Telegraf 1.26.2. Memory Input Plugin - inputs.mem. mem.write_back - zero value.

[syatihoko]

Relevant telegraf.conf

[global_tags]


[agent]
  interval = "30s"
  round_interval = false

  metric_batch_size = 10000

  metric_buffer_limit = 100000

  collection_jitter = "0s"

  flush_interval = "10s"
  flush_jitter = "0s"

  precision = ""

  debug = true


  logfile = "/var/log/telegraf/telegraf.log"



  hostname = "name_host"
  omit_hostname = false


 [[outputs.graphite]]
   servers = ["MY_HIDDEN_HOST:2003"]
   prefix = "servers"
   template = "host.measurement.tags.field"
   timeout = 17

Logs from Telegraf

2023-05-22T07:52:46Z I! [agent] Hang on, flushing any cached metrics before shutdown
2023-05-22T07:52:46Z I! [agent] Stopping running outputs
2023-05-22T07:52:46Z I! Starting Telegraf 1.26.2
2023-05-22T07:52:46Z I! Available plugins: 235 inputs, 9 aggregators, 27 processors, 22 parsers, 57 outputs, 2 secret-stores
@@@

System info

Telegraf 1.26.2. Ubuntu 18.04, 22.04

Docker

No response

Steps to reproduce

1. telegraf --test -config /etc/telegraf/telegraf.conf -config-directory /etc/telegraf/telegraf.d/common.conf --input-filter=mem

Expected behavior

write_back= and write_back_tmp more 0

Actual behavior

write_back=0i,write_back_tmp=0i

Additional info

After update to Telegraf 1.26.2, the metric "mem.write_back" began to accept only a zero value on all servers.
OS - Ubuntu 18.04, 22.04

telegraf --test -config /etc/telegraf/telegraf.conf -config-directory /etc/telegraf/telegraf.d/ --input-filter=mem:
active=24328122368i,available=25401221120i,available_percent=75.5993608248227,buffered=135442432i,cached=25453924352i,commit_limit=18947371008i,committed_as=8156454912i,dirty=37199872i,free=299200512i,high_free=0i,high_total=0i,huge_page_size=2097152i,huge_pages_free=0i,huge_pages_total=0i,inactive=7386312704i,low_free=0i,low_total=0i,mapped=164855808i,page_tables=23384064i,shared=917504i,slab=1301299200i,sreclaimable=1240989696i,sunreclaim=60309504i,swap_cached=278528i,swap_free=2143285248i,swap_total=2147479552i,total=33599782912i,used=7711215616i,used_percent=22.95019475630593,vmalloc_chunk=0i,vmalloc_total=35184372087808i,vmalloc_used=0i,write_back=0i,write_back_tmp=0i 1684741645000000000

[syatihoko]

Unfortunately, there is no whole story - the version where it continues to work:

Telegraf 1.20.4
Telegraf 1.20.0
Telegraf 1.19.1
Telegraf 1.15.2

[syatihoko]

Run command grep -e Dirty: -e Writeback: /proc/meminfo on Host where problem:

11:38
Dirty:             35584 kB
Writeback:       3450572 kB
11:44
Dirty:             31928 kB
Writeback:       3450608 kB
11:46 
Dirty:             81264 kB
Writeback:       3450616 kB

[powersj]

Hi,

The memory plugin data comes from gopsutil. It would help greatly if you could narrow down to what version of telegraf was the last working and the first broken. From there we can determine which version of gopsutil no longer reports the value correctly.

My guess is that it is because upstream is looking for WriteBack and not Writeback, note the lower-case b.

Would you be willing to file an upstream issue?

[syatihoko]

Hello.

Would you be willing to file an upstream issue?
Sure.

I'll try to calculate the specific version that had the problem

[powersj]

Thanks - please do let us know what version you track it down to and if you file an upstream issue.

[syatihoko]

Hi, finished the tests:
telegraf_1.20.4-1_amd64.deb. - work
..
telegraf_1.21.0~rc1-1_amd64.deb. - doesn't work

[powersj]

Looking at our diff, it looks like there were two changes to our gopsutil usage between v1.20.4 and 1.21.0-rc1:

• fix: bump github.com/shirou/gopsutil from 3.21.6-0.20210624221800-cb512c850043+incompatible to 3.21.8+incompatible #9760
• fix: update shirou/gopsutil to v3 #10119

I went through the diff for v3.21.8 and v3.21.10 and was able to compare the v3 and previous version:

• Previously the case looked for Writeback
• Now it looks for WriteBack

I have filed an upstream issue shirou/gopsutil#1471 and put up shirou/gopsutil#1472 to fix the capitalization.

[powersj]

My upstream PR was just merged! I think the maintainer should do a release soon, then we can get that in telegraf with the fix.