https://bugs.openldap.org/show_bug.cgi?id=9538
--- Comment #1 from Ondřej Kuzník <ondra(a)mistotebe.net> ---
This is an example of what can happen, accesslog is set up to log all ops:
1. op A, a write, comes in and locks accesslog
2. op B, like a Bind, comes in, it's not a write can proceed even though
accesslog is locked
3. op B finishes, we decide to log it, create the accesslog entry and
call be_add() (there's still no CSN assigned for it)
4. op A is now scheduled gets a CSN assigned (csnA) and gets committed
into main DB
5. be_add() adding the log entry for op B gets a CSN assigned (csnB) and
gets stored in the log DB
6. accesslog_response() is called for op A and unlocks accesslog, logs
the write, which is now out of CSN order (csnA is before csnB already
committed)
Right now that is what looks the most challenging sequence of events to
sort.
--
You are receiving this mail because:
You are on the CC list for the issue.