diff options
Diffstat (limited to 'doc/bugs/transfer_lock_file_removal_bug.mdwn')
-rw-r--r-- | doc/bugs/transfer_lock_file_removal_bug.mdwn | 26 |
1 files changed, 26 insertions, 0 deletions
diff --git a/doc/bugs/transfer_lock_file_removal_bug.mdwn b/doc/bugs/transfer_lock_file_removal_bug.mdwn index ca698b231..021ce08a6 100644 --- a/doc/bugs/transfer_lock_file_removal_bug.mdwn +++ b/doc/bugs/transfer_lock_file_removal_bug.mdwn @@ -33,3 +33,29 @@ Oh BTW, this race cannot happen on windows, since on windows, an open lock file cannot be deleted by another process. > [[fixed|done]] + +Let's also consider if this change will allow for safely expiring stale +lock files.. + +Situation before with expiry would have been buggy (which is why it never +tried to expire them): + +1. first process creates lock file, takes lock, is interrupted (no more + lock) +2. second process wants to delete stale lock file; checks if it's locked; + isn't +3. third process opens existing lock file, prepares to take lock +4. second process deletes lock file +5. third process takes lock +6. third process is now doing a transfer w/o a (visible) lock + +But, after this bug fix, the third process checks if it's lock file +exists after taking the lock. Since the second process deleted it, +the third process fails with an error "transfer in progress" +which is perhaps not accurate, but at least it stopped. + +For this to work though, the second process needs to actually take +the lock, in non-blocking mode. If it succeeds, it can keep the lock +held while deleting the lock file. This ensures that when the third process +takes the lock, the lock file will already be deleted by the time it checks +if it's there. |