In alert log there are a lot of "Transaction recovery: lock conflict caught and ignored" statements.
I've found some information, that this error is caused when there is an active dead transaction. For example, long running transaction was forcefully stopped and it did not managed to recover(caused by no free space in UNDO tablespace, or aborting database...)
When I query the following:
select b.name useg, b.inst# instid, b.status$ status, a.ktuxeusn xid_usn, a.ktuxeslt xid_slot, a.ktuxesqn xid_seq, a.ktuxesiz undoblocks, a.ktuxesta txstatus from x$ktuxe a, undo$ b where a.ktuxecfl like '%DEAD%' and a.ktuxeusn = b.us#;
It returns 23 rows , so there are active dead transactions.
I tried to DUMP UNDO HEADER, dig in the trace files and find object ids that needed recovery. I found these objects and re-created them(was suggested in some blog) but still did not help.
Note: When there is a supplemental logging enabled , Oracle creates incident files and core dump files and writes mentioned error in alert log, which causes server space to be exhausted. If you turn the supplemental logging off, just alert log contains this error message and NO incident/core dump files are created, better variant, but still disturbing when entries are a lot.