The Files Deleted Years Ago Came Back — JavaScript Bug Hunt

Briefing

Modelled on Dropbox in January 2017, when users reported that files and folders they had deleted — some of them years earlier — had reappeared in their accounts. Dropbox said a bug had caused a small number of users' deleted files to be restored and that it was contacting the people affected.

This reconstruction models one plausible way such a bug happens: a compaction job trims each path's event log before the log is replayed into the live file tree. It throws away old delete tombstones but keeps the old uploads they cancelled, so the next replay brings the files back. (Dropbox's own internal root cause was not published in this form; the project is a reconstruction of the failure, not of their code.)

Fix compact so compaction never changes which files are live.

Bug report

BUG-RESURRECT · Priority: Critical · Reported by: support (users)

compact(events, now, retentionMs) — events are { path, op: "put"|"delete", at } in the order they happened. It returns the events to keep, in their original order:

• keep every event with now - at <= retentionMs
• additionally keep, for each path whose LATEST event is a put, that latest put (however old — a live file is never compacted away)
• drop everything else

Invariant: replay.livePaths(compact(events, ...)) === replay.livePaths(events).

Observed: a folder deleted two years ago is live again after the nightly compaction.

Logs

[compact] 3 tombstones older than 30d dropped
[replay] /photos/2015 restored (latest event: put @ day 10)

The code as shipped

src/sync/compact.js (editable)

// Compacts the event log before it is replayed into the live file tree.
// events: [{ path, op: "put" | "delete", at }] in the order they happened.
exports.compact = function (events, now, retentionMs) {
  return events.filter(function (e) {
    if (e.op === "put") return true;
    return now - e.at <= retentionMs;
  });
};

Read-only context: src/sync/replay.js.