Bug #3887
closedkernel client: small object memory leak
0%
Description
In testing my new request code for rbd (issue 3741 and related)
I tried paying special attention to Linux slab usage in order
to avoid introducing memory leaks.
I ran one test for 4000+ iterations using user-mode-Linux (UML).
I didn't look at all slabs but the generic ones were all reasonable
except "size-32", which had grown by on the order of 50000 objects.
Looking at a sample of the allocated entries in the slab it looked
as though the allocations were something consisting of two pointers.
All the entries seemed to have the same two values, and the first was
an odd pointer so I expect it was a string. (It's possible I
misinterpreted this, but hopefully not. My UML is gone now so I
can't double-check.)
I don't believe any of the rbd data structures I'm now using would
fit this description, so it may be that the leak is somewhere else,
like the osd client or the messenger.
For now I'm just making a note of this. It's not a furiously fast
leak and they're small objects. But it ought to get tracked down.