Broflake Go widget consumes a lot of CPU and Memory

[oxtoacart]

While downloading a very large file and watching a YouTube video, CPU usage on my Broflake widget (free peer) was much higher than I'd expect.

Cumulative PPROF shows that we're spending a lot of time locking.

Showing top 20 nodes out of 138
      flat  flat%   sum%        cum   cum%
     0.01s 0.024% 0.024%     15.67s 37.91%  runtime.systemstack
     0.02s 0.048% 0.073%     15.52s 37.55%  runtime.wakep
         0     0% 0.073%     15.50s 37.50%  runtime.startm
    15.48s 37.45% 37.53%     15.48s 37.45%  runtime.pthread_cond_signal
         0     0% 37.53%     15.48s 37.45%  runtime.semawakeup
         0     0% 37.53%     15.46s 37.41%  runtime.notewakeup
     0.01s 0.024% 37.55%     14.65s 35.45%  runtime.goready.func1
         0     0% 37.55%     14.64s 35.42%  runtime.ready
         0     0% 37.55%      6.14s 14.86%  runtime.mcall
         0     0% 37.55%      6.08s 14.71%  runtime.schedule
         0     0% 37.55%      5.82s 14.08%  runtime.park_m
     0.01s 0.024% 37.58%      5.60s 13.55%  runtime.findRunnable
         0     0% 37.58%      5.37s 12.99%  runtime._System
         0     0% 37.58%      4.78s 11.57%  github.com/pion/sctp.(*Association).readLoop
         0     0% 37.58%      4.76s 11.52%  github.com/pion/sctp.(*Association).handleInbound
         0     0% 37.58%      4.74s 11.47%  github.com/pion/sctp.(*Association).handleChunk
         0     0% 37.58%      4.74s 11.47%  github.com/pion/sctp.(*Association).handleSack
     0.02s 0.048% 37.62%      4.72s 11.42%  github.com/pion/sctp.(*Association).processSelectiveAck
         0     0% 37.62%      4.68s 11.32%  github.com/pion/sctp.(*payloadQueue).pop
     0.01s 0.024% 37.65%      4.68s 11.32%  github.com/pion/sctp.(*payloadQueue).updateSortedKeys

This is also confirmed by non-cumulative pprof:

Showing top 20 nodes out of 138
      flat  flat%   sum%        cum   cum%
    15.48s 37.45% 37.45%     15.48s 37.45%  runtime.pthread_cond_signal
     3.06s  7.40% 44.86%      3.06s  7.40%  runtime.pthread_cond_wait

Possibly this is due to lots of channel sends and receives.

[oxtoacart]

Apparently, it's getting OOM Killed too

May  9 15:05:38 broflake-uncensored-peer-1 kernel: [508700.096948] sshd invoked oom-killer: gfp_mask=0x140cca(GFP_HIGHUSER_MOVABLE|__GFP_COMP), order=0, oom_score_adj=0
May  9 15:05:38 broflake-uncensored-peer-1 kernel: [508700.097079]  oom_kill_process.cold+0xb/0x10
May  9 15:05:38 broflake-uncensored-peer-1 kernel: [508700.097103]  __alloc_pages_may_oom+0x112/0x1e0
May  9 15:05:38 broflake-uncensored-peer-1 kernel: [508700.097347] [  pid  ]   uid  tgid total_vm      rss pgtables_bytes swapents oom_score_adj name
May  9 15:05:38 broflake-uncensored-peer-1 kernel: [508700.097449] oom-kill:constraint=CONSTRAINT_NONE,nodemask=(null),cpuset=ssh.service,mems_allowed=0,global_oom,task_memcg=/user.slice/user-0.slice/session-152.scope,task=widget,pid=69952,uid=0
May  9 15:05:38 broflake-uncensored-peer-1 kernel: [508700.097479] Out of memory: Killed process 69952 (widget) total-vm:1880484044kB, anon-rss:282084kB, file-rss:0kB, shmem-rss:0kB, UID:0 pgtables:920kB oom_score_adj:0