facebook/flow · error

PidLog::log: failed to write

Error message

PidLog::log: failed to write

What it means

Panics when writeln! to the already-opened pids log fails. The pid log records every flow process PID and reason so orphaned daemons can be detected and killed later; the writer is a lazily-initialized global behind a mutex (pid_log::init opens the file). Note that no_fail=true only suppresses the 'uninitialized writer' panic, not a failed write.

Source

Thrown at rust_port/crates/flow_daemon/src/pid_log.rs:61

            .write(true)
            .open(pids_file)?;
        *guard = Some(oc);
        Ok(())
    })
}

pub fn log(reason: Option<&str>, no_fail: bool, pid: u32) {
    if !*enabled().lock().expect("pid_log enabled mutex poisoned") {
        return;
    }
    let pid = sys_utils::pid_of_handle(pid);
    let reason = reason.unwrap_or("unknown");
    let mut guard = log_oc().lock().expect("pid_log log_oc mutex poisoned");
    match guard.as_mut() {
        None if no_fail => {}
        None => panic!("Can't write pid to uninitialized pids log"),
        Some(oc) => {
            writeln!(oc, "{}\t{}", pid, reason).expect("PidLog::log: failed to write");
            oc.flush().expect("PidLog::log: failed to flush");
        }
    }
}

#[derive(Debug)]
pub struct FailedToGetPids;

impl std::fmt::Display for FailedToGetPids {
    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
        write!(f, "FailedToGetPids")
    }
}

impl std::error::Error for FailedToGetPids {}

pub fn get_pids(pids_file: &Path) -> Result<Vec<(u32, String)>, FailedToGetPids> {
    let ic = File::open(pids_file).map_err(|_| FailedToGetPids)?;

View on GitHub (pinned to f88ac94bcf)

Solutions

1. Free space on the volume holding the pids file and let the process log again
2. Point the pids file at a stable local path (pid_log::init argument) not subject to rotation or unmounting
3. If the write error is from a moved/truncated file, restart the process so init reopens a fresh handle

// Confirm the pids-log target still accepts writes before relying on it
let probe = pids_file.with_extension("probe");
if std::fs::write(&probe, b"x").is_err() {
    // volume full/unwritable: skip enabling the pid log rather than crash later
    pid_log::disable();
}
std::fs::remove_file(&probe).ok();

When it happens

Trigger: Calling pid_log::log after the disk holding the pids file is full; the log file was truncated/moved out from under the open handle (log rotation); the filesystem with the pids file was unmounted (NFS/sshfs home) while the process ran.

Common situations: Long-lived flow daemons on CI boxes with full disks; pids.log on a network mount that dropped; external log rotation tools moving the file.

Related errors

• PidLog::log: failed to flush
• Daemon::set_context: bincode serialize context
• failed to write {}: {}
• Daemon::flush failed
• failed to write flowlib file

AI-assisted analysis of facebook/flow@f88ac94bcf (2026-08-20). Data as JSON: /api/errors/5daf57df5dda5e12. Report an issue: GitHub.