[Bug 3162] crash on retry database lookup

https://bugs.exim.org/show_bug.cgi?id=3162

Jeremy Harris <[email protected]> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Priority|high                        |medium

--- Comment #1 from Jeremy Harris <[email protected]> ---
Assumptions from discussions to date :
- No messages are being lost, only delayed - and only in cases where the
  transport target is unavailable anyway.  Downgrading the bug Importance
  on that basis.
- We do not have a repeat-by, which will hinder investigations
- Possibly the system is under memory-pressure load.  So far unclear if
  this is a factor.
- Syndrome was not observed under 4.98.2 - based systems; only 4.98-RC.
- Immediate fault location is in a cache layer filtering access to the
  retry hints DB; the DB implementation in use is probably not a factor.
- Coredump investigation shows (in one identified case; others differ)
  a tree node struct that is uninitialized or has been deliberate-wiped :-

   $2 = {left = 0xdbdbdbdbdbdbdbdb, right = 0xdbdbdbdbdbdbdbdb, data = {ptr = 
   0xdbdbdbdbdbdbdbdb, val = -606348325}, balance = 219 '\333',
   name = <incomplete sequence \333>}

Suggestion for investigation: set main-config option "debug_store = true"
and watch mainlog for "destroyed by reset_store".

Working notes:
- node is created by remote_post_process() call to tree_add_unusable()

-- 
You are receiving this mail because:
You are on the CC list for the bug.