start_command/
exit_reason.rs

1//! Exit reason hints for finished executions.
2//!
3//! A numeric exit code alone is frequently misleading: a Bun/Node process that
4//! exhausts its heap aborts itself, so the container reports `exitCode 139`
5//! with `OOMKilled false` even though the run died of memory exhaustion (issue
6//! #162, related to #144, #148 and #151). `start` already reads the tail of the
7//! log to find the terminal footer, so the same tail is scanned for well-known
8//! fatal markers and the finding is surfaced as an extra `exitReason` field.
9//!
10//! The same scan answers a narrower question for consumers that key off
11//! `oomKilled`: was this run killed by memory exhaustion at all? That answer is
12//! surfaced as `memoryExhausted` plus `memoryExhaustedReason`, the log line that
13//! carried the evidence (issue #165).
14//!
15//! Both fields are hints, never verdicts: they never change `status`,
16//! `exitCode` or `oomKilled`.
17
18use regex::Regex;
19
20/// Fatal markers, most specific first. The first matching entry wins.
21const EXIT_REASON_MARKERS: [(&str, &str); 5] = [
22    (
23        "memory-exhaustion (v8-heap-limit)",
24        r"(?i)FATAL ERROR:[^\r\n]*(?:Reached heap limit|JavaScript heap out of memory)",
25    ),
26    (
27        "memory-exhaustion (v8-heap-limit)",
28        r"(?i)JavaScript heap out of memory",
29    ),
30    (
31        "memory-exhaustion (kernel-oom-killer)",
32        r"(?i)Out of memory: Kill(?:ed)? process|oom-kill(?:er)?[: ]|Killed process \d+",
33    ),
34    (
35        "memory-exhaustion (go-runtime)",
36        r"(?i)fatal error: runtime: out of memory",
37    ),
38    (
39        "memory-exhaustion (allocation-failure)",
40        r"(?i)std::bad_alloc|Cannot allocate memory|memory allocation of \d+ bytes failed|Allocation failed - process out of memory|Array buffer allocation failed",
41    ),
42];
43
44/// Exit codes above 128 encode the signal that killed the process.
45/// Only the signals that actually show up in command logs are named.
46///
47/// Public because the detached completion watcher generates a POSIX `case`
48/// from this very table, so the signal the watcher writes into the log and the
49/// one `--status` reports can never disagree (issue #171.4).
50pub const SIGNAL_NAMES: [(i32, &str); 10] = [
51    (1, "SIGHUP"),
52    (2, "SIGINT"),
53    (3, "SIGQUIT"),
54    (4, "SIGILL"),
55    (6, "SIGABRT"),
56    (8, "SIGFPE"),
57    (9, "SIGKILL"),
58    (11, "SIGSEGV"),
59    (13, "SIGPIPE"),
60    (15, "SIGTERM"),
61];
62
63/// Reasons carrying this prefix report memory exhaustion, whatever the
64/// mechanism (runtime self-abort, kernel OOM killer, failed allocation).
65const MEMORY_EXHAUSTION_PREFIX: &str = "memory-exhaustion";
66
67/// A matched marker line longer than this is truncated: the evidence travels in
68/// `--status` output, and some runtimes print very long single-line traces.
69const MAX_MARKER_LINE_LENGTH: usize = 300;
70
71/// A fatal marker found in the log: its category and the line that carried it.
72#[derive(Debug, Clone, PartialEq, Eq)]
73pub struct ExitReasonMarker {
74    pub reason: String,
75    pub line: String,
76}
77
78/// The memory-exhaustion observation derived from a log tail (issue #165).
79#[derive(Debug, Clone, PartialEq, Eq)]
80pub struct MemoryExhaustion {
81    pub memory_exhausted: bool,
82    pub memory_exhausted_reason: String,
83}
84
85/// Extract the whole (trimmed, length-bounded) line containing byte `index`.
86fn extract_line_at(text: &str, index: usize) -> String {
87    let start = text[..index].rfind('\n').map(|i| i + 1).unwrap_or(0);
88    let end = text[index..]
89        .find('\n')
90        .map(|i| index + i)
91        .unwrap_or(text.len());
92    let line = text[start..end].trim();
93    if line.chars().count() <= MAX_MARKER_LINE_LENGTH {
94        return line.to_string();
95    }
96    let truncated: String = line.chars().take(MAX_MARKER_LINE_LENGTH).collect();
97    format!("{}...", truncated)
98}
99
100/// Scan log text for the first accepted fatal marker, most specific first.
101pub fn find_exit_reason_marker(
102    text: Option<&str>,
103    accept: impl Fn(&str) -> bool,
104) -> Option<ExitReasonMarker> {
105    let text = text?;
106    if text.is_empty() {
107        return None;
108    }
109    for (reason, pattern) in EXIT_REASON_MARKERS {
110        if !accept(reason) {
111            continue;
112        }
113        if let Ok(re) = Regex::new(pattern) {
114            if let Some(found) = re.find(text) {
115                return Some(ExitReasonMarker {
116                    reason: reason.to_string(),
117                    line: extract_line_at(text, found.start()),
118                });
119            }
120        }
121    }
122    None
123}
124
125/// Scan log text for a known fatal marker.
126pub fn detect_exit_reason(text: Option<&str>) -> Option<String> {
127    find_exit_reason_marker(text, |_| true).map(|marker| marker.reason)
128}
129
130/// Scan log text for a marker that reports memory exhaustion specifically.
131pub fn detect_memory_marker(text: Option<&str>) -> Option<ExitReasonMarker> {
132    find_exit_reason_marker(text, |reason| reason.starts_with(MEMORY_EXHAUSTION_PREFIX))
133}
134
135/// Decide whether an execution died of memory exhaustion, and say why.
136///
137/// Only abnormal exits are explained: a command that merely printed a fatal
138/// marker (an `rg` dump, a quoted incident log) and then succeeded is not a
139/// memory failure. An observation, never a verdict (issue #151).
140pub fn resolve_memory_exhaustion(
141    exit_code: Option<i32>,
142    log_tail: Option<&str>,
143    oom_killed: Option<bool>,
144) -> Option<MemoryExhaustion> {
145    match exit_code {
146        Some(code) if code != 0 => {}
147        _ => return None,
148    }
149    if let Some(marker) = detect_memory_marker(log_tail) {
150        return Some(MemoryExhaustion {
151            memory_exhausted: true,
152            memory_exhausted_reason: marker.line,
153        });
154    }
155    if oom_killed == Some(true) {
156        return Some(MemoryExhaustion {
157            memory_exhausted: true,
158            memory_exhausted_reason: "Docker reported State.OOMKilled=true".to_string(),
159        });
160    }
161    None
162}
163
164/// Text used wherever a fact could not be observed at all.
165pub const UNKNOWN_EXIT_CODE: &str = "unknown";
166
167/// A decoded exit code: the numeric value, the signal it encodes (if any), and
168/// the human-readable rendering both the container post-mortem and `--status`
169/// print.
170#[derive(Debug, Clone, PartialEq, Eq)]
171pub struct ExitCodeDescription {
172    pub code: Option<i32>,
173    pub signal: Option<String>,
174    pub text: String,
175}
176
177/// Decode an exit code once, for every caller (issue #171.4).
178///
179/// `128 + n` was previously decoded in the status formatter and described in
180/// prose by the watcher, which is how the same `137` could be reported as
181/// `SIGKILL` in one place and as a bare number in the other.
182pub fn describe_exit_code(exit_code: Option<i32>) -> ExitCodeDescription {
183    let code = match exit_code {
184        Some(code) => code,
185        None => {
186            return ExitCodeDescription {
187                code: None,
188                signal: None,
189                text: UNKNOWN_EXIT_CODE.to_string(),
190            }
191        }
192    };
193    let signal = signal_name_for_exit_code(Some(code));
194    let text = match &signal {
195        Some(name) => format!("{} ({} - 128+{})", code, name, code - 128),
196        None => code.to_string(),
197    };
198    ExitCodeDescription {
199        code: Some(code),
200        signal,
201        text,
202    }
203}
204
205/// Decode an exit code that arrived as text (e.g. from the watcher's shell).
206pub fn describe_exit_code_str(value: &str) -> ExitCodeDescription {
207    describe_exit_code(value.trim().parse::<i32>().ok())
208}
209
210/// Map a shell exit code to the signal name it encodes.
211pub fn signal_name_for_exit_code(exit_code: Option<i32>) -> Option<String> {
212    let code = exit_code?;
213    if code <= 128 || code > 128 + 64 {
214        return None;
215    }
216    SIGNAL_NAMES
217        .iter()
218        .find(|(signal, _)| *signal == code - 128)
219        .map(|(_, name)| name.to_string())
220}
221
222/// Resolve the best available hint for why an execution ended.
223///
224/// Precedence: the log marker (evidence written by the command itself), then
225/// the cgroup OOM observation, then the signal encoded in the exit code.
226pub fn resolve_exit_reason(
227    exit_code: Option<i32>,
228    log_tail: Option<&str>,
229    oom_killed: Option<bool>,
230) -> Option<String> {
231    if let Some(reason) = detect_exit_reason(log_tail) {
232        return Some(reason);
233    }
234
235    if oom_killed == Some(true) {
236        return Some("memory-exhaustion (cgroup-oom-killer)".to_string());
237    }
238
239    signal_name_for_exit_code(exit_code).map(|name| format!("signal ({})", name))
240}
241
242#[cfg(test)]
243mod tests {
244    use super::*;
245
246    #[test]
247    fn detects_v8_heap_limit_from_the_incident_log() {
248        let tail =
249            "FATAL ERROR: Reached heap limit Allocation failed - JavaScript heap out of memory";
250        assert_eq!(
251            detect_exit_reason(Some(tail)),
252            Some("memory-exhaustion (v8-heap-limit)".to_string())
253        );
254    }
255
256    #[test]
257    fn detects_kernel_oom_killer() {
258        assert_eq!(
259            detect_exit_reason(Some("Out of memory: Killed process 1234 (bun)")),
260            Some("memory-exhaustion (kernel-oom-killer)".to_string())
261        );
262    }
263
264    #[test]
265    fn detects_other_runtimes_reporting_their_own_exhaustion() {
266        assert_eq!(
267            detect_exit_reason(Some("fatal error: runtime: out of memory")),
268            Some("memory-exhaustion (go-runtime)".to_string())
269        );
270        assert_eq!(
271            detect_exit_reason(Some("Array buffer allocation failed")),
272            Some("memory-exhaustion (allocation-failure)".to_string())
273        );
274    }
275
276    #[test]
277    fn returns_none_for_ordinary_output() {
278        assert_eq!(detect_exit_reason(Some("all tests passed")), None);
279        assert_eq!(detect_exit_reason(None), None);
280    }
281
282    #[test]
283    fn maps_signal_exit_codes() {
284        assert_eq!(signal_name_for_exit_code(Some(137)), Some("SIGKILL".into()));
285        assert_eq!(signal_name_for_exit_code(Some(139)), Some("SIGSEGV".into()));
286        assert_eq!(signal_name_for_exit_code(Some(0)), None);
287        assert_eq!(signal_name_for_exit_code(Some(1)), None);
288        assert_eq!(signal_name_for_exit_code(None), None);
289    }
290
291    #[test]
292    fn log_marker_wins_over_signal_and_cgroup_flag() {
293        let tail =
294            "FATAL ERROR: Reached heap limit Allocation failed - JavaScript heap out of memory";
295        assert_eq!(
296            resolve_exit_reason(Some(139), Some(tail), Some(false)),
297            Some("memory-exhaustion (v8-heap-limit)".to_string())
298        );
299    }
300
301    #[test]
302    fn memory_marker_returns_the_whole_matched_line() {
303        let tail = "building...\nFATAL ERROR: Reached heap limit Allocation failed - JavaScript heap out of memory\n----- Native stack trace -----\n";
304        let marker = detect_memory_marker(Some(tail)).expect("marker");
305        assert_eq!(marker.reason, "memory-exhaustion (v8-heap-limit)");
306        assert_eq!(
307            marker.line,
308            "FATAL ERROR: Reached heap limit Allocation failed - JavaScript heap out of memory"
309        );
310    }
311
312    #[test]
313    fn memory_marker_line_is_length_bounded() {
314        let tail = format!(
315            "{}JavaScript heap out of memory{}",
316            "x".repeat(400),
317            "y".repeat(400)
318        );
319        let marker = detect_memory_marker(Some(&tail)).expect("marker");
320        assert!(marker.line.chars().count() <= MAX_MARKER_LINE_LENGTH + 3);
321        assert!(marker.line.ends_with("..."));
322    }
323
324    #[test]
325    fn memory_exhaustion_explains_only_abnormal_exits() {
326        let tail =
327            "FATAL ERROR: Reached heap limit Allocation failed - JavaScript heap out of memory";
328        let observed =
329            resolve_memory_exhaustion(Some(139), Some(tail), Some(false)).expect("observed");
330        assert!(observed.memory_exhausted);
331        assert_eq!(observed.memory_exhausted_reason, tail);
332        // The same marker with a successful exit is just output, not a failure.
333        assert_eq!(
334            resolve_memory_exhaustion(Some(0), Some(tail), Some(false)),
335            None
336        );
337        assert_eq!(resolve_memory_exhaustion(None, Some(tail), None), None);
338    }
339
340    #[test]
341    fn memory_exhaustion_falls_back_to_the_container_flag() {
342        let observed =
343            resolve_memory_exhaustion(Some(137), Some("no marker"), Some(true)).expect("observed");
344        assert_eq!(
345            observed.memory_exhausted_reason,
346            "Docker reported State.OOMKilled=true"
347        );
348        assert_eq!(
349            resolve_memory_exhaustion(Some(1), Some("no marker"), Some(false)),
350            None
351        );
352    }
353
354    #[test]
355    fn falls_back_to_cgroup_observation_then_signal() {
356        assert_eq!(
357            resolve_exit_reason(Some(137), None, Some(true)),
358            Some("memory-exhaustion (cgroup-oom-killer)".to_string())
359        );
360        assert_eq!(
361            resolve_exit_reason(Some(139), Some("no marker here"), Some(false)),
362            Some("signal (SIGSEGV)".to_string())
363        );
364        assert_eq!(resolve_exit_reason(Some(0), Some("fine"), None), None);
365    }
366}
start_command/exit_reason.rs

start_command/
exit_reason.rs
