On Jun 29, 2009, at 10:58 AM, Peter Secor wrote: > It looks ok now, this happens every once in awhile when the db gets > locked. I think Zooko checked into it once and may have more details.
Yes, I foolishly upgraded the trac software and the TracDarcs plugin without first making a backup copy of the sqlite db. The sqlite db got corrupted. (Note: sqlite is not supposed to get corrupted no matter what the Python code that is sending sql commands to it does, nor even if sqlite gets killed in the middle of a process of writing updates, or the underlying operating system crashes or if. So this indicates that there was a bug in the version of sqlite on our server. I didn't investigate.) It took 14 hours to repopulate the sqlite db with all of the darcs history. I assume that the slowness of this process was mostly darcs's fault since darcs is so slow, but I didn't investigate. The entire db remained locked through the repopulating process, so the whole trac was unavailable for most of the day. Okay, so please let me know if you see anything else funny about the new version of trac (0.11.4 instead of the 0.11.2 that we were using until now) or the new TracDarcs. Also, I'm going to make backup copies of the sqlite db before changing anything, from now on. Sorry about that! Regards, Zooko _______________________________________________ tahoe-dev mailing list [email protected] http://allmydata.org/cgi-bin/mailman/listinfo/tahoe-dev
