|
When Temboz is run on SQLite 3.x/PySQLite 2.x, sometimes when you do an action that updates the database like clicking on "Thumbs down" while the hourly fetch is done, Temboz will deadlock, and has to be killed.
|
2006-Jan-14 11:41:42 by majid:
It seems like the update thread (thread 2) loses the lock while it's doing its updates, but it is not clear to whom.
2006-Feb-13 00:36:11 by majid:
These articles may be relevant:
- ¤http://www.third-bit.com/trac/argon/wiki/Locking
- ¤http://initd.org/tracker/pysqlite/ticket/126
- ¤http://permalink.gmane.org/gmane.comp.db.sqlite.general/9489
- ¤http://thread.gmane.org/gmane.comp.db.sqlite.general/13001
2006-Feb-26 11:31:54 by majid:
The deadlocks are easily reproduced by running the ApacheBench ab command to load the /view page (which also updates the article viewed timestamp, so it is not a read-only transaction). With a concurrency of 1, it works, with a concurrency of 2 or higher, it immediately deadlocks.
|
Type: bug Version: Status: fixed Created: 2006-Jan-14 11:40 Severity: 1 Last Change: 2006-Feb-26 11:32 Priority: 1 Subsystem: Assigned To: majid Derived From: Creator: majid
| 2006-Feb-26 11:32 | • | Check-in [290]: fix #31 by using a lock to serialize all write transactions, as SQLite3 seems to have deadlocks when writers contend. (By majid) |