Brett @Google writes:
Ok, please proceed with testing RE24.
I think it will bring forth more bugs yet. Re-reading ITS#5340, it lists a number of times REP_ENTRY_MUSTRELEASE and other flags are not properly reset. Combined with overlay, callback and error handling code I have a hard time seeing just what's supposed to happen though.
Seems to work ok for me first time through, but will let it run a few more times.
Would the issue you describe require memory profiling to detect ?
The discussion (message #8 in the ITS) refers to "some" code which does this or that incorrectly, so I'm not sure offhand how to provoke a problem. Possibly both memory leaks and double frees of entries. Or locks that get held too long, so you'll have problems if you pull the plug on the network during a memory operation.
I suppose we could make a "try everything at once"-test: Stacks a bunch of overlays on top of each other, including chaining/back-ldap to the same server, then run some operations which will succeed and some which will fail at various levels in the overlay stack. Not me yet, though.