On Mon, May 22, 2017 at 10:48:31PM +0200, niftybunny wrote:
Same with 0.3.0.5. Upgrading to 0.3.0.7 helped on most relays.
We didn't change anything between 0.3.0.5 and 0.3.0.7 that would have helped. If somebody with a really really fast CPU wants to run their relay under valgrind --leak-check for a while, that would be grand. https://gitweb.torproject.org/tor.git/tree/doc/HACKING/HelpfulTools.md The question to try to get some answers on is what sort of bloat problem we have: A) Maybe it's a memory leak? B) Maybe it's an inefficiency of the memory allocator, e.g. fragmentation where nothing is technically leaked yet there's a whole lot of wasted space? C) Maybe it is some behavior by clients that causes the relay to use a lot of memory, e.g. by having a bunch of stuff in buffers? The behavior could be accidental / normal, or it could be intentional / malicious. My first guess is "C, accidental". But it could easily be 'B', and it would be great if it's 'A' because then we can just fix it. --Roger