On 02/05/13 18:17, Marc von Holzen wrote:
Hi Andy,
Thank you so much for your quick response. I successfully recovered the
data. I will definitely upgrade to the latest.
Regarding understanding the circumstances, my application interacts
entirely via Fuseki services. Nothing more than a series of queries and
updates. The logs don't show much. The only slightly unusual thing
might be a few of "INSERT DATA {<> <> <>}" but they seem to have worked
fine in the past.
Thanks for your work,
Marc
One of the problems with the 0.2.5 codebase was that any problem in a
section of the commit process caused the same symptom.
I looked at the svn history and found a possible reason a server with
multiple overlapping operations might break and cause the problem your
seeing. The fix did not make the last release - it is in the current
development snapshot builds.
I doubt "INSERT DATA {<> <> <>}" is related - more likely is two
transactions at the same time in a race-condition trying to flush the
journal. These can be read transactions - a read transaction can cause
the journal to be written back if a queue of pending commits has built
up and the read transaction is the last block on flushing the
write-ahead log. There was a bug (a small timing hole) here.
Glad you managed to recover the data,
Andy
