Alexandre Fayolle <alexandre.fayolle@logilab.fr> [Wed, 02 Jun 2010 16:30:36 +0200] rev 5641
backported to stable some changes made on site for a customer
Alexandre Fayolle <alexandre.fayolle@logilab.fr> [Wed, 02 Jun 2010 16:25:12 +0000] rev 5640
logging settings
* document helpful log format when dealing with MT issues
* on Win32, use a logrotate equivalent so that logs get a bit lighter (requires updates logilab.common)
Alexandre Fayolle <alexandre.fayolle@logilab.fr> [Wed, 02 Jun 2010 16:12:18 +0000] rev 5639
[win32] fix deadlock occuring on the sequence tables with SQLServer
actually, this deadlock would occur with any db backend other that PostgreSQL
as the previous code was heavily relying on PG's SEQUENCE facility, not available
elsewhere.
Deadlock description:
Thread1 starts creating entities (and therefore calls create_eid):
-> this creates a DB-level lock on the entities_id_seq table, which
will last until end of transaction
Thread2 calls create_eid, which acquires the Python lock object, but
updating the entities_id_seq is held by the DB lock
Thread1 wants to create a new entity, calls create_eid, and is stuck by
the Python lock object held by Thread2.
Solution:
use a separate connection to read and write the entities_id_seq table.