| 1 |
|
| 2 |
|
| 3 |
|
| 4 |
|
| 5 |
|
| 6 |
use crate::home::{Home, RunnerConfig}; |
| 7 |
use crate::{microvm, workspace}; |
| 8 |
use anyhow::{bail, Context, Result}; |
| 9 |
use flow_core::api::{Claim, ClaimRequest, Finish, LogBatch, Record, MAX_RECORD_TEXT}; |
| 10 |
use flow_core::{scope_tree_hash, Receipt}; |
| 11 |
use oxy_core::hash::ChangeHash; |
| 12 |
use oxy_core::repo::Repo; |
| 13 |
use std::path::PathBuf; |
| 14 |
use std::time::{Duration, Instant}; |
| 15 |
|
| 16 |
pub struct Hub { |
| 17 |
base: String, |
| 18 |
token: String, |
| 19 |
agent: ureq::Agent, |
| 20 |
} |
| 21 |
|
| 22 |
impl Hub { |
| 23 |
pub fn new(cfg: &RunnerConfig) -> Result<Hub> { |
| 24 |
let tls = oxy_core::protocol::tls_client_config(cfg.ca.as_deref()).map_err(|e| anyhow::anyhow!("{e}"))?; |
| 25 |
let agent = ureq::AgentBuilder::new().tls_config(tls).timeout(Duration::from_secs(60)).build(); |
| 26 |
Ok(Hub { base: cfg.hub.trim_end_matches('/').to_string(), token: cfg.token.clone(), agent }) |
| 27 |
} |
| 28 |
|
| 29 |
fn post(&self, path: &str, body: &impl serde::Serialize) -> Result<Option<ureq::Response>> { |
| 30 |
let url = format!("{}/api/v1{path}", self.base); |
| 31 |
match self.agent.post(&url).set("Authorization", &format!("Bearer {}", self.token)).send_json(body) { |
| 32 |
Ok(r) if r.status() == 204 => Ok(None), |
| 33 |
Ok(r) => Ok(Some(r)), |
| 34 |
Err(ureq::Error::Status(code, r)) => bail!("{url}: HTTP {code}: {}", r.into_string().unwrap_or_default()), |
| 35 |
Err(e) => bail!("{url}: {e}"), |
| 36 |
} |
| 37 |
} |
| 38 |
|
| 39 |
pub fn claim(&self, req: &ClaimRequest) -> Result<Option<Claim>> { |
| 40 |
match self.post("/flow/claim", req)? { |
| 41 |
Some(r) => Ok(Some(r.into_json()?)), |
| 42 |
None => Ok(None), |
| 43 |
} |
| 44 |
} |
| 45 |
|
| 46 |
pub fn log(&self, job: i64, records: Vec<Record>) -> Result<()> { |
| 47 |
self.post(&format!("/flow/jobs/{job}/log"), &LogBatch { records })?; |
| 48 |
Ok(()) |
| 49 |
} |
| 50 |
|
| 51 |
pub fn finish(&self, job: i64, f: &Finish) -> Result<()> { |
| 52 |
self.post(&format!("/flow/jobs/{job}/finish"), f)?; |
| 53 |
Ok(()) |
| 54 |
} |
| 55 |
} |
| 56 |
|
| 57 |
pub fn serve(home: &Home) -> Result<()> { |
| 58 |
let cfg = home.config.runner.clone().context("flow.toml has no [runner] section")?; |
| 59 |
let hub = Hub::new(&cfg)?; |
| 60 |
let name = cfg.name.clone().unwrap_or_else(|| "flow".into()); |
| 61 |
let req = ClaimRequest { runner: name.clone(), runtimes: vec!["microvm".into()] }; |
| 62 |
eprintln!("flow runner {name}: polling {}", cfg.hub); |
| 63 |
let mut idle = Duration::from_secs(2); |
| 64 |
loop { |
| 65 |
match hub.claim(&req) { |
| 66 |
Ok(Some(claim)) => { |
| 67 |
idle = Duration::from_secs(2); |
| 68 |
let id = claim.job_id; |
| 69 |
eprintln!("job {id}: {} {} on {}", claim.repo, claim.job.name, claim.heads.first().map(|h| &h[..12]).unwrap_or("-")); |
| 70 |
let finish = run_claim(home, &hub, &cfg, &claim); |
| 71 |
let finish = |
| 72 |
finish.unwrap_or_else(|e| Finish { exit_code: None, duration_ms: 0, receipt: None, error: Some(format!("{e:#}")) }); |
| 73 |
eprintln!( |
| 74 |
"job {id}: exit {:?}{}", |
| 75 |
finish.exit_code, |
| 76 |
finish.error.as_deref().map(|e| format!(" ({e})")).unwrap_or_default() |
| 77 |
); |
| 78 |
if let Err(e) = hub.finish(id, &finish) { |
| 79 |
eprintln!("job {id}: reporting failed: {e:#}"); |
| 80 |
} |
| 81 |
} |
| 82 |
Ok(None) => std::thread::sleep(idle), |
| 83 |
Err(e) => { |
| 84 |
eprintln!("claim: {e:#}"); |
| 85 |
std::thread::sleep(idle); |
| 86 |
idle = (idle * 2).min(Duration::from_secs(60)); |
| 87 |
} |
| 88 |
} |
| 89 |
} |
| 90 |
} |
| 91 |
|
| 92 |
|
| 93 |
struct LogSink<'a> { |
| 94 |
hub: &'a Hub, |
| 95 |
job: i64, |
| 96 |
buf: Vec<Record>, |
| 97 |
last: Instant, |
| 98 |
} |
| 99 |
|
| 100 |
impl LogSink<'_> { |
| 101 |
fn push(&mut self, mut r: Record) { |
| 102 |
if r.text.len() > MAX_RECORD_TEXT { |
| 103 |
let mut cut = MAX_RECORD_TEXT; |
| 104 |
while !r.text.is_char_boundary(cut) { |
| 105 |
cut -= 1; |
| 106 |
} |
| 107 |
r.text.truncate(cut); |
| 108 |
} |
| 109 |
self.buf.push(r); |
| 110 |
if self.buf.len() >= 200 || self.last.elapsed() >= Duration::from_secs(1) { |
| 111 |
self.flush(); |
| 112 |
} |
| 113 |
} |
| 114 |
|
| 115 |
fn flush(&mut self) { |
| 116 |
if !self.buf.is_empty() { |
| 117 |
if let Err(e) = self.hub.log(self.job, std::mem::take(&mut self.buf)) { |
| 118 |
eprintln!("job {}: log: {e:#}", self.job); |
| 119 |
} |
| 120 |
} |
| 121 |
self.last = Instant::now(); |
| 122 |
} |
| 123 |
} |
| 124 |
|
| 125 |
fn run_claim(home: &Home, hub: &Hub, cfg: &RunnerConfig, claim: &Claim) -> Result<Finish> { |
| 126 |
let started = Instant::now(); |
| 127 |
let mut sink = LogSink { hub, job: claim.job_id, buf: Vec::new(), last: Instant::now() }; |
| 128 |
let note = |
| 129 |
|sink: &mut LogSink, text: String| sink.push(Record { t: started.elapsed().as_millis() as u64, stream: "flow".into(), text }); |
| 130 |
let heads: Vec<ChangeHash> = claim.heads.iter().map(|h| ChangeHash::from_hex(h).context("bad head")).collect::<Result<_>>()?; |
| 131 |
note(&mut sink, format!("fetching {} for state {}", claim.repo, claim.heads.iter().map(|h| &h[..12]).collect::<Vec<_>>().join("+"))); |
| 132 |
let repo = cached_repo(home, cfg, claim)?; |
| 133 |
let scratch = home.root.join("work").join(format!("job-{}", claim.job_id)); |
| 134 |
let _ = std::fs::remove_dir_all(&scratch); |
| 135 |
std::fs::create_dir_all(scratch.join("tree"))?; |
| 136 |
let result = (|| -> Result<Finish> { |
| 137 |
let mut files = workspace::export_state(&repo, &heads, &scratch.join("tree"))?; |
| 138 |
|
| 139 |
|
| 140 |
for sub in &claim.subrepos { |
| 141 |
if sub.path.split('/').any(|p| p.is_empty() || p == "." || p == ".." || p == ".oxy") { |
| 142 |
bail!("subrepo path {:?} is not a clean relative path", sub.path); |
| 143 |
} |
| 144 |
note( |
| 145 |
&mut sink, |
| 146 |
format!( |
| 147 |
"fetching subrepo {} at {} for {}", |
| 148 |
sub.repo, |
| 149 |
sub.heads.iter().map(|h| &h[..12]).collect::<Vec<_>>().join("+"), |
| 150 |
sub.path |
| 151 |
), |
| 152 |
); |
| 153 |
let sub_repo = cached_repo_at(home, cfg, &sub.repo, &sub.clone_url, &claim.token)?; |
| 154 |
let sub_heads: Vec<ChangeHash> = |
| 155 |
sub.heads.iter().map(|h| ChangeHash::from_hex(h).context("bad subrepo head")).collect::<Result<_>>()?; |
| 156 |
let dir = scratch.join("tree").join(&sub.path); |
| 157 |
std::fs::create_dir_all(&dir)?; |
| 158 |
let sub_files = workspace::export_state(&sub_repo, &sub_heads, &dir)?; |
| 159 |
files.extend(sub_files.into_iter().map(|(p, h)| (format!("{}/{p}", sub.path), h))); |
| 160 |
} |
| 161 |
let scope_hash = scope_tree_hash(&claim.job.scope, files.iter().map(|(p, h)| (p.as_str(), *h))); |
| 162 |
workspace::image(&scratch.join("tree"), &scratch.join("workspace.ext4"))?; |
| 163 |
std::fs::remove_dir_all(scratch.join("tree"))?; |
| 164 |
note( |
| 165 |
&mut sink, |
| 166 |
format!( |
| 167 |
"booting a microVM: {} vCPU, {} MiB, rootfs {}", |
| 168 |
claim.sandbox.cpus, |
| 169 |
claim.sandbox.memory_mib, |
| 170 |
&claim.sandbox.rootfs[..12] |
| 171 |
), |
| 172 |
); |
| 173 |
let outcome = microvm::run( |
| 174 |
home, |
| 175 |
&claim.sandbox, |
| 176 |
&claim.job, |
| 177 |
&scratch.join("workspace.ext4"), |
| 178 |
&scratch, |
| 179 |
Duration::from_secs(claim.timeout_s), |
| 180 |
&mut |r| sink.push(r), |
| 181 |
)?; |
| 182 |
let receipt = outcome.exit_code.map(|code| Receipt { |
| 183 |
job_name: claim.job.name.clone(), |
| 184 |
pipeline_hash: hex32(&claim.pipeline_hash), |
| 185 |
scope_tree_hash: scope_hash, |
| 186 |
inputs: vec![], |
| 187 |
environment: outcome.environment.hash(), |
| 188 |
outputs: vec![], |
| 189 |
exit_code: code as u32, |
| 190 |
duration_ms: outcome.duration_ms, |
| 191 |
}); |
| 192 |
Ok(Finish { |
| 193 |
exit_code: outcome.exit_code, |
| 194 |
duration_ms: started.elapsed().as_millis() as u64, |
| 195 |
receipt, |
| 196 |
error: outcome.timed_out.then(|| "stopped at the time limit".to_string()), |
| 197 |
}) |
| 198 |
})(); |
| 199 |
let _ = std::fs::remove_dir_all(&scratch); |
| 200 |
sink.flush(); |
| 201 |
result |
| 202 |
} |
| 203 |
|
| 204 |
fn hex32(s: &str) -> [u8; 32] { |
| 205 |
let mut out = [0; 32]; |
| 206 |
if let Ok(v) = hex::decode(s) { |
| 207 |
if v.len() == 32 { |
| 208 |
out.copy_from_slice(&v); |
| 209 |
} |
| 210 |
} |
| 211 |
out |
| 212 |
} |
| 213 |
|
| 214 |
|
| 215 |
fn cached_repo(home: &Home, cfg: &RunnerConfig, claim: &Claim) -> Result<Repo> { |
| 216 |
cached_repo_at(home, cfg, &claim.repo, &claim.clone_url, &claim.token) |
| 217 |
} |
| 218 |
|
| 219 |
|
| 220 |
fn cached_repo_at(home: &Home, cfg: &RunnerConfig, repo_name: &str, clone_url: &str, token: &str) -> Result<Repo> { |
| 221 |
let dir: PathBuf = home.root.join("repos").join(repo_name.replace('/', "__")); |
| 222 |
let set_remote = |repo: &mut Repo| -> Result<()> { |
| 223 |
repo.config.set("remote.origin.url", clone_url).map_err(|e| anyhow::anyhow!("{e}"))?; |
| 224 |
repo.config.set("remote.origin.token", token).map_err(|e| anyhow::anyhow!("{e}"))?; |
| 225 |
if let Some(ca) = &cfg.ca { |
| 226 |
repo.config.set("remote.origin.ca", &ca.to_string_lossy()).map_err(|e| anyhow::anyhow!("{e}"))?; |
| 227 |
} |
| 228 |
repo.save_config().map_err(|e| anyhow::anyhow!("{e}")) |
| 229 |
}; |
| 230 |
if dir.join(".oxy").is_dir() { |
| 231 |
let mut repo = Repo::open(&dir).map_err(|e| anyhow::anyhow!("{e}"))?; |
| 232 |
set_remote(&mut repo)?; |
| 233 |
repo.fetch("origin").map_err(|e| anyhow::anyhow!("fetch {clone_url}: {e}"))?; |
| 234 |
return Ok(repo); |
| 235 |
} |
| 236 |
|
| 237 |
std::env::set_var("OXY_TOKEN", token); |
| 238 |
if let Some(ca) = &cfg.ca { |
| 239 |
std::env::set_var("OXY_CA", ca); |
| 240 |
} |
| 241 |
let cloned = Repo::clone(clone_url, &dir); |
| 242 |
std::env::remove_var("OXY_TOKEN"); |
| 243 |
std::env::remove_var("OXY_CA"); |
| 244 |
let mut repo = cloned.map_err(|e| { |
| 245 |
let _ = std::fs::remove_dir_all(&dir); |
| 246 |
anyhow::anyhow!("clone {}: {e}", clone_url) |
| 247 |
})?; |
| 248 |
set_remote(&mut repo)?; |
| 249 |
Ok(repo) |
| 250 |
} |