On 20/01/15 14:18, Rob Hall wrote:
The fix isn't in the mongo code, but rather the is os itself requires the
hotfix. Check the mongodb windows installation instructions for ref.
Got it, thanks. It's about flushing files.
That's a hotfix to Windows7, not java. Maybe useful to apply but it
does not affect the fact that the JVM on windows simply does not have
the functionality to delete mmap files mid-process.
FWIW MMap files on Windows don't seem to be faster than TDB in direct
(old fashioned, in-JVM caching) or so I'm told. Nowadays it possible to
tune the internal TDB caches. Moving to Guava concurrent caches will
help (JENA-801, which principally about the node table but if in direct
mode on Windows, the same applies).
Andy
On Jan 20, 2015 8:45 AM, "Andy Seaborne" <[email protected]> wrote:
On 20/01/15 12:15, Rob Hall wrote:
I did randomly notice that mongo db had to have a hot fix for windows 7+
users that solved a bug associated with Windows memory mapped files.
If this is a Java+Windows environment problem, that might be a clue as to
what's going on.
The build problem looks like resources, not java.
The reason why TDB has Windows specific testing is the (in)famous:
http://bugs.java.com/view_bug.do?bug_id=4715154
(and several others)
Long time, won't fix, bug. (There are pragmatic, limited, JVM-specific
semi-workarounds.)
If MongoDB have had a fix recently, I'd be interested to know more - my
Google foo wasn't strong enough to find it.
TDB's test setup goes through hoops to workaround it and it leads to large
amounts of temporary file space that does not go away until end of JVM.
Plus users do sometimes want to delete a database awhile running.
Andy
Relevant hot fix: http://support.microsoft.com/kb/2731284
On Jan 20, 2015 6:04 AM, "Andy Seaborne" <[email protected]> wrote:
This is repeatedly going wrong and I'm not sure why.
On the surface, it looks like there is a rogue Jetty sitting on port 3535
which then blocks the tests.
But:
1/ Tracing back to the last job to run at all (#507), it's some TDB/JDBC
tests failing for weird low level reasons.
2/ Switch between windows1 and windows2 slaves today gets nasty file
errors (various flavours - some clearly jenkins problems)
3/ I can't even wipe the workspace at the moment.
so my guess is that OS resources are the problem and we can't fix them.
I've email builds@ to ask for help and disabled the job for now.
We might be contributing to the problems; the windows builds being
resource hungry for mmap files, but the job was running fine for quite
some
time until #507 and I can't see any change that might be connected.
Andy
