currency": "NOK",
},
{
"env_key": "GOOGLE_CREDIT_DIALOGFLOW_NOK",
"label": "Dialogflow CX Trial",
"bq_types": ["PROMOTION"],
"skus": ["Dialogflow"],
"expires_days": None,
"currency": "NOK",
},
]
class BillingAgent:
def __init__(self):
self.project_id = os.environ.get("GOOGLE_CLOUD_PROJECT")
if not self.project_id:
raise ValueError("GOOGLE_CLOUD_PROJECT environment variable not set.")
self.billing_table = os.environ.get("BILLING_TABLE")
if not self.billing_table:
raise ValueError("BILLING_TABLE environment variable not set.")
self.bq_client = bigquery.Client(project=self.project_id)
# ── summary ────────────────────────────────────────────────────────────────────
def get_summary(self):
query = f"""
SELECT
DATE(usage_start_time) AS usage_date,
project.id AS project_id,
service.description AS service,
SUM(cost) AS daily_cost
FROM `{self.billing_table}`
WHERE _PARTITIONTIME >= TIMESTAMP_SUB(CURRENT_TIMESTAMP(), INTERVAL 30 DAY)
GROUP BY usage_date, project_id, service
ORDER BY usage_date DESC, daily_cost DESC
LIMIT 100
"""
results = self.bq_client.query(query).result()
summary = [
{
"usage_date": str(row.usage_date),
"project_id": row.project_id,
"service": row.service,
"daily_cost": row.daily_cost,
}
for row in results
]
if not summary:
return {
"onboarding_status": {
"state": "awaiting_data",
"message": "Fakturaeksport er aktiv, men ingen data for siste 30 dager ennå.",
}
}
return {"summary": summary}
# ── forecast ──────────────────────────────────────────────────────────────────
def get_forecast(self):
q7 = f"""
SELECT SUM(cost) + SUM(IFNULL(
(SELECT SUM(c.amount) FROM UNNEST(credits) c), 0)) AS total_cost
FROM `{self.billing_table}`
WHERE _PARTITIONTIME >= TIMESTAMP_SUB(CURRENT_TIMESTAMP(), INTERVAL 7 DAY)
"""
total_7d = list(self.bq_client.query(q7).result())[0].total_cost or 0
daily_average = total_7d / 7
today = datetime.date.today()
next_month = datetime.date(
today.year + (1 if today.month == 12 else 0),
(today.month % 12) + 1, 1
)
remaining_days = (next_month - today).days
q_mtd = f"""
SELECT SUM(cost) + SUM(IFNULL(
(SELECT SUM(c.amount) FROM UNNEST(credits) c), 0)) AS total_cost
FROM `{self.billing_table}`
WHERE EXTRACT(MONTH FROM _PARTITIONTIME) = EXTRACT(MONTH FROM CURRENT_DATE())
AND EXTRACT(YEAR FROM _PARTITIONTIME) = EXTRACT(YEAR FROM CURRENT_DATE())
"""
mtd_cost = list(self.bq_client.query(q_mtd).result())[0].total_cost or 0
return {
"daily_average_last_7_days": daily_average,
"month_to_date_cost": mtd_cost,
"forecasted_remaining_cost": daily_average * remaining_days,
"total_monthly_forecast": mtd_cost + daily_average * remaining_days,
"remaining_days_in_month": remaining_days,
"data_note": "Prognose basert på siste 7 dager. BigQuery kan ha 24-48 timers forsinkelse.",
}
# ── discover_credits: aktiv scanner for alle kreditttyper i BQ ───────────────
def discover_credits(self, days: int = 90):
"""
Scanner BQ-billing for alle unike kreditttyper siste {days} dager.
Returnerer liste med type, full_name, total brukt, og forslag til env-var.
Nyttig for å oppdage nye kreditter automatisk.
"""
q = f"""
SELECT
cr.type AS credit_type,
cr.full_name AS full_name,
ROUND(SUM(cr.amount), 2) AS total_used,
COUNT(DISTINCT DATE(_PARTITIONTIME)) AS active_days
FROM `{self.billing_table}`,
UNNEST(credits) AS cr
WHERE _PARTITIONTIME >= TIMESTAMP_SUB(CURRENT_TIMESTAMP(), INTERVAL {int(days)} DAY)
GROUP BY credit_type, full_name
ORDER BY total_used ASC
"""
rows = list(self.bq_client.query(q).result())
discovered = []
for row in rows:
label = (row.full_name or row.credit_type or "Ukjent").strip()
env_suggestion = "GOOGLE_CREDIT_" + label.upper().replace(" ", "_").replace("-", "_")[:30] + "_NOK"
discovered.append({
"credit_type": row.credit_type,
"full_name": row.full_name,
"total_used_nok": abs(float(row.total_used or 0)),
"active_days": row.active_days,
"env_suggestion": env_suggestion,
"configured": any(
os.environ.get(c["env_key"]) for c in CREDIT_CONFIG
if row.credit_type in c["bq_types"]
),
})
return {
"discovered": discovered,
"period_days": days,
"note": "Sett env-vars i Cloud Run for nøyaktig runway-beregning per kreditttype.",
}
# ── credits-status (oppdatert) ────────────────────────────────────────────────
def get_credits_status(self, days: int = 90):
"""
Henter faktiske kreditter fra BigQuery og beregner separat runway
per kreditttype (INFRA, VERTEX, DIALOGFLOW) basert på env-vars i NOK.
Returnerer:
- credits_by_type: [{type, full_name, amount}]
- credit_pools: [{label, total_nok, used_nok, remaining_nok,
runway_days, exhaustion_date, expires_days, warning}]
- credits_used_total_nok: total brukt (alle typer)
- gross_cost_total_nok: bruttokostnad
- net_cost_total_nok: nettokostnad etter kreditter
- daily_gross_burn_nok: daglig bruttokostnad (7d snitt)
- daily_credit_burn_nok: daglig kredittforbruk (7d snitt)
- data_as_of: siste dato med data i BQ
- warning: kritisk advarsel (Free Trial 27d)
"""
q_credits = f"""
SELECT
cr.type AS credit_type,
cr.full_name AS full_name,
ROUND(SUM(cr.amount), 4) AS total_amount
FROM `{self.billing_table}`,
UNNEST(credits) AS cr
WHERE _PARTITIONTIME >= TIMESTAMP_SUB(CURRENT_TIMESTAMP(), INTERVAL {int(days)} DAY)
GROUP BY credit_type, full_name
ORDER BY total_amount ASC
"""
q_burn = f"""
SELECT
ROUND(SUM(cost) / 7, 6) AS daily_gross,
ROUND(SUM(IFNULL(
(SELECT SUM(c.amount) FROM UNNEST(credits) c), 0)) / 7, 6) AS daily_credit
FROM `{self.billing_table}`
WHERE _PARTITIONTIME >= TIMESTAMP_SUB(CURRENT_TIMESTAMP(), INTERVAL 7 DAY)
"""
q_totals = f"""
SELECT
ROUND(SUM(cost), 4) AS gross_total,
ROUND(SUM(cost) + SUM(IFNULL(
(SELECT SUM(c.amount) FROM UNNEST(credits) c), 0)), 4) AS net_total
FROM `{self.billing_table}`
WHERE _PARTITIONTIME >= TIMESTAMP_SUB(CURRENT_TIMESTAMP(), INTERVAL {int(days)} DAY)
"""
q_latest = f"""
SELECT MAX(DATE(usage_start_time)) AS latest_date
FROM `{self.billing_table}`
"""
credits_rows = list(self.bq_client.query(q_credits).result())
burn_row = list(self.bq_client.query(q_burn).result())[0]
totals_row = list(self.bq_client.query(q_totals).result())[0]
latest_row = list(self.bq_client.query(q_latest).result())[0]
credits_by_type = [
{
"type": row.credit_type,
"full_name": row.full_name,
"amount": float(row.total_amount),
}
for row in credits_rows
]
credits_used_total = abs(sum(r["amount"] for r in credits_by_type))
gross_total = float(totals_row.gross_total or 0)
net_total = float(totals_row.net_total or 0)
daily_gross = float(burn_row.daily_gross or 0)
daily_credit = abs(float(burn_row.daily_credit or 0))
data_as_of = str(latest_row.latest_date) if latest_row.latest_date else None
# ── Per-pool runway ───────────────────────────────────────────────────────
credit_pools = []
top_warning = None
for cfg in CREDIT_CONFIG:
total_nok = float(os.environ.get(cfg["env_key"], "0") or "0")
# Brukt beregnes som andel av total BQ-kreditter (alle PROMOTION er NOK)
used_nok = credits_used_total # samme BQ-total deles; raffineres hvis SKU-filter legges til
remaining_nok = round(total_nok - used_nok, 2) if total_nok > 0 else None
runway_days = None
exhaustion_date = None
pool_warning = None
if total_nok > 0 and daily_credit > 0 and remaining_nok is not None:
runway_days = int(remaining_nok / daily_credit) if remaining_nok > 0 else 0
exhaustion_date = str(
datetime.date.today() + datetime.timedelta(days=runway_days)
) if runway_days > 0 else str(datetime.date.today())
# Utløpsadvarsel fra Console (kjent)
expires_days = cfg.get("expires_days")
if expires_days is not None and expires_days <= 30:
pool_warning = f"⚠️ Utløper om {expires_days} dager — kr {total_nok:,.2f} går tapt!"
top_warning = pool_warning
elif runway_days is not None and runway_days < 30:
pool_warning = f"⚠️ Estimert tom om {runway_days} dager ({exhaustion_date})."
if top_warning is None:
top_warning = pool_warning
elif runway_days is not None and runway_days < 60:
pool_warning = f"⚠️ Kreditter estimert tom om {runway_days} dager."
credit_pools.append({
"env_key": cfg["env_key"],
"label": cfg["label"],
"total_nok": total_nok if total_nok > 0 else None,
"used_nok": round(used_nok, 2),
"remaining_nok": remaining_nok,
"runway_days": runway_days,
"exhaustion_date": exhaustion_date,
"expires_days": expires_days,
"configured": total_nok > 0,
"warning": pool_warning,
})
return {
"credits_used_total_nok": round(credits_used_total, 4),
"gross_cost_total_nok": gross_total,
"net_cost_total_nok": net_total,
"daily_gross_burn_nok": round(daily_gross, 6),
"daily_credit_burn_nok": round(daily_credit, 6),
"credit_pools": credit_pools,
"credits_by_type": credits_by_type,
"data_as_of": data_as_of,
"period_days": days,
"warning": top_warning,
"setup_note": None if any(p["configured"] for p in credit_pools) else (
"Sett GOOGLE_CREDIT_INFRA_NOK, GOOGLE_CREDIT_VERTEX_NOK og "
"GOOGLE_CREDIT_DIALOGFLOW_NOK i Cloud Run env for nøyaktig runway per kreditttype."
),
}
# ── anomalier ──────────────────────────────────────────────────────────────────
def get_anomalies(self):
query = f"""
WITH daily_costs AS (
SELECT
service.description AS service,
DATE(_PARTITIONTIME) AS usage_date,
SUM(cost) + SUM(IFNULL(
(SELECT SUM(c.amount) FROM UNNEST(credits) c), 0)) AS daily_cost
FROM `{self.billing_table}`
WHERE _PARTITIONTIME >= TIMESTAMP_SUB(CURRENT_TIMESTAMP(), INTERVAL 14 DAY)
GROUP BY 1, 2
),
costs_with_avg AS (
SELECT
service, usage_date, daily_cost,
AVG(daily_cost) OVER (
PARTITION BY service ORDER BY usage_date
ROWS BETWEEN 7 PRECEDING AND 1 PRECEDING
) AS avg_7day
FROM daily_costs
)
SELECT service, daily_cost AS today_cost, avg_7day,
(daily_cost / avg_7day) AS ratio
FROM costs_with_avg
WHERE usage_date = CURRENT_DATE()
AND avg_7day > 0
AND daily_cost > (2.0 * avg_7day)
"""
results = self.bq_client.query(query).result()
return {
"anomalies": [
{
"service": row.service,
"today_cost": row.today_cost,
"avg_7d": row.avg_7day,
"ratio": row.ratio,
}
for row in results
]
}
# ── daglig historikk for bar-chart (CG6) ───────────────────────────────────
def get_daily_history(self, days: int = 30):
query = f"""
SELECT
DATE(usage_start_time) AS usage_date,
SUM(cost) + SUM(IFNULL(
(SELECT SUM(c.amount) FROM UNNEST(credits) c), 0)) AS day_cost
FROM `{self.billing_table}`
WHERE _PARTITIONTIME >= TIMESTAMP_SUB(CURRENT_TIMESTAMP(),
INTERVAL {int(days)} DAY)
GROUP BY usage_date
ORDER BY usage_date ASC
"""
results = list(self.bq_client.query(query).result())
history = []
mtd = 0.0
for row in results:
if history and row.usage_date.day == 1:
mtd = 0.0
mtd += float(row.day_cost or 0)
history.append({"date": str(row.usage_date), "mtd": round(mtd, 6)})
return history
# ── tjenester gruppert per service (CG5+CG6) ─────────────────────────────────
def get_service_totals(self, days: int = 30):
query = f"""
SELECT
service.description AS service,
sku.description AS sku,
SUM(cost) + SUM(IFNULL(
(SELECT SUM(c.amount) FROM UNNEST(credits) c), 0)) AS sku_cost
FROM `{self.billing_table}`
WHERE _PARTITIONTIME >= TIMESTAMP_SUB(CURRENT_TIMESTAMP(),
INTERVAL {int(days)} DAY)
GROUP BY service, sku
HAVING sku_cost > 0
ORDER BY service, sku_cost DESC
"""
results = self.bq_client.query(query).result()
services: dict = {}
for row in results:
svc = row.service
if svc not in services:
services[svc] = {"service": svc, "total_cost": 0.0, "skus": []}
services[svc]["total_cost"] = round(
services[svc]["total_cost"] + float(row.sku_cost), 6
)
services[svc]["skus"].append({
"sku": row.sku,
"sku_cost": round(float(row.sku_cost), 6),
})
return sorted(services.values(), key=lambda x: x["total_cost"], reverse=True)
if __name__ == '__main__':
agent = BillingAgent()
print("Summary:", agent.get_summary())
print("Forecast:", agent.get_forecast())
print("Credits status:", agent.get_credits_status())
print("Credits discover:",agent.discover_credits())
print("Anomalies:", agent.get_anomalies())
print("By-service:", agent.get_service_totals())
print("History:", agent.get_daily_history())