agentguard-local 0.1.0
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/Cargo.lock +727 -0
- package/Cargo.toml +22 -0
- package/README.md +140 -0
- package/crates/agentguard/Cargo.toml +23 -0
- package/crates/agentguard/src/bin/agentguard.rs +460 -0
- package/crates/agentguard/src/bin/agentguardd.rs +23 -0
- package/crates/agentguard/tests/integration.rs +384 -0
- package/crates/agentguard-core/Cargo.toml +19 -0
- package/crates/agentguard-core/src/adapter.rs +69 -0
- package/crates/agentguard-core/src/lib.rs +14 -0
- package/crates/agentguard-core/src/lifecycle.rs +41 -0
- package/crates/agentguard-core/src/policy.rs +117 -0
- package/crates/agentguard-core/src/pressure.rs +186 -0
- package/crates/agentguard-core/src/process.rs +100 -0
- package/crates/agentguard-core/src/protocol.rs +186 -0
- package/crates/agentguard-core/src/scheduler.rs +132 -0
- package/crates/agentguard-core/src/server.rs +613 -0
- package/crates/agentguard-core/src/store.rs +434 -0
- package/crates/agentguard-core/src/types.rs +248 -0
- package/docs/DECISIONS.md +37 -0
- package/docs/TESTING.md +75 -0
- package/npm/bin/agentguard +9 -0
- package/npm/bin/agentguardd +9 -0
- package/npm/bin/darwin-arm64/agentguard +0 -0
- package/npm/bin/darwin-arm64/agentguardd +0 -0
- package/package.json +44 -0
|
@@ -0,0 +1,384 @@
|
|
|
1
|
+
use serde_json::Value;
|
|
2
|
+
use std::fs;
|
|
3
|
+
use std::io;
|
|
4
|
+
use std::os::unix::fs::PermissionsExt;
|
|
5
|
+
use std::path::{Path, PathBuf};
|
|
6
|
+
use std::process::{Child, Command, Output};
|
|
7
|
+
use std::thread;
|
|
8
|
+
use std::time::{Duration, Instant};
|
|
9
|
+
use tempfile::{TempDir, tempdir};
|
|
10
|
+
|
|
11
|
+
struct DaemonGuard {
|
|
12
|
+
child: Child,
|
|
13
|
+
_dir: TempDir,
|
|
14
|
+
socket: PathBuf,
|
|
15
|
+
db: PathBuf,
|
|
16
|
+
pressure_file: PathBuf,
|
|
17
|
+
}
|
|
18
|
+
|
|
19
|
+
impl DaemonGuard {
|
|
20
|
+
fn start() -> Self {
|
|
21
|
+
let dir = tempdir().unwrap();
|
|
22
|
+
let socket = dir.path().join("agentguard.sock");
|
|
23
|
+
let db = dir.path().join("state.db");
|
|
24
|
+
let pressure_file = dir.path().join("pressure.txt");
|
|
25
|
+
fs::write(&pressure_file, "normal").unwrap();
|
|
26
|
+
let child = Command::new(bin("agentguardd"))
|
|
27
|
+
.arg("--socket")
|
|
28
|
+
.arg(&socket)
|
|
29
|
+
.arg("--db")
|
|
30
|
+
.arg(&db)
|
|
31
|
+
.arg("--foreground")
|
|
32
|
+
.env("AGENTGUARD_TEST_MODE", "1")
|
|
33
|
+
.env("AGENTGUARD_FAKE_PRESSURE_FILE", &pressure_file)
|
|
34
|
+
.spawn()
|
|
35
|
+
.unwrap();
|
|
36
|
+
let daemon = Self {
|
|
37
|
+
child,
|
|
38
|
+
_dir: dir,
|
|
39
|
+
socket,
|
|
40
|
+
db,
|
|
41
|
+
pressure_file,
|
|
42
|
+
};
|
|
43
|
+
wait_for(
|
|
44
|
+
|| daemon.status_json_result().is_ok(),
|
|
45
|
+
Duration::from_secs(5),
|
|
46
|
+
)
|
|
47
|
+
.unwrap();
|
|
48
|
+
let output = daemon
|
|
49
|
+
.cli()
|
|
50
|
+
.args(["policy", "set", "recovery_window_seconds", "1"])
|
|
51
|
+
.output()
|
|
52
|
+
.unwrap();
|
|
53
|
+
assert_success(&output);
|
|
54
|
+
daemon
|
|
55
|
+
}
|
|
56
|
+
|
|
57
|
+
fn restart(&mut self) {
|
|
58
|
+
let _ = self.child.kill();
|
|
59
|
+
let _ = self.child.wait();
|
|
60
|
+
self.child = Command::new(bin("agentguardd"))
|
|
61
|
+
.arg("--socket")
|
|
62
|
+
.arg(&self.socket)
|
|
63
|
+
.arg("--db")
|
|
64
|
+
.arg(&self.db)
|
|
65
|
+
.arg("--foreground")
|
|
66
|
+
.env("AGENTGUARD_TEST_MODE", "1")
|
|
67
|
+
.env("AGENTGUARD_FAKE_PRESSURE_FILE", &self.pressure_file)
|
|
68
|
+
.spawn()
|
|
69
|
+
.unwrap();
|
|
70
|
+
wait_for(|| self.status_json_result().is_ok(), Duration::from_secs(5)).unwrap();
|
|
71
|
+
}
|
|
72
|
+
|
|
73
|
+
fn set_pressure(&self, pressure: &str) {
|
|
74
|
+
fs::write(&self.pressure_file, pressure).unwrap();
|
|
75
|
+
}
|
|
76
|
+
|
|
77
|
+
fn cli(&self) -> Command {
|
|
78
|
+
let mut command = Command::new(bin("agentguard"));
|
|
79
|
+
command.arg("--socket").arg(&self.socket);
|
|
80
|
+
command
|
|
81
|
+
}
|
|
82
|
+
|
|
83
|
+
fn status_json(&self) -> Value {
|
|
84
|
+
self.status_json_result().unwrap()
|
|
85
|
+
}
|
|
86
|
+
|
|
87
|
+
fn status_json_result(&self) -> Result<Value, String> {
|
|
88
|
+
let output = self.cli().arg("status").arg("--json").output().unwrap();
|
|
89
|
+
if !output.status.success() {
|
|
90
|
+
return Err(format!(
|
|
91
|
+
"status={:?}\nstdout={}\nstderr={}",
|
|
92
|
+
output.status.code(),
|
|
93
|
+
String::from_utf8_lossy(&output.stdout),
|
|
94
|
+
String::from_utf8_lossy(&output.stderr)
|
|
95
|
+
));
|
|
96
|
+
}
|
|
97
|
+
serde_json::from_slice(&output.stdout).map_err(|err| err.to_string())
|
|
98
|
+
}
|
|
99
|
+
}
|
|
100
|
+
|
|
101
|
+
impl Drop for DaemonGuard {
|
|
102
|
+
fn drop(&mut self) {
|
|
103
|
+
let _ = self.child.kill();
|
|
104
|
+
let _ = self.child.wait();
|
|
105
|
+
}
|
|
106
|
+
}
|
|
107
|
+
|
|
108
|
+
#[test]
|
|
109
|
+
fn status_json_reports_pressure_policy_and_empty_job_state() {
|
|
110
|
+
let daemon = DaemonGuard::start();
|
|
111
|
+
let status = daemon.status_json();
|
|
112
|
+
assert_eq!(status["pressure"], "normal");
|
|
113
|
+
assert_eq!(status["policy"]["max_active_jobs"], 1);
|
|
114
|
+
assert!(status["jobs"].as_array().unwrap().is_empty());
|
|
115
|
+
}
|
|
116
|
+
|
|
117
|
+
#[test]
|
|
118
|
+
fn run_fake_tool_records_job_and_preserves_stdio_and_exit_code() {
|
|
119
|
+
let daemon = DaemonGuard::start();
|
|
120
|
+
let output = daemon
|
|
121
|
+
.cli()
|
|
122
|
+
.args([
|
|
123
|
+
"run",
|
|
124
|
+
"--",
|
|
125
|
+
"/bin/sh",
|
|
126
|
+
"-c",
|
|
127
|
+
"echo out; echo err >&2; exit 7",
|
|
128
|
+
])
|
|
129
|
+
.output()
|
|
130
|
+
.unwrap();
|
|
131
|
+
assert_eq!(output.status.code(), Some(7));
|
|
132
|
+
assert_eq!(String::from_utf8_lossy(&output.stdout).trim(), "out");
|
|
133
|
+
assert_eq!(String::from_utf8_lossy(&output.stderr).trim(), "err");
|
|
134
|
+
|
|
135
|
+
let status = daemon.status_json();
|
|
136
|
+
let jobs = status["jobs"].as_array().unwrap();
|
|
137
|
+
assert_eq!(jobs[0]["status"], "failed");
|
|
138
|
+
assert_eq!(jobs[0]["exit_code"], 7);
|
|
139
|
+
}
|
|
140
|
+
|
|
141
|
+
#[test]
|
|
142
|
+
fn active_job_limit_queues_second_job_until_first_exits() {
|
|
143
|
+
let daemon = DaemonGuard::start();
|
|
144
|
+
let mut first = daemon
|
|
145
|
+
.cli()
|
|
146
|
+
.args(["run", "--", "/bin/sleep", "1"])
|
|
147
|
+
.spawn()
|
|
148
|
+
.unwrap();
|
|
149
|
+
wait_for(
|
|
150
|
+
|| status_has_state(&daemon, "running"),
|
|
151
|
+
Duration::from_secs(5),
|
|
152
|
+
)
|
|
153
|
+
.unwrap();
|
|
154
|
+
|
|
155
|
+
let mut second = daemon
|
|
156
|
+
.cli()
|
|
157
|
+
.args(["run", "--", "/bin/sh", "-c", "echo second"])
|
|
158
|
+
.spawn()
|
|
159
|
+
.unwrap();
|
|
160
|
+
wait_for(
|
|
161
|
+
|| status_has_state(&daemon, "queued"),
|
|
162
|
+
Duration::from_secs(5),
|
|
163
|
+
)
|
|
164
|
+
.unwrap();
|
|
165
|
+
|
|
166
|
+
assert_eq!(
|
|
167
|
+
wait_child(&mut first, Duration::from_secs(5)).unwrap(),
|
|
168
|
+
Some(0)
|
|
169
|
+
);
|
|
170
|
+
assert_eq!(
|
|
171
|
+
wait_child(&mut second, Duration::from_secs(5)).unwrap(),
|
|
172
|
+
Some(0)
|
|
173
|
+
);
|
|
174
|
+
let status = daemon.status_json();
|
|
175
|
+
assert!(status["jobs"].as_array().unwrap().iter().any(|job| {
|
|
176
|
+
job["command"] == "/bin/sh" && job["status"] == "exited" && job["exit_code"] == 0
|
|
177
|
+
}));
|
|
178
|
+
}
|
|
179
|
+
|
|
180
|
+
#[test]
|
|
181
|
+
fn simulated_soft_pressure_queues_then_recovery_admits_job() {
|
|
182
|
+
let daemon = DaemonGuard::start();
|
|
183
|
+
daemon.set_pressure("soft");
|
|
184
|
+
let mut queued = daemon
|
|
185
|
+
.cli()
|
|
186
|
+
.args(["run", "--", "/bin/sh", "-c", "echo recovered"])
|
|
187
|
+
.stdout(std::process::Stdio::piped())
|
|
188
|
+
.spawn()
|
|
189
|
+
.unwrap();
|
|
190
|
+
wait_for(
|
|
191
|
+
|| status_has_state(&daemon, "queued"),
|
|
192
|
+
Duration::from_secs(5),
|
|
193
|
+
)
|
|
194
|
+
.unwrap();
|
|
195
|
+
let queued_status = daemon.status_json();
|
|
196
|
+
assert!(queued_status["jobs"].as_array().unwrap().iter().any(|job| {
|
|
197
|
+
job["queued_reason"]
|
|
198
|
+
.as_str()
|
|
199
|
+
.is_some_and(|reason| reason.contains("soft memory pressure"))
|
|
200
|
+
}));
|
|
201
|
+
|
|
202
|
+
daemon.set_pressure("normal");
|
|
203
|
+
let code = wait_child(&mut queued, Duration::from_secs(5)).unwrap();
|
|
204
|
+
assert_eq!(code, Some(0));
|
|
205
|
+
}
|
|
206
|
+
|
|
207
|
+
#[test]
|
|
208
|
+
fn simulated_critical_pressure_pauses_and_recovery_resumes_guarded_group() {
|
|
209
|
+
let daemon = DaemonGuard::start();
|
|
210
|
+
let mut child = daemon
|
|
211
|
+
.cli()
|
|
212
|
+
.args(["run", "--priority", "low", "--", "/bin/sleep", "3"])
|
|
213
|
+
.spawn()
|
|
214
|
+
.unwrap();
|
|
215
|
+
wait_for(
|
|
216
|
+
|| status_has_state(&daemon, "running"),
|
|
217
|
+
Duration::from_secs(5),
|
|
218
|
+
)
|
|
219
|
+
.unwrap();
|
|
220
|
+
daemon.set_pressure("critical");
|
|
221
|
+
wait_for(
|
|
222
|
+
|| {
|
|
223
|
+
let _ = daemon.status_json();
|
|
224
|
+
status_has_state(&daemon, "paused_by_guard")
|
|
225
|
+
},
|
|
226
|
+
Duration::from_secs(5),
|
|
227
|
+
)
|
|
228
|
+
.unwrap();
|
|
229
|
+
|
|
230
|
+
daemon.set_pressure("normal");
|
|
231
|
+
wait_for(
|
|
232
|
+
|| {
|
|
233
|
+
let _ = daemon.status_json();
|
|
234
|
+
status_has_state(&daemon, "running") || child.try_wait().unwrap().is_some()
|
|
235
|
+
},
|
|
236
|
+
Duration::from_secs(5),
|
|
237
|
+
)
|
|
238
|
+
.unwrap();
|
|
239
|
+
let _ = child.kill();
|
|
240
|
+
let _ = child.wait();
|
|
241
|
+
}
|
|
242
|
+
|
|
243
|
+
#[test]
|
|
244
|
+
fn daemon_restart_reconciles_still_running_job() {
|
|
245
|
+
let mut daemon = DaemonGuard::start();
|
|
246
|
+
let mut child = daemon
|
|
247
|
+
.cli()
|
|
248
|
+
.args(["run", "--", "/bin/sleep", "3"])
|
|
249
|
+
.spawn()
|
|
250
|
+
.unwrap();
|
|
251
|
+
wait_for(
|
|
252
|
+
|| status_has_state(&daemon, "running"),
|
|
253
|
+
Duration::from_secs(5),
|
|
254
|
+
)
|
|
255
|
+
.unwrap();
|
|
256
|
+
|
|
257
|
+
daemon.restart();
|
|
258
|
+
wait_for(
|
|
259
|
+
|| status_has_event(&daemon, "daemon_restart_reconciled"),
|
|
260
|
+
Duration::from_secs(5),
|
|
261
|
+
)
|
|
262
|
+
.unwrap();
|
|
263
|
+
let status = daemon.status_json();
|
|
264
|
+
assert!(
|
|
265
|
+
status["jobs"]
|
|
266
|
+
.as_array()
|
|
267
|
+
.unwrap()
|
|
268
|
+
.iter()
|
|
269
|
+
.any(|job| { job["status"] == "running" || job["status"] == "exited" })
|
|
270
|
+
);
|
|
271
|
+
let _ = child.kill();
|
|
272
|
+
let _ = child.wait();
|
|
273
|
+
}
|
|
274
|
+
|
|
275
|
+
#[test]
|
|
276
|
+
fn shim_wraps_fake_codex_without_real_codex_and_injects_thread_cap() {
|
|
277
|
+
let daemon = DaemonGuard::start();
|
|
278
|
+
let dir = tempdir().unwrap();
|
|
279
|
+
let real_codex = dir.path().join("codex-real");
|
|
280
|
+
fs::write(&real_codex, "#!/bin/sh\nprintf '%s\\n' \"$@\"\n").unwrap();
|
|
281
|
+
fs::set_permissions(&real_codex, fs::Permissions::from_mode(0o755)).unwrap();
|
|
282
|
+
let shim_dir = dir.path().join("shim");
|
|
283
|
+
|
|
284
|
+
let install = daemon
|
|
285
|
+
.cli()
|
|
286
|
+
.arg("install-shim")
|
|
287
|
+
.arg("codex")
|
|
288
|
+
.arg("--real-path")
|
|
289
|
+
.arg(&real_codex)
|
|
290
|
+
.arg("--shim-dir")
|
|
291
|
+
.arg(&shim_dir)
|
|
292
|
+
.output()
|
|
293
|
+
.unwrap();
|
|
294
|
+
assert_success(&install);
|
|
295
|
+
|
|
296
|
+
let output = Command::new(shim_dir.join("codex"))
|
|
297
|
+
.arg("exec")
|
|
298
|
+
.arg("hello")
|
|
299
|
+
.output()
|
|
300
|
+
.unwrap();
|
|
301
|
+
assert_success(&output);
|
|
302
|
+
let stdout = String::from_utf8_lossy(&output.stdout);
|
|
303
|
+
assert!(stdout.contains("agents.max_threads=1"));
|
|
304
|
+
assert!(stdout.contains("exec"));
|
|
305
|
+
|
|
306
|
+
let output = Command::new(shim_dir.join("codex"))
|
|
307
|
+
.args(["-c", "agents.max_threads=9", "exec"])
|
|
308
|
+
.output()
|
|
309
|
+
.unwrap();
|
|
310
|
+
assert_success(&output);
|
|
311
|
+
let stdout = String::from_utf8_lossy(&output.stdout);
|
|
312
|
+
assert!(stdout.contains("agents.max_threads=9"));
|
|
313
|
+
assert!(!stdout.contains("agents.max_threads=1"));
|
|
314
|
+
}
|
|
315
|
+
|
|
316
|
+
fn bin(name: &str) -> PathBuf {
|
|
317
|
+
let key = format!("CARGO_BIN_EXE_{name}");
|
|
318
|
+
std::env::var_os(&key)
|
|
319
|
+
.map(PathBuf::from)
|
|
320
|
+
.unwrap_or_else(|| {
|
|
321
|
+
let exe = std::env::current_exe().unwrap();
|
|
322
|
+
exe.parent().and_then(Path::parent).unwrap().join(name)
|
|
323
|
+
})
|
|
324
|
+
}
|
|
325
|
+
|
|
326
|
+
fn wait_for(mut predicate: impl FnMut() -> bool, timeout: Duration) -> io::Result<()> {
|
|
327
|
+
let start = Instant::now();
|
|
328
|
+
while start.elapsed() < timeout {
|
|
329
|
+
if predicate() {
|
|
330
|
+
return Ok(());
|
|
331
|
+
}
|
|
332
|
+
thread::sleep(Duration::from_millis(50));
|
|
333
|
+
}
|
|
334
|
+
Err(io::Error::new(
|
|
335
|
+
io::ErrorKind::TimedOut,
|
|
336
|
+
"condition timed out",
|
|
337
|
+
))
|
|
338
|
+
}
|
|
339
|
+
|
|
340
|
+
fn status_has_state(daemon: &DaemonGuard, state: &str) -> bool {
|
|
341
|
+
let Ok(status) = daemon.status_json_result() else {
|
|
342
|
+
return false;
|
|
343
|
+
};
|
|
344
|
+
status["jobs"]
|
|
345
|
+
.as_array()
|
|
346
|
+
.unwrap()
|
|
347
|
+
.iter()
|
|
348
|
+
.any(|job| job["status"] == state)
|
|
349
|
+
}
|
|
350
|
+
|
|
351
|
+
fn status_has_event(daemon: &DaemonGuard, event_type: &str) -> bool {
|
|
352
|
+
let Ok(status) = daemon.status_json_result() else {
|
|
353
|
+
return false;
|
|
354
|
+
};
|
|
355
|
+
status["events"]
|
|
356
|
+
.as_array()
|
|
357
|
+
.unwrap()
|
|
358
|
+
.iter()
|
|
359
|
+
.any(|event| event["event_type"] == event_type)
|
|
360
|
+
}
|
|
361
|
+
|
|
362
|
+
fn wait_child(child: &mut Child, timeout: Duration) -> io::Result<Option<i32>> {
|
|
363
|
+
let start = Instant::now();
|
|
364
|
+
loop {
|
|
365
|
+
if let Some(status) = child.try_wait()? {
|
|
366
|
+
return Ok(status.code());
|
|
367
|
+
}
|
|
368
|
+
if start.elapsed() >= timeout {
|
|
369
|
+
let _ = child.kill();
|
|
370
|
+
return Ok(None);
|
|
371
|
+
}
|
|
372
|
+
thread::sleep(Duration::from_millis(50));
|
|
373
|
+
}
|
|
374
|
+
}
|
|
375
|
+
|
|
376
|
+
fn assert_success(output: &Output) {
|
|
377
|
+
assert!(
|
|
378
|
+
output.status.success(),
|
|
379
|
+
"status={:?}\nstdout={}\nstderr={}",
|
|
380
|
+
output.status.code(),
|
|
381
|
+
String::from_utf8_lossy(&output.stdout),
|
|
382
|
+
String::from_utf8_lossy(&output.stderr)
|
|
383
|
+
);
|
|
384
|
+
}
|
|
@@ -0,0 +1,19 @@
|
|
|
1
|
+
[package]
|
|
2
|
+
name = "agentguard-core"
|
|
3
|
+
version.workspace = true
|
|
4
|
+
edition.workspace = true
|
|
5
|
+
license.workspace = true
|
|
6
|
+
authors.workspace = true
|
|
7
|
+
|
|
8
|
+
[dependencies]
|
|
9
|
+
anyhow.workspace = true
|
|
10
|
+
chrono.workspace = true
|
|
11
|
+
libc.workspace = true
|
|
12
|
+
rusqlite.workspace = true
|
|
13
|
+
serde.workspace = true
|
|
14
|
+
serde_json.workspace = true
|
|
15
|
+
thiserror.workspace = true
|
|
16
|
+
uuid.workspace = true
|
|
17
|
+
|
|
18
|
+
[dev-dependencies]
|
|
19
|
+
tempfile.workspace = true
|
|
@@ -0,0 +1,69 @@
|
|
|
1
|
+
pub mod codex {
|
|
2
|
+
pub fn apply_thread_cap(args: &[String], cap: u32) -> Vec<String> {
|
|
3
|
+
if has_explicit_thread_cap(args) {
|
|
4
|
+
return args.to_vec();
|
|
5
|
+
}
|
|
6
|
+
let mut adjusted = vec!["-c".to_string(), format!("agents.max_threads={cap}")];
|
|
7
|
+
adjusted.extend_from_slice(args);
|
|
8
|
+
adjusted
|
|
9
|
+
}
|
|
10
|
+
|
|
11
|
+
pub fn has_explicit_thread_cap(args: &[String]) -> bool {
|
|
12
|
+
let mut index = 0;
|
|
13
|
+
while index < args.len() {
|
|
14
|
+
let arg = &args[index];
|
|
15
|
+
if arg == "-c" || arg == "--config" {
|
|
16
|
+
if args
|
|
17
|
+
.get(index + 1)
|
|
18
|
+
.is_some_and(|next| config_sets_threads(next))
|
|
19
|
+
{
|
|
20
|
+
return true;
|
|
21
|
+
}
|
|
22
|
+
index += 2;
|
|
23
|
+
continue;
|
|
24
|
+
}
|
|
25
|
+
if arg.strip_prefix("-c=").is_some_and(config_sets_threads)
|
|
26
|
+
|| arg
|
|
27
|
+
.strip_prefix("--config=")
|
|
28
|
+
.is_some_and(config_sets_threads)
|
|
29
|
+
|| arg.strip_prefix("-c").is_some_and(config_sets_threads)
|
|
30
|
+
{
|
|
31
|
+
return true;
|
|
32
|
+
}
|
|
33
|
+
index += 1;
|
|
34
|
+
}
|
|
35
|
+
false
|
|
36
|
+
}
|
|
37
|
+
|
|
38
|
+
fn config_sets_threads(value: &str) -> bool {
|
|
39
|
+
let trimmed = value.trim();
|
|
40
|
+
trimmed == "agents.max_threads" || trimmed.starts_with("agents.max_threads=")
|
|
41
|
+
}
|
|
42
|
+
|
|
43
|
+
#[cfg(test)]
|
|
44
|
+
mod tests {
|
|
45
|
+
use super::*;
|
|
46
|
+
|
|
47
|
+
#[test]
|
|
48
|
+
fn codex_adapter_injects_thread_cap_when_absent() {
|
|
49
|
+
let args = vec!["exec".to_string(), "fix tests".to_string()];
|
|
50
|
+
let adjusted = apply_thread_cap(&args, 1);
|
|
51
|
+
assert_eq!(adjusted[0], "-c");
|
|
52
|
+
assert_eq!(adjusted[1], "agents.max_threads=1");
|
|
53
|
+
assert_eq!(adjusted[2], "exec");
|
|
54
|
+
}
|
|
55
|
+
|
|
56
|
+
#[test]
|
|
57
|
+
fn codex_adapter_does_not_override_explicit_config() {
|
|
58
|
+
let args = vec![
|
|
59
|
+
"-c".to_string(),
|
|
60
|
+
"agents.max_threads=4".to_string(),
|
|
61
|
+
"exec".to_string(),
|
|
62
|
+
];
|
|
63
|
+
assert_eq!(apply_thread_cap(&args, 1), args);
|
|
64
|
+
|
|
65
|
+
let args = vec!["--config=agents.max_threads=3".to_string()];
|
|
66
|
+
assert_eq!(apply_thread_cap(&args, 1), args);
|
|
67
|
+
}
|
|
68
|
+
}
|
|
69
|
+
}
|
|
@@ -0,0 +1,14 @@
|
|
|
1
|
+
pub mod adapter;
|
|
2
|
+
pub mod lifecycle;
|
|
3
|
+
pub mod policy;
|
|
4
|
+
pub mod pressure;
|
|
5
|
+
pub mod process;
|
|
6
|
+
pub mod protocol;
|
|
7
|
+
pub mod scheduler;
|
|
8
|
+
pub mod server;
|
|
9
|
+
pub mod store;
|
|
10
|
+
pub mod types;
|
|
11
|
+
|
|
12
|
+
pub use types::{
|
|
13
|
+
EventRecord, JobRecord, JobState, Policy, PressureState, Priority, StatusSnapshot, ToolPath,
|
|
14
|
+
};
|
|
@@ -0,0 +1,41 @@
|
|
|
1
|
+
use crate::types::JobState;
|
|
2
|
+
|
|
3
|
+
pub fn valid_transition(from: JobState, to: JobState) -> bool {
|
|
4
|
+
use JobState::*;
|
|
5
|
+
match (from, to) {
|
|
6
|
+
(PendingAdmission, Queued | Starting | Cancelled | Failed) => true,
|
|
7
|
+
(Queued, Starting | Cancelled | Lost) => true,
|
|
8
|
+
(Starting, Running | Failed | Lost | Cancelled) => true,
|
|
9
|
+
(Running, PausedByGuard | Exiting | Exited | Failed | Lost | Cancelled) => true,
|
|
10
|
+
(PausedByGuard, Running | Exiting | Lost | Cancelled) => true,
|
|
11
|
+
(Exiting, Exited | Failed | Lost | Cancelled) => true,
|
|
12
|
+
(Exited | Failed | Lost | Cancelled, _) => false,
|
|
13
|
+
(same_from, same_to) if same_from == same_to => true,
|
|
14
|
+
_ => false,
|
|
15
|
+
}
|
|
16
|
+
}
|
|
17
|
+
|
|
18
|
+
pub fn ensure_transition(from: JobState, to: JobState) -> anyhow::Result<()> {
|
|
19
|
+
anyhow::ensure!(
|
|
20
|
+
valid_transition(from, to),
|
|
21
|
+
"invalid job lifecycle transition from {from} to {to}"
|
|
22
|
+
);
|
|
23
|
+
Ok(())
|
|
24
|
+
}
|
|
25
|
+
|
|
26
|
+
#[cfg(test)]
|
|
27
|
+
mod tests {
|
|
28
|
+
use super::*;
|
|
29
|
+
|
|
30
|
+
#[test]
|
|
31
|
+
fn validates_job_lifecycle_transitions() {
|
|
32
|
+
assert!(valid_transition(
|
|
33
|
+
JobState::PendingAdmission,
|
|
34
|
+
JobState::Queued
|
|
35
|
+
));
|
|
36
|
+
assert!(valid_transition(JobState::Queued, JobState::Starting));
|
|
37
|
+
assert!(valid_transition(JobState::Running, JobState::PausedByGuard));
|
|
38
|
+
assert!(!valid_transition(JobState::Exited, JobState::Running));
|
|
39
|
+
assert!(!valid_transition(JobState::Queued, JobState::Exited));
|
|
40
|
+
}
|
|
41
|
+
}
|
|
@@ -0,0 +1,117 @@
|
|
|
1
|
+
use crate::types::Policy;
|
|
2
|
+
|
|
3
|
+
impl Default for Policy {
|
|
4
|
+
fn default() -> Self {
|
|
5
|
+
Self::for_physical_memory(8 * 1024 * 1024 * 1024)
|
|
6
|
+
}
|
|
7
|
+
}
|
|
8
|
+
|
|
9
|
+
impl Policy {
|
|
10
|
+
pub fn for_physical_memory(bytes: u64) -> Self {
|
|
11
|
+
let gib = bytes / 1024 / 1024 / 1024;
|
|
12
|
+
if gib <= 8 {
|
|
13
|
+
Self {
|
|
14
|
+
max_active_jobs: 1,
|
|
15
|
+
soft_memory_pressure_percent: 72,
|
|
16
|
+
critical_memory_pressure_percent: 88,
|
|
17
|
+
recovery_memory_pressure_percent: 62,
|
|
18
|
+
default_codex_max_threads: 1,
|
|
19
|
+
pause_enabled: true,
|
|
20
|
+
queue_timeout_seconds: 1800,
|
|
21
|
+
recovery_window_seconds: 60,
|
|
22
|
+
}
|
|
23
|
+
} else if gib <= 16 {
|
|
24
|
+
Self {
|
|
25
|
+
max_active_jobs: 2,
|
|
26
|
+
soft_memory_pressure_percent: 80,
|
|
27
|
+
critical_memory_pressure_percent: 92,
|
|
28
|
+
recovery_memory_pressure_percent: 70,
|
|
29
|
+
default_codex_max_threads: 2,
|
|
30
|
+
pause_enabled: true,
|
|
31
|
+
queue_timeout_seconds: 1800,
|
|
32
|
+
recovery_window_seconds: 30,
|
|
33
|
+
}
|
|
34
|
+
} else {
|
|
35
|
+
Self {
|
|
36
|
+
max_active_jobs: 3,
|
|
37
|
+
soft_memory_pressure_percent: 85,
|
|
38
|
+
critical_memory_pressure_percent: 94,
|
|
39
|
+
recovery_memory_pressure_percent: 75,
|
|
40
|
+
default_codex_max_threads: 3,
|
|
41
|
+
pause_enabled: true,
|
|
42
|
+
queue_timeout_seconds: 1800,
|
|
43
|
+
recovery_window_seconds: 30,
|
|
44
|
+
}
|
|
45
|
+
}
|
|
46
|
+
}
|
|
47
|
+
|
|
48
|
+
pub fn set_key(&mut self, key: &str, value: &str) -> anyhow::Result<()> {
|
|
49
|
+
match key {
|
|
50
|
+
"max_active_jobs" => self.max_active_jobs = value.parse()?,
|
|
51
|
+
"soft_memory_pressure_percent" => self.soft_memory_pressure_percent = value.parse()?,
|
|
52
|
+
"critical_memory_pressure_percent" => {
|
|
53
|
+
self.critical_memory_pressure_percent = value.parse()?;
|
|
54
|
+
}
|
|
55
|
+
"recovery_memory_pressure_percent" => {
|
|
56
|
+
self.recovery_memory_pressure_percent = value.parse()?;
|
|
57
|
+
}
|
|
58
|
+
"default_codex_max_threads" => self.default_codex_max_threads = value.parse()?,
|
|
59
|
+
"pause_enabled" => self.pause_enabled = parse_bool(value)?,
|
|
60
|
+
"queue_timeout_seconds" => self.queue_timeout_seconds = value.parse()?,
|
|
61
|
+
"recovery_window_seconds" => self.recovery_window_seconds = value.parse()?,
|
|
62
|
+
other => anyhow::bail!("unknown policy key {other:?}"),
|
|
63
|
+
}
|
|
64
|
+
self.validate()
|
|
65
|
+
}
|
|
66
|
+
|
|
67
|
+
pub fn validate(&self) -> anyhow::Result<()> {
|
|
68
|
+
anyhow::ensure!(self.max_active_jobs > 0, "max_active_jobs must be > 0");
|
|
69
|
+
anyhow::ensure!(
|
|
70
|
+
self.recovery_memory_pressure_percent < self.soft_memory_pressure_percent,
|
|
71
|
+
"recovery threshold must be below soft threshold"
|
|
72
|
+
);
|
|
73
|
+
anyhow::ensure!(
|
|
74
|
+
self.soft_memory_pressure_percent < self.critical_memory_pressure_percent,
|
|
75
|
+
"soft threshold must be below critical threshold"
|
|
76
|
+
);
|
|
77
|
+
anyhow::ensure!(
|
|
78
|
+
self.critical_memory_pressure_percent <= 100,
|
|
79
|
+
"critical threshold must be <= 100"
|
|
80
|
+
);
|
|
81
|
+
anyhow::ensure!(
|
|
82
|
+
self.default_codex_max_threads > 0,
|
|
83
|
+
"default_codex_max_threads must be > 0"
|
|
84
|
+
);
|
|
85
|
+
Ok(())
|
|
86
|
+
}
|
|
87
|
+
}
|
|
88
|
+
|
|
89
|
+
fn parse_bool(value: &str) -> anyhow::Result<bool> {
|
|
90
|
+
match value {
|
|
91
|
+
"true" | "1" | "yes" | "on" => Ok(true),
|
|
92
|
+
"false" | "0" | "no" | "off" => Ok(false),
|
|
93
|
+
_ => anyhow::bail!("expected boolean value, got {value:?}"),
|
|
94
|
+
}
|
|
95
|
+
}
|
|
96
|
+
|
|
97
|
+
#[cfg(test)]
|
|
98
|
+
mod tests {
|
|
99
|
+
use super::*;
|
|
100
|
+
|
|
101
|
+
#[test]
|
|
102
|
+
fn selects_8gb_defaults_for_small_machines() {
|
|
103
|
+
let policy = Policy::for_physical_memory(8 * 1024 * 1024 * 1024);
|
|
104
|
+
assert_eq!(policy.max_active_jobs, 1);
|
|
105
|
+
assert_eq!(policy.default_codex_max_threads, 1);
|
|
106
|
+
assert!(policy.pause_enabled);
|
|
107
|
+
}
|
|
108
|
+
|
|
109
|
+
#[test]
|
|
110
|
+
fn updates_mvp_policy_keys() {
|
|
111
|
+
let mut policy = Policy::default();
|
|
112
|
+
policy.set_key("max_active_jobs", "2").unwrap();
|
|
113
|
+
policy.set_key("pause_enabled", "false").unwrap();
|
|
114
|
+
assert_eq!(policy.max_active_jobs, 2);
|
|
115
|
+
assert!(!policy.pause_enabled);
|
|
116
|
+
}
|
|
117
|
+
}
|