Ticket #3187 (closed defect: fixed)
Oracle FastCGI memory leak with 5.6.0-betas
|Reported by:||jmckenna||Owned by:||aboudreault|
|Component:||Input - Native Oracle Spatial Support||Version:||unspecified|
|Cc:||jimk, msmitherdc, aaime, dmorissette, pramsey, aboudreault, sdlime|
Initially discovered in the benchmarking exercise:
Mike and I spent some time looking into the leak tonight to no real compelling effect I'm afraid. Mike confirmed the problem is just the oracle case, and only occurs in 5.6, not 5.4. I ran mapserv for one oracle map request under valgrind and there was no obvious sign of big leakage beyond the one time connection setup costs. I guessed there might be problems with the connection pooling, and did some fastcgi debug logging, but after Mike pointed out an obvious point I concluded that the connection pooling did seem to be working ok and there is no particular reason to believe there is a connection level leak. The changes between 5.4 and 5.6 are huge.
From Steve Lime:
Jim found potential leaks in the template processing but that shouldn't come into play in this case and if a map is "free'd" then I believe those all go away. If not then you'd see problems with the other drivers as well. He's still looking...