On Wed, Mar 22, 2023 at 02:39:15PM +0100, Miroslav Zagorac wrote: > On 22. 03. 2023. 14:33, Willy Tarreau wrote: > >> Also, in that case, when calling 'haproxy -vv', it is no longer printed > >> that > >> malloc_trim is enabled. > > > > I'm not sure what you mean here, because for me it's never printed that > > malloc_trim() is enabled when building with jemalloc. > > > > Hello Willy, > > without the patch: > > % make -j8 ADDLIB="-ljemalloc" TARGET="linux-glibc" IGNOREGIT=1 > USE_GETADDRINFO=1 USE_LIBCRYPT=1 USE_LUA=1 USE_MEMORY_PROFILING=1 > USE_MODULES=1 USE_NS=1 USE_OPENSSL=1 USE_PCRE=1 USE_PCRE_JIT=1 USE_TFO=1 > USE_THREAD=1 USE_ZLIB=1 > > % ldd haproxy | grep jemalloc > libjemalloc.so.2 => /lib/x86_64-linux-gnu/libjemalloc.so.2 > (0x00007f1319e3a000) > > ./haproxy -vv | grep trim > Support for malloc_trim() is enabled.
Ah thanks now I understand. For me it wasn't the case because I LD_PRELOADed it. However by doing so you've changed the semantics of this variable "using_default_allocator" whose role was exactly to detect if the allocator had changed from the built-in one. That's why it wasn't called "using_glibc_allocator" or so. Admittedly that part was still seriously lacking comments... I'd rather turn it back to what it was and fix the problem where the message is emitted instead. Anyway on latest changes we do emulate malloc_trim() using the equivalent in the other libs so in the end I'll just remove that message as it doesn't provide any value anymore. Thanks for the explanation! Willy
