host_session/main.py

import os
import re
import time

import requests

SIDECAR_URL = os.getenv("SIDECAR_URL", "http://sidecar:80")
DB_URL = os.getenv("DB_CONTROLLER_URL", "http://sidecar:8081")
TABLE = "host_session"
SIGNALS_FILE = "/app/signals.yml"
MINUTE_MS = 60000
HOUR_MS = 60 * MINUTE_MS
LOOKBACK_MS = 60 * 24 * HOUR_MS
MATCH_MS = 2 * MINUTE_MS


def read_history(signals, start_ms, end_ms, resolution):
    names = ",".join(str(signal) for signal in signals)
    path = f"/api/history/?signals={names}&start={int(start_ms)}&end={int(end_ms)}&resolution={resolution}"
    response = requests.get(SIDECAR_URL + path, timeout=(5, 80))
    response.raise_for_status()
    return {item["name"]: item for item in response.json()["data"]["series"]}


def signal_hosts():
    with open(SIGNALS_FILE) as file:
        names = re.findall(r"name:\s*'([^']+)'", file.read())
    return sorted(name[: -len("-uptime")] for name in names if name.endswith("-uptime"))


def owns(host, now_ms):
    path = f"/api/history/?signals={host}-uptime&start={int(now_ms - MINUTE_MS)}&end={int(now_ms)}&resolution=1min"
    status = requests.get(SIDECAR_URL + path, timeout=(5, 80)).status_code
    if status not in (200, 403):
        raise RuntimeError(f"history probe for {host} answered {status}")
    return status == 200


def own_host():
    now_ms = time.time() * 1000
    host = next((name for name in signal_hosts() if owns(name, now_ms)), None)
    if host is None:
        raise RuntimeError("no uptime signal of this gateway is listed in signals.yml")
    return host


def samples(series):
    start, step = series["startMs"], series["stepMs"]
    return [(start + i * step, value) for i, value in enumerate(series["values"]) if value is not None]


def last_samples(signal, start_ms, end_ms, resolution):
    if end_ms <= start_ms:
        return []
    found = samples(read_history([signal], start_ms, end_ms, resolution)[signal])
    return [(t, v) for t, v in found if t < end_ms]


def current_boot_ms():
    with open("/proc/uptime") as file:
        uptime_s = float(file.read().split()[0])
    return int(time.time() * 1000 - uptime_s * 1000)


def previous_session(host, boot_ms):
    uptime = f"{host}-uptime"
    hours = last_samples(uptime, boot_ms - LOOKBACK_MS, boot_ms, "1hr")
    if not hours:
        return None
    minutes = last_samples(uptime, hours[-1][0], min(hours[-1][0] + HOUR_MS, boot_ms), "1min")
    if not minutes:
        return None
    seconds = last_samples(uptime, minutes[-1][0], min(minutes[-1][0] + MINUTE_MS, boot_ms), "1s")
    end_ms, uptime_s = (seconds or minutes)[-1]
    return int(end_ms - uptime_s * 1000), int(end_ms)


def db_call(method, path, payload):
    response = requests.request(method, DB_URL + path, json=payload, timeout=(5, 30))
    response.raise_for_status()
    body = response.json()
    if body.get("statusCode", 200) != 200:
        raise RuntimeError(f"db-controller {body.get('statusCode')}: {body.get('errorMessage')}")
    return body.get("data")


def stored_rows():
    return db_call("POST", f"/api/db-query/{TABLE}/", {"filters": [], "orderBy": ["start_time"], "limit": 1000}) or []


def find_row(rows, start_ms):
    return next((row for row in rows if abs(row["start_time"] - start_ms) < MATCH_MS), None)


def open_row(start_ms):
    row = {"start_time": start_ms, "end_time": start_ms, "duration_s": 0, "coverage": 0}
    for prefix in ("cpu", "ram", "disk"):
        row[f"{prefix}_mean"], row[f"{prefix}_max"] = 0, 0
    return row


def mean_max(series, start_ms, end_ms):
    values = [v for t, v in samples(series) if start_ms <= t < end_ms]
    if not values:
        return None, None
    return sum(values) / len(values), max(values)


def coverage(series, start_ms, end_ms):
    total = max(1, (end_ms - start_ms) // MINUTE_MS)
    present = sum(1 for t, _ in samples(series) if start_ms <= t < end_ms)
    return min(1.0, present / total)


def build_row(host, start_ms, end_ms):
    cpu, ram, disk = f"{host}-cpu", f"{host}-ram", f"{host}-disk"
    history = read_history([cpu, ram, disk], start_ms, end_ms, "1min")
    row = {"start_time": start_ms, "end_time": end_ms, "duration_s": (end_ms - start_ms) // 1000}
    for prefix, signal in (("cpu", cpu), ("ram", ram), ("disk", disk)):
        row[f"{prefix}_mean"], row[f"{prefix}_max"] = mean_max(history[signal], start_ms, end_ms)
    row["coverage"] = coverage(history[cpu], start_ms, end_ms)
    return row


def close_previous(host, rows, boot_ms):
    session = previous_session(host, boot_ms)
    if session is None:
        print("no previous session in the uptime history", flush=True)
        return
    row = build_row(host, *session)
    stored = find_row(rows, session[0])
    if stored is None:
        print("previous session stored", db_call("POST", f"/api/db/{TABLE}/", row), row, flush=True)
    elif stored["end_time"] == stored["start_time"]:
        db_call("PUT", f"/api/db/{TABLE}/{stored['id']}/", row)
        print("previous session closed", stored["id"], row, flush=True)


def main():
    boot_ms = current_boot_ms()
    rows = stored_rows()
    if find_row(rows, boot_ms) is not None:
        return
    close_previous(own_host(), rows, boot_ms)
    print("current session opened", db_call("POST", f"/api/db/{TABLE}/", open_row(boot_ms)), flush=True)


if __name__ == "__main__":
    main()