Description of problem: memory-leak in crypt xlator (glusterfs client) Version-Release number of selected component (if applicable): 3.4 How reproducible: create/delete large amounts of files Steps to Reproduce: 1. 2. 3. Actual results: I created ~70K files and memory (RSS) started up from 435208 to 1663336 then rm and leave only 10K in the dir but RSS remained. Expected results: RSS should decrease after file deletion. Additional info: From the client statedumps I see that inode cache is released [encryption/crypt.oam_enc-crypt - usage-type gf_crypt_mt_inode memusage] size=9293896 num_allocs=11279 max_size=64594184 max_num_allocs=78391 total_allocs=78560 but local allocations seem still present... after deletion: [encryption/crypt.oam_enc-crypt - usage-type gf_crypt_mt_local memusage] size=547946008 num_allocs=678151 max_size=547946008 max_num_allocs=678151 total_allocs=678151 As i have seen in crypt xlator GF_CALLOC allocates memory from gf_crypt_mt_local pool. local = GF_CALLOC (sizeof (*local), 1, gf_crypt_mt_local); but in FRAME_DESTROY only mem_put(local) is called thus gf_common_mt_mem_pool is released which results of errors like 0-mem-pool: mem-pool ptr is NULL from glusterfs client.
with latest glusterfs builds, crypt xlator is no more loaded in graph, and is present in codebase.