[Gluster-devel] Questions on Statedump results

[Kaushal M]

On Sat, May 14, 2016 at 3:07 PM, Kremmyda, Olympia (Nokia - GR/Athens)
<olympia.kremmyda at nokia.com> wrote:
> Hi,
>
> We have observed in some statedumps, these entries during our tests:
> We use Gluster 3.6.9 and our tests include the creation of 4000 files, 1MB
> each, in one replicated volume, with two bricks.
> pool-name=fuse:dentry_t
> hot-count=78
> cold-count=32690
> padded_sizeof=84
> alloc-count=40121
> max-alloc=32768
> pool-misses=7347
> cur-stdalloc=0
> max-stdalloc=7347
> -----=-----
> pool-name=fuse:inode_t
> hot-count=87
> cold-count=32681
> padded_sizeof=156
> alloc-count=165684
> max-alloc=32768
> pool-misses=19253
> cur-stdalloc=1
> max-stdalloc=7350
> Are these values of pool-misses expected? Is there an indication of memory
> leak?

Gluster maintains pre-allocated memory pools for most of a lot of data
types of fixed capacity. Memory can be obtained from these pools using
the mem-pool api.
If enough memory concurrent allocations have been done to use up the
pool, memory is allocated using the normal allocation methods
(calloc).
This is what you observe as pool-misses. This is not an indication of
memory leak and is expected when under heavier load.

>
> Also in the data structure allocation stats:
> [mount/fuse.fuse - usage-type gf_fuse_mt_gids_t memusage]
> size=968
> num_allocs=1024
> max_size=8200
> max_num_allocs=1025
> total_allocs=106176
> If the size is “num_allocs times the sizeof(data‐type) i.e. num_allocs *
> sizeof (data‐type)”, how can it be in this case the num_allocs to be 1024
> and the size to be smaller than 1024?

The size here is the size of the data-type, not the current allocated
memory of this type.

>
> And one other question, can you explain to us what is the gf_common_mt_long
> pool, and why it is initialized only in production code (#ifndef DEBUG) in
> function mem_pool_new_fn?
>
> Thank you for your time,
> Olia
>
>
>
>
> _______________________________________________
> Gluster-devel mailing list
> Gluster-devel at gluster.org
> http://www.gluster.org/mailman/listinfo/gluster-devel