Steadily increasing memory usage on version 1.2.5

[hsychla]

We see a slow but steady increase of memory usage on UPGs running version 1.2.5.

2022-04-11T22:00Z: 1.164 GB 2022-05-20T00:00Z: 1.479 GB

The number of sessions did not change significantly during that increase:

[hsychla]

Output of show memory map of one of the affected instances:

StartAddr          size   FD PageSz  Pages Numa0 Numa1 NotMap Name
00007fa7ad0d1000     1m          4K    256     2     0    254 main heap
00007fa6ed0d0000     3g          4K 786432 11430405812 369177 main heap
00007fa6e9862000     2m          4K    512     2     1    509 thread stack: thread 0
00007fa6c9861000   512m    4     4K 131072     4  1600      0 stat segment
00007fa6c9858000    32k          4K      8     0     2      6 process stack: wg-timer-manager
00007fa6c984f000    32k          4K      8     0     2      6 process stack: vrrp-periodic-process
00007fa6c9846000    32k          4K      8     0     2      6 process stack: pfcp-server-process
00007fa6c9835000    64k          4K     16     0     3     13 process stack: pfcp-api
00007fa6c982c000    32k          4K      8     0     2      6 process stack: perfmon-periodic-process
00007fa6c9823000    32k          4K      8     0     2      6 process stack: nsh-md2-ioam-export-process
00007fa6c981a000    32k          4K      8     0     2      6 process stack: nat-ha-process
00007fa6c9811000    32k          4K      8     0     2      6 process stack: memif-process
00007fa6c9808000    32k          4K      8     0     2      6 process stack: lldp-process
00007fa6c97ff000    32k          4K      8     0     2      6 process stack: udp-ping-process
00007fa6c97f6000    32k          4K      8     0     2      6 process stack: vxlan-gpe-ioam-export-process
00007fa6c97ed000    32k          4K      8     0     2      6 process stack: ioam-export-process
00007fa6c97e4000    32k          4K      8     0     2      6 process stack: ikev2-manager-process
00007fa6c97db000    32k          4K      8     0     2      6 process stack: igmp-timer-process
00007fa6c97d2000    32k          4K      8     0     1      7 process stack: static-http-server-process
00007fa6c97c9000    32k          4K      8     0     1      7 process stack: http-server-process
00007fa6c97c0000    32k          4K      8     0     2      6 process stack: gbp-scanner
00007fa6c97b7000    32k          4K      8     0     2      6 process stack: flowprobe-timer-process
00007fa6c97ae000    32k          4K      8     0     2      6 process stack: send-dhcp6-pd-client-message-process
00007fa6c97a5000    32k          4K      8     0     2      6 process stack: dhcp6-pd-client-cp-process
00007fa6c979c000    32k          4K      8     0     2      6 process stack: dhcp6-client-cp-process
00007fa6c9793000    32k          4K      8     0     2      6 process stack: send-dhcp6-client-message-process
00007fa6c978a000    32k          4K      8     0     2      6 process stack: dhcp6-pd-reply-publisher-process
00007fa6c9781000    32k          4K      8     0     2      6 process stack: dhcp6-reply-publisher-process
00007fa6c9770000    64k          4K     16     0     2     14 process stack: dhcp-client-process
00007fa6c9767000    32k          4K      8     0     2      6 process stack: cnat-scanner-process
00007fa6c975e000    32k          4K      8     0     2      6 process stack: avf-process
00007fa6c9755000    32k          4K      8     0     2      6 process stack: acl-plugin-fa-cleaner-process
00007fa6c974c000    32k          4K      8     0     2      6 process stack: statseg-collector-process
00007fa6c970b000   256k          4K     64     0     3     61 process stack: api-rx-from-ring
00007fa6c9702000    32k          4K      8     0     2      6 process stack: rd-cp-process
00007fa6c96f9000    32k          4K      8     0     2      6 process stack: ip6-ra-process
00007fa6c96f0000    32k          4K      8     0     2      6 process stack: ip6-rs-process
00007fa6c96e7000    32k          4K      8     0     2      6 process stack: ip6-mld-process
00007fa6c96de000    32k          4K      8     0     2      6 process stack: fib-walk
00007fa6c96d5000    32k          4K      8     0     1      7 process stack: session-queue-process
00007fa6c96cc000    32k          4K      8     0     2      6 process stack: virtio-send-interrupt-process
00007fa6c96c3000    32k          4K      8     0     2      6 process stack: vhost-user-process
00007fa6c96ba000    32k          4K      8     0     2      6 process stack: vhost-user-send-interrupt-process
00007fa6c96b1000    32k          4K      8     0     2      6 process stack: flow-report-process
00007fa6c96a8000    32k          4K      8     0     2      6 process stack: bfd-process
00007fa6c969f000    32k          4K      8     0     2      6 process stack: ip-neighbor-event
00007fa6c9696000    32k          4K      8     0     2      6 process stack: ip6-neighbor-age-process
00007fa6c968d000    32k          4K      8     0     3      5 process stack: ip4-neighbor-age-process
00007fa6c9684000    32k          4K      8     0     2      6 process stack: ip6-sv-reassembly-expire-walk
00007fa6c967b000    32k          4K      8     0     2      6 process stack: ip6-full-reassembly-expire-walk
00007fa6c9672000    32k          4K      8     0     2      6 process stack: ip4-sv-reassembly-expire-walk
00007fa6c9669000    32k          4K      8     0     2      6 process stack: ip4-full-reassembly-expire-walk
00007fa6c9660000    32k          4K      8     0     2      6 process stack: bond-process
00007fa6c9657000    32k          4K      8     0     2      6 process stack: l2fib-mac-age-scanner-process
00007fa6c964e000    32k          4K      8     0     2      6 process stack: l2-arp-term-publisher
00007fa6c9645000    32k          4K      8     0     2      6 process stack: vpe-link-state-process
00007fa6c9604000   256k          4K     64     0     3     61 process stack: startup-config-process
00007fa6c95c3000   256k          4K     64     0     2     62 process stack: unix-cli-stdin
0000002000001000    64m   10     4K  16384     0    17      0 session: evt-qs-segment
00007fa6c75c1000 32.00m          4K   8193     0     4   8189 tls segment
00007fa6a75bf000512.00m          4K 131073     0     4 131069 upf-proxy-server segment
00007fa69f5bd000128.00m          4K  32769     0     4  32765 upf-proxy-active-open segment
00007fa69b799000  1.00m          4K    257     0    86    171 upf-pfcp-server segment
00007fa69b744000   256k          4K     64     0     3     61 process stack: unix-cli-127.0.0.1:38520
00007fa69b733000    64k          4K     16     0     2     14 process stack: unix-cli-new-session
00007fa69b6f2000   256k          4K     64     0     2     62 process stack: unix-cli-127.0.0.1:39170

[hsychla]

Output of show memory main-heap

Thread 0 vpp_main
  base 0x7fa6ed0d0000, size 3g, locked, unmap-on-destroy, name 'main heap'
    page stats: page-size 4K, total 786432, mapped 417243, not-mapped 369176, unknown 13
      numa 0: 11430 pages, 44.65m bytes
      numa 1: 405813 pages, 1.55g bytes
    total: 2.99G, used: 1.46G, free: 1.54G, trimmable: 1.48G

[ivan4th]

I suggest that we run show memory main-heap a couple of days later and see if the reported heap usage is increasing. If it isn't, then the problem can be explained by internal heap usage patterns and likely will not cause any problems (VPP doesn't free the pages allocated for its heap when the parts of heap are freed internally; but this memory will be reused). If heap usage will be growing, we might have to resort to memory tracing.

[mgumz]

@ivan4th side-question: is main-heap available as metric exposed via vpp/upg-agent?

[hsychla]

@ivan4th new show memory main-heap:

Thread 0 vpp_main
  base 0x7fa6ed0d0000, size 3g, locked, unmap-on-destroy, name 'main heap'
    page stats: page-size 4K, total 786432, mapped 426532, not-mapped 359887, unknown 13
      numa 0: 11430 pages, 44.65m bytes
      numa 1: 415102 pages, 1.58g bytes
    total: 2.99G, used: 1.53G, free: 1.47G, trimmable: 1.44G

[mgumz]

Since upg-vpp:1.9.0 is released and running stable: i am closing this issue