| 1 | |
|
| 2 | |
|
| 3 | |
|
| 4 | |
|
| 5 | |
|
| 6 | | use crate::home::Home; |
| 7 | | use anyhow::{bail, Context, Result}; |
| 8 | | use flow_core::api::{Record, SecretInput}; |
| 9 | | use flow_core::{Environment, Job, MountSource, Sandbox}; |
| 10 | | use std::io::{BufRead, BufReader, Read}; |
| 11 | | use std::path::Path; |
| 12 | | use std::sync::mpsc; |
| 13 | | use std::time::{Duration, Instant}; |
| 14 | |
|
| 15 | | pub struct Outcome { |
| 16 | | |
| 17 | | pub exit_code: Option<i32>, |
| 18 | | pub duration_ms: u64, |
| 19 | | pub environment: Environment, |
| 20 | | pub timed_out: bool, |
| 21 | | |
| 22 | | pub cancelled: bool, |
| 23 | | |
| 24 | | pub peak_memory_mib: Option<u64>, |
| 25 | | |
| 26 | | |
| 27 | | pub pruned: Vec<(String, flow_init::prune::Pruned)>, |
| 28 | | } |
| 29 | |
|
| 30 | |
|
| 31 | |
|
| 32 | | pub trait Sink { |
| 33 | | fn record(&mut self, r: Record); |
| 34 | 15× | fn cancelled(&mut self) -> bool { |
| 35 | 15× | false |
| 36 | 15× | } |
| 37 | | } |
| 38 | |
|
| 39 | | impl<F: FnMut(Record)> Sink for F { |
| 40 | 17× | fn record(&mut self, r: Record) { |
| 41 | 17× | self(r) |
| 42 | 17× | } |
| 43 | | } |
| 44 | |
|
| 45 | |
|
| 46 | 0× | pub fn environment(home: &Home, sandbox: &Sandbox, job: &Job) -> Result<Environment> { |
| 47 | 0× | let cfg = home.config.microvm.as_ref().context("flow.toml has no [microvm] section")?; |
| 48 | | Ok(Environment { |
| 49 | 0× | runtime: "microvm".into(), |
| 50 | 0× | parts: vec![ |
| 51 | 0× | ("firecracker".into(), cfg.firecracker.blake3.clone()), |
| 52 | 0× | ("kernel".into(), cfg.kernel.blake3.clone()), |
| 53 | 0× | ("init".into(), cfg.init.blake3.clone()), |
| 54 | 0× | ("memory_mib".into(), sandbox.memory_mib.to_string()), |
| 55 | 0× | ("cpus".into(), sandbox.cpus.to_string()), |
| 56 | | ] |
| 57 | 0× | .into_iter() |
| 58 | 0× | .chain(job.mounts.iter().map(|m| (format!("mount {}", m.target), format!("{} {} MiB", m.source.as_str(), m.size_mib)))) |
| 59 | 0× | .collect(), |
| 60 | 0× | rootfs: sandbox.rootfs.clone(), |
| 61 | 0× | argv: job.exec.clone(), |
| 62 | 0× | env: job.env.iter().map(|(k, v)| (k.clone(), v.clone())).collect(), |
| 63 | | }) |
| 64 | 0× | } |
| 65 | |
|
| 66 | |
|
| 67 | | #[allow(clippy::too_many_arguments)] |
| 68 | 0× | pub fn run( |
| 69 | 0× | home: &Home, |
| 70 | 0× | sandbox: &Sandbox, |
| 71 | 0× | job: &Job, |
| 72 | 0× | secrets: &[SecretInput], |
| 73 | 0× | workspace: &Path, |
| 74 | 0× | scratch: &Path, |
| 75 | 0× | timeout: Duration, |
| 76 | 0× | sink: &mut dyn Sink, |
| 77 | 0× | ) -> Result<Outcome> { |
| 78 | 0× | let cfg = home.config.microvm.as_ref().context("flow.toml has no [microvm] section")?; |
| 79 | 0× | let firecracker = home.verified(&home.resolve(&cfg.firecracker.path), &cfg.firecracker.blake3)?; |
| 80 | 0× | let kernel = home.verified(&home.resolve(&cfg.kernel.path), &cfg.kernel.blake3)?; |
| 81 | 0× | let init = home.verified(&home.resolve(&cfg.init.path), &cfg.init.blake3)?; |
| 82 | 0× | let rootfs = home.rootfs(&sandbox.rootfs); |
| 83 | 0× | if !rootfs.is_file() { |
| 84 | 0× | bail!("rootfs {} is not on this runner (import it with `flow rootfs import`)", sandbox.rootfs); |
| 85 | 0× | } |
| 86 | 0× | let rootfs = home.verified(&rootfs, &sandbox.rootfs)?; |
| 87 | 0× | let environment = environment(home, sandbox, job)?; |
| 88 | |
|
| 89 | 0× | let percent = home.config.runner.as_ref().and_then(|r| r.cache_prune_percent).unwrap_or(flow_init::prune::DEFAULT_PERCENT); |
| 90 | 0× | let mut spec = guest_spec(job, percent); |
| 91 | 0× | spec["loopback"] = serde_json::Value::Bool(sandbox.network == flow_core::Network::Loopback); |
| 92 | | |
| 93 | | |
| 94 | 0× | let egress = sandbox.network == flow_core::Network::Egress; |
| 95 | 0× | let (egress_tx, egress_rx) = mpsc::channel::<String>(); |
| 96 | 0× | let mut _proxy = None; |
| 97 | 0× | if egress { |
| 98 | 0× | if !home.config.runner.as_ref().is_some_and(|r| r.egress) { |
| 99 | 0× | bail!("this runner does not allow network egress (a runner allows it with `egress = true` under [runner] in flow.toml)"); |
| 100 | 0× | } |
| 101 | 0× | _proxy = Some( |
| 102 | 0× | crate::egress::start(&scratch.join(format!("v.sock_{}", crate::egress::VSOCK_PORT)), sandbox.egress_allow.clone(), egress_tx) |
| 103 | 0× | .context("start the egress proxy")?, |
| 104 | | ); |
| 105 | 0× | spec["egress"] = serde_json::Value::Bool(true); |
| 106 | 0× | add_proxy_env(&mut spec); |
| 107 | 0× | sink.record(Record { |
| 108 | 0× | t: 0, |
| 109 | 0× | stream: "flow".into(), |
| 110 | 0× | text: format!("egress: through this runner, to {} only", sandbox.egress_allow.join(", ")), |
| 111 | 0× | }); |
| 112 | 0× | } |
| 113 | | |
| 114 | | |
| 115 | 0× | if !secrets.is_empty() { |
| 116 | 0× | spec["secrets"] = |
| 117 | 0× | serde_json::Value::Array(secrets.iter().map(|s| serde_json::json!({ "path": s.path, "value": s.value })).collect()); |
| 118 | 0× | } |
| 119 | | |
| 120 | | |
| 121 | 0× | let mut disks = Vec::new(); |
| 122 | 0× | for (i, m) in job.mounts.iter().filter(|m| m.source == MountSource::Disk).enumerate() { |
| 123 | 0× | let path = scratch.join(format!("mount-{i}.ext4")); |
| 124 | | |
| 125 | 0× | if !path.exists() { |
| 126 | 0× | crate::workspace::empty_ext4(&path, m.size_mib, "scratch")?; |
| 127 | 0× | } |
| 128 | 0× | disks.push( |
| 129 | 0× | serde_json::json!({ "drive_id": format!("mount{i}"), "path_on_host": path, "is_root_device": false, "is_read_only": false }), |
| 130 | | ); |
| 131 | | } |
| 132 | 0× | let init_bytes = std::fs::read(&init)?; |
| 133 | 0× | let spec_bytes = serde_json::to_vec(&spec)?; |
| 134 | 0× | let initrd = crate::cpio::newc(&[ |
| 135 | 0× | crate::cpio::Entry { name: "init", mode: 0o755, data: &init_bytes }, |
| 136 | 0× | crate::cpio::Entry { name: "job.json", mode: 0o644, data: &spec_bytes }, |
| 137 | 0× | ]); |
| 138 | 0× | let initrd_path = scratch.join("initrd.cpio"); |
| 139 | 0× | std::fs::write(&initrd_path, initrd)?; |
| 140 | 0× | let drives: Vec<serde_json::Value> = [ |
| 141 | 0× | serde_json::json!({ "drive_id": "rootfs", "path_on_host": rootfs, "is_root_device": false, "is_read_only": true }), |
| 142 | 0× | serde_json::json!({ "drive_id": "workspace", "path_on_host": workspace, "is_root_device": false, "is_read_only": false }), |
| 143 | 0× | ] |
| 144 | 0× | .into_iter() |
| 145 | 0× | .chain(disks) |
| 146 | 0× | .collect(); |
| 147 | 0× | let vm = serde_json::json!({ |
| 148 | 0× | "boot-source": { |
| 149 | 0× | "kernel_image_path": kernel, |
| 150 | 0× | "initrd_path": initrd_path, |
| 151 | 0× | "boot_args": "console=ttyS0 reboot=k panic=-1 pci=off quiet loglevel=0 rdinit=/init", |
| 152 | | }, |
| 153 | 0× | "drives": drives, |
| 154 | 0× | "machine-config": { "vcpu_count": sandbox.cpus, "mem_size_mib": sandbox.memory_mib }, |
| 155 | | }); |
| 156 | 0× | let mut vm = vm; |
| 157 | 0× | if egress { |
| 158 | 0× | |
| 159 | 0× | |
| 160 | 0× | vm["vsock"] = serde_json::json!({ "guest_cid": 3, "uds_path": "v.sock" }); |
| 161 | 0× | } |
| 162 | 0× | let vm_path = scratch.join("vm.json"); |
| 163 | 0× | std::fs::write(&vm_path, serde_json::to_vec_pretty(&vm)?)?; |
| 164 | |
|
| 165 | 0× | let started = Instant::now(); |
| 166 | 0× | let mut child = std::process::Command::new(&firecracker) |
| 167 | 0× | .arg("--no-api") |
| 168 | 0× | .arg("--config-file") |
| 169 | 0× | .arg(&vm_path) |
| 170 | 0× | .args(["--level", "Error"]) |
| 171 | 0× | .current_dir(scratch) |
| 172 | 0× | .stdin(std::process::Stdio::null()) |
| 173 | 0× | .stdout(std::process::Stdio::piped()) |
| 174 | 0× | .stderr(std::process::Stdio::piped()) |
| 175 | 0× | .spawn() |
| 176 | 0× | .with_context(|| format!("start {}", firecracker.display()))?; |
| 177 | 0× | let (tx, rx) = mpsc::channel::<(bool, Vec<u8>)>(); |
| 178 | 0× | let out = child.stdout.take().unwrap(); |
| 179 | 0× | let err = child.stderr.take().unwrap(); |
| 180 | 0× | let tx2 = tx.clone(); |
| 181 | 0× | let readers = [std::thread::spawn(move || lines(out, true, tx)), std::thread::spawn(move || lines(err, false, tx2))]; |
| 182 | 0× | let mut exit_code = None; |
| 183 | 0× | let mut timed_out = false; |
| 184 | 0× | let mut cancelled = false; |
| 185 | 0× | let deadline = started + timeout; |
| 186 | 0× | let t = |s: &Instant| s.elapsed().as_millis() as u64; |
| 187 | | |
| 188 | | |
| 189 | 0× | let mut peak_memory_mib = None; |
| 190 | 0× | let mut pruned = Vec::new(); |
| 191 | | loop { |
| 192 | 0× | if let Some(m) = crate::facts::peak_rss_mib(child.id()) { |
| 193 | 0× | peak_memory_mib = Some(m); |
| 194 | 0× | } |
| 195 | 0× | while let Ok(line) = egress_rx.try_recv() { |
| 196 | 0× | sink.record(Record { t: t(&started), stream: "flow".into(), text: line }); |
| 197 | 0× | } |
| 198 | 0× | match rx.recv_timeout(Duration::from_millis(200)) { |
| 199 | 0× | Ok((from_console, line)) => { |
| 200 | 0× | let text = String::from_utf8_lossy(&line).trim_end_matches('\r').to_string(); |
| 201 | 0× | let rec = |stream: &str, text: &str| Record { t: t(&started), stream: stream.into(), text: text.into() }; |
| 202 | 0× | if !from_console { |
| 203 | 0× | sink.record(rec("console", &format!("firecracker: {text}"))); |
| 204 | 0× | } else if let Some(rest) = text.strip_prefix('\u{1e}') { |
| 205 | 0× | let (kind, body) = rest.split_at(rest.len().min(1)); |
| 206 | 0× | let body = body.strip_prefix(' ').unwrap_or(body); |
| 207 | 0× | match kind { |
| 208 | 0× | "O" => sink.record(rec("out", body)), |
| 209 | 0× | "E" => sink.record(rec("err", body)), |
| 210 | 0× | "X" => exit_code = body.trim().parse().ok(), |
| 211 | 0× | "F" => sink.record(rec("flow", &format!("guest init failed: {body}"))), |
| 212 | 0× | "C" => match pruned_frame(body) { |
| 213 | 0× | Some(p) => pruned.push(p), |
| 214 | 0× | None => sink.record(rec("console", &text)), |
| 215 | | }, |
| 216 | 0× | _ => sink.record(rec("console", &text)), |
| 217 | | } |
| 218 | 0× | } else if !text.is_empty() { |
| 219 | 0× | sink.record(rec("console", &text)); |
| 220 | 0× | } |
| 221 | | } |
| 222 | 0× | Err(mpsc::RecvTimeoutError::Timeout) => {} |
| 223 | 0× | Err(mpsc::RecvTimeoutError::Disconnected) => break, |
| 224 | | } |
| 225 | 0× | if !timed_out && !cancelled && sink.cancelled() { |
| 226 | 0× | cancelled = true; |
| 227 | 0× | let _ = child.kill(); |
| 228 | 0× | sink.record(Record { t: t(&started), stream: "flow".into(), text: "stopped: the run was cancelled".into() }); |
| 229 | 0× | } |
| 230 | 0× | if Instant::now() >= deadline && !timed_out && !cancelled { |
| 231 | 0× | timed_out = true; |
| 232 | 0× | let _ = child.kill(); |
| 233 | 0× | sink.record(Record { |
| 234 | 0× | t: t(&started), |
| 235 | 0× | stream: "flow".into(), |
| 236 | 0× | text: format!("stopped at the time limit of {} s", timeout.as_secs()), |
| 237 | 0× | }); |
| 238 | 0× | } |
| 239 | | } |
| 240 | 0× | let status = child.wait()?; |
| 241 | 0× | for r in readers { |
| 242 | 0× | let _ = r.join(); |
| 243 | 0× | } |
| 244 | 0× | if exit_code.is_none() && !timed_out && !cancelled { |
| 245 | 0× | sink.record(Record { |
| 246 | 0× | t: t(&started), |
| 247 | 0× | stream: "flow".into(), |
| 248 | 0× | text: format!("the microVM ended without the job's exit code ({status})"), |
| 249 | 0× | }); |
| 250 | 0× | } |
| 251 | | Ok(Outcome { |
| 252 | 0× | exit_code: if timed_out || cancelled { None } else { exit_code }, |
| 253 | 0× | duration_ms: t(&started), |
| 254 | 0× | environment, |
| 255 | 0× | timed_out, |
| 256 | 0× | cancelled, |
| 257 | 0× | peak_memory_mib, |
| 258 | 0× | pruned, |
| 259 | | }) |
| 260 | 0× | } |
| 261 | |
|
| 262 | |
|
| 263 | 2× | fn pruned_frame(body: &str) -> Option<(String, flow_init::prune::Pruned)> { |
| 264 | | #[derive(serde::Deserialize)] |
| 265 | | struct Frame { |
| 266 | | target: String, |
| 267 | | pruned: flow_init::prune::Pruned, |
| 268 | | } |
| 269 | 2× | let f: Frame = serde_json::from_str(body).ok()?; |
| 270 | 1× | Some((f.target, f.pruned)) |
| 271 | 2× | } |
| 272 | |
|
| 273 | |
|
| 274 | |
|
| 275 | 1× | fn add_proxy_env(spec: &mut serde_json::Value) { |
| 276 | 1× | let Some(env) = spec["env"].as_array_mut() else { return }; |
| 277 | 1× | let set: Vec<String> = env.iter().filter_map(|kv| kv[0].as_str().map(String::from)).collect(); |
| 278 | 6× | for (k, v) in [ |
| 279 | 1× | ("HTTPS_PROXY", crate::egress::GUEST_PROXY), |
| 280 | 1× | ("https_proxy", crate::egress::GUEST_PROXY), |
| 281 | 1× | ("HTTP_PROXY", crate::egress::GUEST_PROXY), |
| 282 | 1× | ("http_proxy", crate::egress::GUEST_PROXY), |
| 283 | 1× | ("NO_PROXY", "localhost,127.0.0.1"), |
| 284 | 1× | ("no_proxy", "localhost,127.0.0.1"), |
| 285 | 1× | ] { |
| 286 | 6× | if !set.iter().any(|s| s == k) { |
| 287 | 5× | env.push(serde_json::json!([k, v])); |
| 288 | 5× | } |
| 289 | | } |
| 290 | 1× | } |
| 291 | |
|
| 292 | |
|
| 293 | |
|
| 294 | |
|
| 295 | |
|
| 296 | 3× | pub fn guest_spec(job: &Job, percent: u8) -> serde_json::Value { |
| 297 | 3× | let mut next_disk = b'c'; |
| 298 | 3× | let mounts: Vec<serde_json::Value> = job |
| 299 | 3× | .mounts |
| 300 | 3× | .iter() |
| 301 | 5× | .map(|m| match m.source { |
| 302 | 1× | MountSource::Tmpfs => serde_json::json!({ "kind": "tmpfs", "target": m.target, "size_mib": m.size_mib }), |
| 303 | | MountSource::Disk => { |
| 304 | 4× | let dev = format!("/dev/vd{}", next_disk as char); |
| 305 | 4× | next_disk += 1; |
| 306 | 4× | let mut disk = serde_json::json!({ "kind": "disk", "target": m.target, "size_mib": m.size_mib, "device": dev }); |
| 307 | 4× | if m.cache.is_some() { |
| 308 | 1× | let prefix = format!("{}/", m.target); |
| 309 | 2× | let keep: Vec<&str> = job.outputs.iter().filter_map(|o| o.path.strip_prefix(&prefix)).collect(); |
| 310 | 1× | disk["prune"] = percent.into(); |
| 311 | 1× | disk["keep"] = keep.into(); |
| 312 | 3× | } |
| 313 | 4× | disk |
| 314 | | } |
| 315 | 5× | }) |
| 316 | 3× | .collect(); |
| 317 | 3× | serde_json::json!({ "argv": job.exec, "env": job.env.iter().collect::<Vec<_>>(), "mounts": mounts }) |
| 318 | 3× | } |
| 319 | |
|
| 320 | 0× | fn lines(r: impl Read, console: bool, tx: mpsc::Sender<(bool, Vec<u8>)>) { |
| 321 | 0× | let mut r = BufReader::new(r); |
| 322 | 0× | let mut buf = Vec::new(); |
| 323 | | loop { |
| 324 | 0× | buf.clear(); |
| 325 | 0× | match r.read_until(b'\n', &mut buf) { |
| 326 | 0× | Ok(0) | Err(_) => break, |
| 327 | | Ok(_) => { |
| 328 | 0× | if buf.last() == Some(&b'\n') { |
| 329 | 0× | buf.pop(); |
| 330 | 0× | } |
| 331 | 0× | if tx.send((console, buf.clone())).is_err() { |
| 332 | 0× | break; |
| 333 | 0× | } |
| 334 | | } |
| 335 | | } |
| 336 | | } |
| 337 | 0× | } |
| 338 | |
|
| 339 | | #[cfg(test)] |
| 340 | | mod tests { |
| 341 | | use super::*; |
| 342 | | use flow_core::Mount; |
| 343 | |
|
| 344 | | #[test] |
| 345 | 1× | fn a_job_with_egress_gets_the_proxy_variables_unless_it_sets_them() { |
| 346 | 1× | let mut spec = serde_json::json!({ "env": [["HTTPS_PROXY", "http://mine:1"]] }); |
| 347 | 1× | super::add_proxy_env(&mut spec); |
| 348 | 1× | let env = spec["env"].as_array().unwrap(); |
| 349 | 8× | let get = |k: &str| env.iter().find(|kv| kv[0] == k).map(|kv| kv[1].as_str().unwrap().to_string()); |
| 350 | 1× | assert_eq!(get("HTTPS_PROXY").as_deref(), Some("http://mine:1"), "the job's own is kept"); |
| 351 | 1× | assert_eq!(get("https_proxy").as_deref(), Some("http://127.0.0.1:3128")); |
| 352 | 1× | assert_eq!(get("NO_PROXY").as_deref(), Some("localhost,127.0.0.1")); |
| 353 | 1× | } |
| 354 | |
|
| 355 | | #[test] |
| 356 | 1× | fn guest_spec_numbers_disks_after_the_workspace() { |
| 357 | 1× | let mut job: Job = serde_json::from_value(serde_json::json!({ |
| 358 | 1× | "name": "t", "sandbox": "s", "scope": [], "depends_on": [], "env": {"A": "1"}, "exec": ["true"] |
| 359 | | })) |
| 360 | 1× | .unwrap(); |
| 361 | 1× | assert_eq!(guest_spec(&job, 60)["mounts"], serde_json::json!([]), "no mounts"); |
| 362 | 1× | job.mounts = vec![ |
| 363 | 1× | Mount { source: MountSource::Disk, target: "/src/target".into(), size_mib: 8192, cache: None, key: vec![] }, |
| 364 | 1× | Mount { source: MountSource::Tmpfs, target: "/tmp/cache".into(), size_mib: 256, cache: None, key: vec![] }, |
| 365 | 1× | Mount { source: MountSource::Disk, target: "/tmp/z".into(), size_mib: 64, cache: None, key: vec![] }, |
| 366 | | ]; |
| 367 | 1× | let spec = guest_spec(&job, 60); |
| 368 | 1× | assert_eq!(spec["mounts"][0], serde_json::json!({"kind": "disk", "target": "/src/target", "size_mib": 8192, "device": "/dev/vdc"})); |
| 369 | 1× | assert_eq!(spec["mounts"][1], serde_json::json!({"kind": "tmpfs", "target": "/tmp/cache", "size_mib": 256})); |
| 370 | 1× | assert_eq!(spec["mounts"][2]["device"], "/dev/vdd"); |
| 371 | 1× | assert_eq!(spec["env"], serde_json::json!([["A", "1"]])); |
| 372 | 1× | } |
| 373 | |
|
| 374 | | |
| 375 | | |
| 376 | | |
| 377 | | #[test] |
| 378 | 1× | fn a_cache_disk_is_pruned_in_the_guest_but_its_outputs_stay() { |
| 379 | 1× | let mut job: Job = serde_json::from_value(serde_json::json!({ |
| 380 | 1× | "name": "t", "sandbox": "s", "scope": [], "depends_on": [], "env": {}, "exec": ["true"], |
| 381 | 1× | "outputs": [{"name": "bin", "path": "/src/target/release/tool"}, {"name": "w", "path": "/src/w"}] |
| 382 | | })) |
| 383 | 1× | .unwrap(); |
| 384 | 1× | job.mounts = vec![ |
| 385 | 1× | Mount { source: MountSource::Disk, target: "/src/target".into(), size_mib: 8192, cache: Some("t".into()), key: vec![] }, |
| 386 | 1× | Mount { source: MountSource::Disk, target: "/tmp/z".into(), size_mib: 64, cache: None, key: vec![] }, |
| 387 | | ]; |
| 388 | 1× | let spec = guest_spec(&job, 45); |
| 389 | 1× | assert_eq!((spec["mounts"][0]["prune"].as_u64(), &spec["mounts"][0]["keep"]), (Some(45), &serde_json::json!(["release/tool"]))); |
| 390 | 1× | assert!(spec["mounts"][1].get("prune").is_none() && spec["mounts"][1].get("keep").is_none()); |
| 391 | 1× | let (target, p) = |
| 392 | 1× | pruned_frame(r#"{"target":"/src/target","pruned":{"size":100,"percent":60,"before":90,"after":50,"freed":40,"files":3}}"#) |
| 393 | 1× | .unwrap(); |
| 394 | 1× | assert_eq!((target.as_str(), p.after, p.cap(), p.fits(), p.unpruned), ("/src/target", 50, 60, true, None)); |
| 395 | 1× | assert!(pruned_frame("not json").is_none()); |
| 396 | 1× | } |
| 397 | | } |