I have a container leaking memory. Or at least, reported memory consumption goes up fast. If I run top, I get this:
top - 16:56:51 up 6 days, 17:25, 0 users, load average: 0.16, 0.27, 0.31
Tasks: 4 total, 1 running, 3 sleeping, 0 stopped, 0 zombie
%Cpu(s): 1.3 us, 0.7 sy, 0.0 ni, 98.0 id, 0.0 wa, 0.0 hi, 0.0 si, 0.0 st
KiB Mem: 7676380 total, 4089380 used, 3587000 free, 675164 buffers
KiB Swap: 0 total, 0 used, 0 free. 2586496 cached Mem
PID USER PR NI VIRT RES SHR S %CPU %MEM TIME+ COMMAND
1 root 20 0 46924 15196 6456 S 0.0 0.2 0:15.54 supervisord
8 root 20 0 3526084 47976 29660 S 0.0 0.6 0:59.15 dotnet
568 root 20 0 20364 3332 2728 S 0.0 0.0 0:00.09 bash
2800 root 20 0 21956 2420 2032 R 0.0 0.0 0:00.00 top
The number I get reported out at this moment is 90M, which doesn't add up. I'm pretty sure /sys/fs/cgroup/memory/memory.usage_in_bytes is what's reported:
> cd /sys/fs/cgroup/memory/
> cat cgroup.procs
1
8
568
2494
> cat memory.usage_in_bytes
92282880
> pmap -p 1 | tail -n 1
total 46924K
> pmap -p 8 | tail -n 1
total 3599848K
> pmap -p 568 | tail -n 1
total 20364K
> ps 2494
PID TTY STAT TIME COMMAND
To my eyes there's a significant amount of memory 'missing' here... if I cat memory.usage_in_bytes again over the time I've been typing this:
> cat memory.usage_in_bytes
112291840
> pmap -p 1 | tail -n 1
total 46924K
> pmap -p 8 | tail -n 1
total 3452320K
> pmap -p 568 | tail -n 1
total 20368K
Nothing obviously accounts for this memory usage. If I look at memory.stat I see this:
# cat memory.stat
cache 89698304
rss 30699520
rss_huge 0
mapped_file 1552384
writeback 0
pgpgin 102007
pgpgout 72613
pgfault 115021
pgmajfault 8
inactive_anon 1519616
active_anon 30789632
inactive_file 417792
active_file 87654400
unevictable 4096
hierarchical_memory_limit 18446744073709551615
total_cache 89698304
total_rss 30699520
total_rss_huge 0
total_mapped_file 1552384
total_writeback 0
total_pgpgin 102007
total_pgpgout 72613
total_pgfault 115021
total_pgmajfault 8
total_inactive_anon 1519616
total_active_anon 30789632
total_inactive_file 417792
total_active_file 87654400
total_unevictable 4096
Then a short moment later:
# cat memory.stat
cache 89972736
rss 30777344
rss_huge 0
mapped_file 1552384
writeback 0
pgpgin 102316
pgpgout 72836
pgfault 115674
pgmajfault 8
inactive_anon 1519616
active_anon 30867456
inactive_file 417792
active_file 87928832
unevictable 4096
hierarchical_memory_limit 18446744073709551615
total_cache 89972736
total_rss 30777344
total_rss_huge 0
total_mapped_file 1552384
total_writeback 0
total_pgpgin 102316
total_pgpgout 72836
total_pgfault 115674
total_pgmajfault 8
total_inactive_anon 1519616
total_active_anon 30867456
total_inactive_file 417792
total_active_file 87928832
total_unevictable 4096
But I'll be honest; I don't really know what I'm looking at here. I am looking suspicously at active_file but again; I don't really know what I'm looking at.
Some notes and observations:
- The container is scheduled by Kubernetes.
- This program writes a lot of data to stdout.
- Reducing the amount of data written to console to near-zero solves the reported memory leak.
- Deploying a program which only writes lots of data to stdout does not seem exhibit the same memory leak(!)
So! How should I go about finding where this memory is getting consumed? Is there anything obvious to anybody - perhaps something's staring me in the face or I'm not looking at something I should be?
Thanks!
