fix(cli): have diagnose report failing writes, a pause and folders in error

diagnose dropped most of the Status reply and never looked at the synced
folders, so it said "No problems found" while a folder would not sync
(B268, issue #53). It now warns about writes that keep failing, a write
a day old, a pause, folders in error or conflict, and a control socket
nothing listens on.
This commit is contained in:
nvrl committed 2026-10-31 10:57:34 +01:00
1 parent 9460c93b50
commit a617b2b410
2 files changed
+166 -9

No files matched your search

+154 -5
View File
@@ -3369,6 +3369,79 @@ fn cmd_sync_limit(up: Option<u64>, down: Option<u64>) -> Result<()> {
})?)
}
/// How long the oldest write may wait to upload before `diagnose` calls it
/// out. Under a day it is a slow link or a big file; past it, something is
/// holding the queue.
const STAGED_STALE_SECS: u64 = 86_400;
/// `diagnose` findings for the queue and the pause, from the `Status` reply.
/// `paused_for` is how long a timed pause still runs, in seconds.
fn sync_health(
failing_ops: u64,
failing_error: Option<&str>,
staged_bytes: u64,
staged_oldest_secs: u64,
paused: bool,
paused_for: Option<i64>,
) -> Vec<(DiagnoseLevel, &'static str, String)> {
let mut findings = Vec::new();
if failing_ops > 0 {
findings.push((
DiagnoseLevel::Warn,
" failing writes",
match failing_error {
Some(error) => format!("{failing_ops} keep failing; the last error: {error}"),
None => format!("{failing_ops} keep failing"),
},
));
}
if staged_bytes > 0 && staged_oldest_secs >= STAGED_STALE_SECS {
findings.push((
DiagnoseLevel::Warn,
" oldest unsent write",
format!(
"{} old, {} only on this computer",
human_secs(staged_oldest_secs),
human_bytes(staged_bytes)
),
));
}
if paused {
findings.push((
DiagnoseLevel::Warn,
" sync",
match paused_for {
Some(left) if left > 0 => {
format!("paused for another {}", human_secs(left as u64))
}
_ => "paused until resumed (`pdfs sync resume`)".to_string(),
},
));
}
findings
}
/// `diagnose` findings for the synced folders that need someone to look.
fn folder_health(
folders: &[pdfs_core::control::SyncFolderInfo],
) -> Vec<(DiagnoseLevel, String, String)> {
folders
.iter()
.filter_map(|folder| {
let what = match folder.state.as_str() {
"error" => "sync error; `pdfs sync list` and the activity feed say more",
"conflict" => "conflict copies to look at",
_ => return None,
};
Some((
DiagnoseLevel::Warn,
format!(" synced folder [{}]", folder.id),
format!("{}: {what}", folder.local_path),
))
})
.collect()
}
fn unix_now() -> i64 {
std::time::SystemTime::now()
.duration_since(std::time::UNIX_EPOCH)
@@ -3705,21 +3778,26 @@ fn cmd_diagnose() -> Result<()> {
let socket_exists = socket.exists();
let socket_kind_ok =
std::fs::symlink_metadata(&socket).is_ok_and(|m| m.file_type().is_socket());
// A stale socket file with no listener is the common failure, so the reply
// is what decides this rather than the file's existence.
let status = control_request(CtlRequest::Status);
report.finding(
if socket_exists && !socket_kind_ok {
DiagnoseLevel::Fail
} else if socket_exists {
} else if socket_exists && status.is_ok() {
DiagnoseLevel::Ok
} else {
DiagnoseLevel::Warn
},
" control socket",
socket.display().to_string(),
if socket_exists && socket_kind_ok && status.is_err() {
format!("{} (left over; nothing listens on it)", socket.display())
} else {
socket.display().to_string()
},
);
// A stale socket file with no listener is the common failure, so the reply
// is what decides this rather than the file's existence.
let mut daemon_responding = false;
match control_request(CtlRequest::Status) {
match status {
Ok(CtlResponse::Status {
mountpoint,
online,
@@ -3727,6 +3805,12 @@ fn cmd_diagnose() -> Result<()> {
tokens_unsaved,
pending_uploads,
pending_changes,
failing_ops,
failing_error,
staged_bytes,
staged_oldest_secs,
paused,
paused_until,
version,
..
}) => {
@@ -3801,6 +3885,27 @@ fn cmd_diagnose() -> Result<()> {
format!("{pending} waiting to upload")
},
);
// The rest of the daemon's state that the report used to drop:
// writes that keep failing, a pause, a folder in error. Each of
// them stops changes from reaching Drive while everything above
// reads as fine (`docs/BUGS.md` B268).
for (level, check, detail) in sync_health(
failing_ops,
failing_error.as_deref(),
staged_bytes,
staged_oldest_secs,
paused,
paused_until.map(|until| until - unix_now()),
) {
report.finding(level, check, detail);
}
if let Ok(CtlResponse::SyncFolders { items }) =
control_request(CtlRequest::ListSyncFolders)
{
for (level, check, detail) in folder_health(&items) {
report.finding(level, &check, detail);
}
}
match await_integrity_check() {
Ok(CtlResponse::CacheReport {
integrity_problems,
@@ -3982,6 +4087,50 @@ mod pause_tests {
mod diagnose_tests {
use super::*;
/// A healthy queue says nothing; writes that keep failing, an old unsent
/// write and a pause each say so (`docs/BUGS.md` B268).
#[test]
fn a_stuck_queue_or_a_pause_is_reported() {
assert!(sync_health(0, None, 0, 0, false, None).is_empty());
assert!(sync_health(0, None, 4096, STAGED_STALE_SECS - 1, false, None).is_empty());
let checks: Vec<&str> = sync_health(
2,
Some("upload a.txt: quota"),
4096,
STAGED_STALE_SECS,
true,
Some(600),
)
.into_iter()
.map(|(level, check, _)| {
assert!(matches!(level, DiagnoseLevel::Warn));
check.trim()
})
.collect();
assert_eq!(checks, ["failing writes", "oldest unsent write", "sync"]);
}
/// Issue #53: a folder that would not sync, and `diagnose` found nothing.
#[test]
fn a_synced_folder_in_error_is_reported() {
let folder = |id: i64, state: &str| pdfs_core::control::SyncFolderInfo {
id,
local_path: format!("/home/u/f{id}"),
remote_uid: "vol~link".into(),
mode: "mirror".into(),
state: state.into(),
last_sync: 0,
pending_mode: None,
progress: None,
paused: false,
};
let findings =
folder_health(&[folder(1, "idle"), folder(2, "error"), folder(3, "syncing")]);
assert_eq!(findings.len(), 1);
assert_eq!(findings[0].1.trim(), "synced folder [2]");
assert!(findings[0].2.starts_with("/home/u/f2: sync error"));
}
#[test]
fn finding_severity_is_counted_centrally() {
let mut report = DiagnoseReport::default();