|
| 1 | +import importlib.resources |
| 2 | +import logging |
| 3 | +import subprocess |
| 4 | +import tempfile |
| 5 | +from asyncio import Lock |
| 6 | +from pathlib import Path |
| 7 | +from typing import Annotated, Dict, Literal, Union |
| 8 | + |
| 9 | +import jinja2 |
| 10 | +from pydantic import BaseModel, Field |
| 11 | + |
| 12 | +from dstack.gateway.common import run_async |
| 13 | +from dstack.gateway.errors import GatewayError |
| 14 | + |
| 15 | +CONFIGS_DIR = Path("/etc/nginx/sites-enabled") |
| 16 | +GATEWAY_PORT = 8000 |
| 17 | +logger = logging.getLogger(__name__) |
| 18 | + |
| 19 | + |
| 20 | +class SiteConfig(BaseModel): |
| 21 | + type: str |
| 22 | + domain: str |
| 23 | + |
| 24 | + def render(self) -> str: |
| 25 | + template = importlib.resources.read_text( |
| 26 | + "dstack.gateway.resources.nginx", f"{self.type}.jinja2" |
| 27 | + ) |
| 28 | + return jinja2.Template(template).render( |
| 29 | + **self.model_dump(), |
| 30 | + gateway_port=GATEWAY_PORT, |
| 31 | + ) |
| 32 | + |
| 33 | + |
| 34 | +class ServiceConfig(SiteConfig): |
| 35 | + type: Literal["service"] = "service" |
| 36 | + project: str |
| 37 | + service_id: str |
| 38 | + auth: bool |
| 39 | + servers: Dict[str, str] = {} |
| 40 | + |
| 41 | + |
| 42 | +class EntrypointConfig(SiteConfig): |
| 43 | + type: Literal["entrypoint"] = "entrypoint" |
| 44 | + proxy_path: str |
| 45 | + |
| 46 | + |
| 47 | +class Nginx(BaseModel): |
| 48 | + """ |
| 49 | + Nginx keeps track of registered domains, updates nginx config and issues SSL certificates. |
| 50 | + Its internal state could be serialized to a file and restored from it using pydantic. |
| 51 | + """ |
| 52 | + |
| 53 | + configs: Dict[ |
| 54 | + str, Annotated[Union[ServiceConfig, EntrypointConfig], Field(discriminator="type")] |
| 55 | + ] = {} |
| 56 | + _lock: Lock = Lock() |
| 57 | + |
| 58 | + async def register_service(self, project: str, service_id: str, domain: str, auth: bool): |
| 59 | + config_name = self.get_config_name(domain) |
| 60 | + conf = ServiceConfig( |
| 61 | + project=project, |
| 62 | + service_id=service_id, |
| 63 | + domain=domain, |
| 64 | + auth=auth, |
| 65 | + ) |
| 66 | + |
| 67 | + async with self._lock: |
| 68 | + if config_name in self.configs: |
| 69 | + raise GatewayError(f"Domain {domain} is already registered") |
| 70 | + |
| 71 | + logger.debug("Registering service domain %s", domain) |
| 72 | + |
| 73 | + await run_async(self.run_certbot, domain) |
| 74 | + await run_async(self.write_conf, conf.render(), config_name) |
| 75 | + self.configs[config_name] = conf |
| 76 | + |
| 77 | + logger.info("Service domain %s is registered now", domain) |
| 78 | + |
| 79 | + async def register_entrypoint(self, domain: str, prefix: str): |
| 80 | + config_name = self.get_config_name(domain) |
| 81 | + conf = EntrypointConfig( |
| 82 | + domain=domain, |
| 83 | + proxy_path=prefix, |
| 84 | + ) |
| 85 | + |
| 86 | + async with self._lock: |
| 87 | + if config_name in self.configs: |
| 88 | + raise GatewayError(f"Domain {domain} is already registered") |
| 89 | + |
| 90 | + logger.debug("Registering entrypoint domain %s", domain) |
| 91 | + |
| 92 | + await run_async(self.run_certbot, domain) |
| 93 | + await run_async(self.write_conf, conf.render(), config_name) |
| 94 | + self.configs[config_name] = conf |
| 95 | + |
| 96 | + logger.info("Entrypoint domain %s is registered now", domain) |
| 97 | + |
| 98 | + async def unregister_domain(self, domain: str): |
| 99 | + config_name = self.get_config_name(domain) |
| 100 | + |
| 101 | + async with self._lock: |
| 102 | + if config_name not in self.configs: |
| 103 | + raise GatewayError("Domain is not registered") |
| 104 | + |
| 105 | + logger.debug("Unregistering domain %s", domain) |
| 106 | + |
| 107 | + await run_async(sudo_rm, CONFIGS_DIR / config_name) |
| 108 | + await run_async(self.reload) |
| 109 | + self.configs.pop(config_name) |
| 110 | + |
| 111 | + logger.info("Domain %s is unregistered now", domain) |
| 112 | + |
| 113 | + async def add_upstream(self, domain: str, server: str, replica_id: str): |
| 114 | + config_name = self.get_config_name(domain) |
| 115 | + |
| 116 | + async with self._lock: |
| 117 | + if config_name not in self.configs: |
| 118 | + raise GatewayError(f"Domain {domain} is not registered") |
| 119 | + |
| 120 | + logger.debug("Adding upstream %s to domain %s", server, domain) |
| 121 | + |
| 122 | + conf = self.configs[config_name].model_copy(deep=True) |
| 123 | + conf.servers[replica_id] = server |
| 124 | + await run_async(self.write_conf, conf.render(), config_name) |
| 125 | + self.configs[config_name] = conf |
| 126 | + |
| 127 | + logger.debug("Upstream %s is added to domain %s", server, domain) |
| 128 | + |
| 129 | + async def remove_upstream(self, domain: str, replica_id: str): |
| 130 | + config_name = self.get_config_name(domain) |
| 131 | + |
| 132 | + async with self._lock: |
| 133 | + if config_name not in self.configs: |
| 134 | + raise GatewayError(f"Domain {domain} is not registered") |
| 135 | + if replica_id not in self.configs[config_name].servers: |
| 136 | + raise GatewayError(f"Upstream {replica_id} is not registered") |
| 137 | + |
| 138 | + logger.debug("Removing upstream %s from domain %s", replica_id, domain) |
| 139 | + |
| 140 | + conf = self.configs[config_name].model_copy(deep=True) |
| 141 | + conf.servers.pop(replica_id) |
| 142 | + await run_async(self.write_conf, conf.render(), config_name) |
| 143 | + self.configs[config_name] = conf |
| 144 | + |
| 145 | + logger.debug("Upstream %s is removed from domain %s", replica_id, domain) |
| 146 | + |
| 147 | + @staticmethod |
| 148 | + def reload(): |
| 149 | + cmd = ["sudo", "systemctl", "reload", "nginx.service"] |
| 150 | + r = subprocess.run(cmd) |
| 151 | + if r.returncode != 0: |
| 152 | + raise GatewayError("Failed to reload nginx") |
| 153 | + |
| 154 | + @classmethod |
| 155 | + def write_conf(cls, conf: str, conf_name: str): |
| 156 | + """Update config and reload nginx. Rollback changes on error.""" |
| 157 | + conf_path = CONFIGS_DIR / conf_name |
| 158 | + old_conf = conf_path.read_text() if conf_path.exists() else None |
| 159 | + |
| 160 | + sudo_write(conf_path, conf) |
| 161 | + try: |
| 162 | + cls.reload() |
| 163 | + except GatewayError: |
| 164 | + # rollback changes |
| 165 | + if old_conf is not None: |
| 166 | + sudo_write(conf_path, old_conf) |
| 167 | + else: |
| 168 | + sudo_rm(conf_path) |
| 169 | + raise |
| 170 | + |
| 171 | + @staticmethod |
| 172 | + def run_certbot(domain: str): |
| 173 | + logger.info("Running certbot for %s", domain) |
| 174 | + cmd = ["sudo", "certbot", "certonly"] |
| 175 | + cmd += ["--non-interactive", "--agree-tos", "--register-unsafely-without-email"] |
| 176 | + cmd += ["--nginx", "--domain", domain] |
| 177 | + r = subprocess.run(cmd, capture_output=True) |
| 178 | + if r.returncode != 0: |
| 179 | + raise GatewayError(f"Certbot failed:\n{r.stderr.decode()}") |
| 180 | + |
| 181 | + @staticmethod |
| 182 | + def get_config_name(domain: str) -> str: |
| 183 | + return f"443-{domain}.conf" |
| 184 | + |
| 185 | + |
| 186 | +def sudo_write(path: Path, content: str): |
| 187 | + with tempfile.NamedTemporaryFile("w") as temp: |
| 188 | + temp.write(content) |
| 189 | + temp.flush() |
| 190 | + temp.seek(0) |
| 191 | + r = subprocess.run(["sudo", "cp", "-p", temp.name, path]) |
| 192 | + if r.returncode != 0: |
| 193 | + raise GatewayError("Failed to copy file as sudo") |
| 194 | + |
| 195 | + |
| 196 | +def sudo_rm(path: Path): |
| 197 | + r = subprocess.run(["sudo", "rm", path]) |
| 198 | + if r.returncode != 0: |
| 199 | + raise GatewayError("Failed to remove file as sudo") |
0 commit comments