On 11/12/05, robert burrell donkin <[EMAIL PROTECTED]> wrote: > On Sat, 2005-11-12 at 13:44 +0000, Stephen Colebourne wrote: > > robert burrell donkin wrote: > > > has anyone run a long stress test? > > > > > > if not, i'm willing to code up something and set it running on my debian > > > box for a few days. i'd appreciate a second pair of eyes on the code (to > > > avoid mistakes). > > > > My attempt at one is attached to bugzilla - SoakLRUMap. I've only run it > > for an hour or so though.
I ran something similar for two 3-hour runs with no errors, using a mix of Integer and String keys engineered to get a lot of reuse to happen. I did lots of shorter runs modifying the the number of buckets, number of threads, and relative frequency of adds / replaces, etc. It was easy to get the reported errors with synchronization off; but I saw no errors when access was properly synchronized. > > great. it's a simple test but then again, none of the problems were > reported with obscure keys. i've set off two processes on my debian box > one running the single thread, one the synchronised multiple thread > variant. > > > I did wonder if this might be some kind of double checked locking effect > > only seen on multi-processor systems. > > otis reports that he was using a beta JVM. > > the reports (or at least those who are using correctly synchronized > versions of the map) coincide with the time when 1.5 JDKs were in beta. > wonder whether it was some sort of bug in the synchronization code that > was later fixed by sun. > > my debian box is single processor and i don't have access to a double > processor machine. it's on constantly so i can leave the tests running > for a few days. i'll report back if there are any NPEs. > > i suggest that if there are no NPEs in the few days, we work on the > assumption that it's some kind of issue in the synchronization code > (rather than the collection code). so, we then document and add a > warning to the release notes before moving towards shipping the > release. +1 > > i would recommend keeping the bugzilla open but adding a comment > requesting that users ensure that the code is synchronized properly and > that they list the full environment (JVM and platform) plus information > about the keys they are using (and ideally a soak tests we can run). > that may give us enough information to track down where the problem > lies. +1, plus include additional debugging code in release. > > - robert > > > --------------------------------------------------------------------- > To unsubscribe, e-mail: [EMAIL PROTECTED] > For additional commands, e-mail: [EMAIL PROTECTED] > > --------------------------------------------------------------------- To unsubscribe, e-mail: [EMAIL PROTECTED] For additional commands, e-mail: [EMAIL PROTECTED]