[Date Prev][Date Next] [Chronological] [Thread] [Top]

handling cache "not ready" and cache insertion races

I've committed code to both the LDBM and BDB2 backends that should
wait for cached entries to become ready (instead of returning
'no such object').  I also added code to handle concurrent
attempts to add the same fetched entry into the cache.  The
threads still race to insert, but the loser(s) accepts the
winner's insertion instead of failing.

The changes makes the concurrency test behave much better on
the few boxes I've tested it on.  You are encourged to test
these changes!