|
| 1 | +#!/usr/bin/env python3 |
| 2 | +"""Bootstrap per-worker Snuba instances for CI. |
| 3 | +
|
| 4 | +Overlaps the expensive ClickHouse table setup with the devservices |
| 5 | +health-check wait. |
| 6 | +
|
| 7 | +Phase 1 (early): As soon as ClickHouse is accepting queries, create |
| 8 | + per-worker databases and run ``snuba bootstrap --force``. |
| 9 | +Phase 2 (after devservices): Stop snuba-snuba-1 and start per-worker |
| 10 | + API containers. We must wait for devservices to finish first — |
| 11 | + stopping the container while devservices is health-checking it would |
| 12 | + cause a timeout. |
| 13 | +
|
| 14 | +Requires: XDIST_WORKERS env var |
| 15 | +Reads: /tmp/ds-exit (written by setup-devservices/wait.sh) |
| 16 | +Writes: /tmp/snuba-bootstrap-exit |
| 17 | +""" |
| 18 | + |
| 19 | +from __future__ import annotations |
| 20 | + |
| 21 | +import os |
| 22 | +import subprocess |
| 23 | +import sys |
| 24 | +import time |
| 25 | +from concurrent.futures import ThreadPoolExecutor, as_completed |
| 26 | +from functools import partial |
| 27 | +from pathlib import Path |
| 28 | +from typing import Any, Callable |
| 29 | +from urllib.error import URLError |
| 30 | +from urllib.request import urlopen |
| 31 | + |
| 32 | +DS_EXIT = Path("/tmp/ds-exit") |
| 33 | +SNUBA_EXIT = Path("/tmp/snuba-bootstrap-exit") |
| 34 | + |
| 35 | +SNUBA_ENV = { |
| 36 | + "CLICKHOUSE_HOST": "clickhouse", |
| 37 | + "CLICKHOUSE_PORT": "9000", |
| 38 | + "CLICKHOUSE_HTTP_PORT": "8123", |
| 39 | + "DEFAULT_BROKERS": "kafka:9093", |
| 40 | + "REDIS_HOST": "redis", |
| 41 | + "REDIS_PORT": "6379", |
| 42 | + "REDIS_DB": "1", |
| 43 | + "SNUBA_SETTINGS": "docker", |
| 44 | +} |
| 45 | + |
| 46 | +ENV_ARGS = [flag for k, v in SNUBA_ENV.items() for flag in ("-e", f"{k}={v}")] |
| 47 | + |
| 48 | + |
| 49 | +def retry( |
| 50 | + fn: Callable[[], Any], *, attempts: int = 3, delay: int = 5, label: str = "operation" |
| 51 | +) -> Any: |
| 52 | + for i in range(attempts): |
| 53 | + try: |
| 54 | + return fn() |
| 55 | + except Exception: |
| 56 | + if i == attempts - 1: |
| 57 | + raise |
| 58 | + log(f"{label} failed (attempt {i + 1}/{attempts}), retrying in {delay}s...") |
| 59 | + time.sleep(delay) |
| 60 | + |
| 61 | + |
| 62 | +def log(msg: str) -> None: |
| 63 | + print(msg, flush=True) |
| 64 | + |
| 65 | + |
| 66 | +def fail(msg: str) -> None: |
| 67 | + log(f"::error::{msg}") |
| 68 | + SNUBA_EXIT.write_text("1") |
| 69 | + sys.exit(1) |
| 70 | + |
| 71 | + |
| 72 | +def http_ok(url: str) -> bool: |
| 73 | + try: |
| 74 | + with urlopen(url, timeout=3): |
| 75 | + return True |
| 76 | + except (URLError, OSError): |
| 77 | + return False |
| 78 | + |
| 79 | + |
| 80 | +def docker( |
| 81 | + *args: str, check: bool = False, timeout: int | None = None |
| 82 | +) -> subprocess.CompletedProcess[str]: |
| 83 | + return subprocess.run( |
| 84 | + ["docker", *args], capture_output=True, text=True, check=check, timeout=timeout |
| 85 | + ) |
| 86 | + |
| 87 | + |
| 88 | +def docker_inspect(container: str, fmt: str) -> str: |
| 89 | + r = docker("inspect", container, "--format", fmt) |
| 90 | + return r.stdout.strip() if r.returncode == 0 else "" |
| 91 | + |
| 92 | + |
| 93 | +def inspect_snuba_container() -> tuple[str, str]: |
| 94 | + image = docker_inspect("snuba-snuba-1", "{{.Config.Image}}") |
| 95 | + network = docker_inspect( |
| 96 | + "snuba-snuba-1", |
| 97 | + "{{range $k, $v := .NetworkSettings.Networks}}{{$k}}{{end}}", |
| 98 | + ) |
| 99 | + if not image or not network: |
| 100 | + fail("Could not inspect snuba-snuba-1 container") |
| 101 | + return image, network |
| 102 | + |
| 103 | + |
| 104 | +def run_parallel(fn: Callable[[int], Any], workers: range, *, fail_fast: bool = True) -> int: |
| 105 | + """Run fn(i) in parallel for each i in workers. Returns 0 on full success.""" |
| 106 | + rc = 0 |
| 107 | + with ThreadPoolExecutor(max_workers=len(workers)) as pool: |
| 108 | + futs = {pool.submit(fn, i): i for i in workers} |
| 109 | + for fut in as_completed(futs): |
| 110 | + try: |
| 111 | + fut.result() |
| 112 | + except Exception as e: |
| 113 | + if fail_fast: |
| 114 | + fail(str(e)) |
| 115 | + log(f"ERROR: {e}") |
| 116 | + rc = 1 |
| 117 | + return rc |
| 118 | + |
| 119 | + |
| 120 | +def wait_for_prerequisites(timeout: int = 300) -> None: |
| 121 | + log("Waiting for ClickHouse and Snuba container...") |
| 122 | + start = time.monotonic() |
| 123 | + while True: |
| 124 | + if time.monotonic() - start > timeout: |
| 125 | + fail("Timed out waiting for Snuba bootstrap prerequisites") |
| 126 | + if http_ok("http://localhost:8123/") and docker_inspect("snuba-snuba-1", "{{.Id}}"): |
| 127 | + break |
| 128 | + time.sleep(2) |
| 129 | + log(f"Prerequisites ready ({time.monotonic() - start:.0f}s)") |
| 130 | + |
| 131 | + |
| 132 | +def wait_for_devservices(timeout: int = 300) -> None: |
| 133 | + start = time.monotonic() |
| 134 | + while not DS_EXIT.exists(): |
| 135 | + if time.monotonic() - start > timeout: |
| 136 | + fail("Timed out waiting for devservices to finish") |
| 137 | + time.sleep(1) |
| 138 | + rc = int(DS_EXIT.read_text().strip()) |
| 139 | + if rc != 0: |
| 140 | + fail(f"devservices failed (exit {rc}), skipping Phase 2") |
| 141 | + |
| 142 | + |
| 143 | +def bootstrap_worker(worker_id: int, *, image: str, network: str) -> None: |
| 144 | + """Create a ClickHouse database and run snuba bootstrap.""" |
| 145 | + db = f"default_gw{worker_id}" |
| 146 | + |
| 147 | + def create_db() -> None: |
| 148 | + with urlopen( |
| 149 | + "http://localhost:8123/", f"CREATE DATABASE IF NOT EXISTS {db}".encode(), timeout=30 |
| 150 | + ): |
| 151 | + pass |
| 152 | + |
| 153 | + retry(create_db, label=f"CREATE DATABASE {db}") |
| 154 | + |
| 155 | + def run_bootstrap() -> None: |
| 156 | + r = docker( |
| 157 | + "run", |
| 158 | + "--rm", |
| 159 | + "--network", |
| 160 | + network, |
| 161 | + "-e", |
| 162 | + f"CLICKHOUSE_DATABASE={db}", |
| 163 | + *ENV_ARGS, |
| 164 | + image, |
| 165 | + "bootstrap", |
| 166 | + "--force", |
| 167 | + ) |
| 168 | + for line in (r.stdout + r.stderr).strip().splitlines()[-3:]: |
| 169 | + log(line) |
| 170 | + if r.returncode != 0: |
| 171 | + raise RuntimeError(f"snuba bootstrap failed for worker {worker_id}") |
| 172 | + |
| 173 | + retry(run_bootstrap, label=f"snuba bootstrap gw{worker_id}") |
| 174 | + |
| 175 | + |
| 176 | +def start_worker_container(worker_id: int, *, image: str, network: str) -> None: |
| 177 | + """Start a per-worker Snuba API container and wait for health.""" |
| 178 | + db = f"default_gw{worker_id}" |
| 179 | + port = 1230 + worker_id |
| 180 | + name = f"snuba-gw{worker_id}" |
| 181 | + |
| 182 | + docker("rm", "-f", name) |
| 183 | + |
| 184 | + r = docker( |
| 185 | + "run", |
| 186 | + "-d", |
| 187 | + "--name", |
| 188 | + name, |
| 189 | + "--network", |
| 190 | + network, |
| 191 | + "-p", |
| 192 | + f"{port}:1218", |
| 193 | + "-e", |
| 194 | + f"CLICKHOUSE_DATABASE={db}", |
| 195 | + *ENV_ARGS, |
| 196 | + "-e", |
| 197 | + "DEBUG=1", |
| 198 | + image, |
| 199 | + "api", |
| 200 | + ) |
| 201 | + if r.returncode != 0: |
| 202 | + raise RuntimeError(f"docker run {name} failed: {r.stderr.strip()}") |
| 203 | + |
| 204 | + for attempt in range(1, 31): |
| 205 | + if http_ok(f"http://127.0.0.1:{port}/health"): |
| 206 | + log(f"{name} healthy on port {port}") |
| 207 | + return |
| 208 | + if attempt == 30: |
| 209 | + r = docker("logs", name) |
| 210 | + for line in (r.stdout + r.stderr).strip().splitlines()[-20:]: |
| 211 | + log(line) |
| 212 | + raise RuntimeError(f"{name} failed health check after 30 attempts") |
| 213 | + time.sleep(2) |
| 214 | + |
| 215 | + |
| 216 | +def main() -> None: |
| 217 | + workers_str = os.environ.get("XDIST_WORKERS") |
| 218 | + if not workers_str: |
| 219 | + fail("XDIST_WORKERS must be set") |
| 220 | + workers = range(int(workers_str)) |
| 221 | + start = time.monotonic() |
| 222 | + |
| 223 | + wait_for_prerequisites() |
| 224 | + image, network = inspect_snuba_container() |
| 225 | + |
| 226 | + log("Phase 1: bootstrapping ClickHouse databases") |
| 227 | + run_parallel(partial(bootstrap_worker, image=image, network=network), workers) |
| 228 | + log(f"Phase 1 done ({time.monotonic() - start:.0f}s)") |
| 229 | + |
| 230 | + wait_for_devservices() |
| 231 | + try: |
| 232 | + docker("stop", "snuba-snuba-1", timeout=30) |
| 233 | + except subprocess.TimeoutExpired: |
| 234 | + log("WARNING: docker stop snuba-snuba-1 timed out, killing") |
| 235 | + docker("kill", "snuba-snuba-1") |
| 236 | + |
| 237 | + log("Phase 2: starting per-worker Snuba API containers") |
| 238 | + rc = run_parallel( |
| 239 | + partial(start_worker_container, image=image, network=network), |
| 240 | + workers, |
| 241 | + fail_fast=False, |
| 242 | + ) |
| 243 | + |
| 244 | + log(f"Snuba bootstrap complete ({time.monotonic() - start:.0f}s total)") |
| 245 | + SNUBA_EXIT.write_text(str(rc)) |
| 246 | + sys.exit(rc) |
| 247 | + |
| 248 | + |
| 249 | +if __name__ == "__main__": |
| 250 | + main() |
0 commit comments