Memory fragmentation rate is abnormal

[heepy]

 
We have a problem that the memory fragmentation rate of redis instance is lower than 1, but the system resources are enough and there is no swap。

 
System：Red Hat Enterprise 7.4
Redis Server：Redis 3.2.9
model：Redis Cluster
 
info memory
#Memory
used_memeory_human：11.57M
used_memeory_rss_human：7.56M
used_memeory_peak_human：24.39M
used_memeory_lua_human：37.00k
maxmemory_human:1.00G
mem_fragmentation_ratio:0.66
mem_allocator:jemalloc-4.0.3

#free -g

	total	used	free	shared	buff/cache	available
mem	251	2	246	1	3	246
swap	3	0	3

[oranagra]

this can easily happen when allocating a chunk of memory and not actually writing to it. in that case the OS doesn't actually maps any pages for that allocation.
most of the actual allocations in redis (data types) don't do that, but in this case i see redis is nearly empty, so that might be some startup memory allocations.

bottom line, not a lot of memory, not a problem, and not unexpected.
i.e. if you would have shown me that RSS 2GB smaller than used_memory that would make me curious. (although still not really a problem)

btw, what version of redis is it?
if you really want to look into fragmentation, i'd advise to look at:

allocator_frag_ratio
allocator_frag_bytes
allocator_rss_ratio
allocator_rss_bytes
rss_overhead_ratio
rss_overhead_bytes

[heepy]

this can easily happen when allocating a chunk of memory and not actually writing to it. in that case the OS doesn't actually maps any pages for that allocation.
most of the actual allocations in redis (data types) don't do that, but in this case i see redis is nearly empty, so that might be some startup memory allocations.

bottom line, not a lot of memory, not a problem, and not unexpected.
i.e. if you would have shown me that RSS 2GB smaller than used_memory that would make me curious. (although still not really a problem)

btw, what version of redis is it?
if you really want to look into fragmentation, i'd advise to look at:

allocator_frag_ratio
allocator_frag_bytes
allocator_rss_ratio
allocator_rss_bytes
rss_overhead_ratio
rss_overhead_bytes

redis server version 3.2.9

[oranagra]

so you don't have these fields in that version.
if you feel adventurous, you can get that info manually by running DEBUG jemalloc info