Cloudron makes it easy to run web apps like WordPress, Nextcloud, GitLab on your server. Find out more or install now.


Navigation

    Cloudron Forum

    • Register
    • Login
    • Search
    • Categories
    • Recent
    • Tags
    • Popular

    SOLVED CPU usage breakdown?

    Support
    contabo cpu graphs
    4
    16
    102
    Loading More Posts
    • Oldest to Newest
    • Newest to Oldest
    • Most Votes
    Reply
    • Reply as topic
    Log in to reply
    This topic has been deleted. Only users with topic management privileges can see it.
    • nebulon
      nebulon Staff last edited by

      One way is to determine the busy container right now first by using systemd-cgtop. It would would display the docker containers like /docker/003d4ff9290ce4b92223b3bb57195a73fe8a4f5b8e473e74f6e54575e4bdadcb where the uuid then can be grepped for to find the container with

      docker ps --no-trunc | grep 003d4ff9290ce4b92223b3bb57195a73fe8a4f5b8e473e74f6e54575e4bdadcb
      
      ruihildt 1 Reply Last reply Reply Quote 0
      • ruihildt
        ruihildt @nebulon last edited by ruihildt

        @nebulon Thanks, I'll do that.

        Is there a way to keep track of that in time or it only works in real time?

        Would it be possible to have it in the same way as we can keep track of ram in each app and in the system page per app, somewhere in the future? I have no idea if that's possible to implement or not. 😄

        1 Reply Last reply Reply Quote 0
        • rmdes
          rmdes last edited by

          What does it mean ?
          6a0fa665-3ea3-4f8a-9959-0d9a6754bbc3-image.png

          1 Reply Last reply Reply Quote 0
          • nebulon
            nebulon Staff last edited by

            The output is similar to top or htop, but that tool is aware of cgroups in linux, so it is easier to determine the app based on the container id.

            rmdes 1 Reply Last reply Reply Quote 0
            • rmdes
              rmdes last edited by

              Problem is the parts that are using the most CPU are :

              / is using 793-796 % CPU 14.3 Gb MEMORY
              cron.service 300+ % CPU 13 GB MEMORY
              system.slice
              docker.service
              box.service
              user.slice
              containerd.service

              Another problem or discrepancies I'm seeing if I dive into containers themselves is that it seems the containers that are using more CPU, the ones that often comes on top, are not tied to any domain/app in particular.

              if I docker ps a specific container i see on the systemd-cgtop no matter what I try to match them I cannot associate these containers ID to my list of ID-Domains when I type cloudron list

              it's like our box is running more containers than the number of app we have ?
              can a container be on the loose and still running and using resources ?

              It does not make sense I know 😕

              1 Reply Last reply Reply Quote 0
              • rmdes
                rmdes last edited by rmdes

                I keep having this drop_caches process eating all my CPU

                terminal.png

                I tried to kill this process but no matter the way I choose to terminate/kill it it just keeps going, something is occupying the system,it seems to me, outside of the docker environnement.

                1 Reply Last reply Reply Quote 0
                • rmdes
                  rmdes @nebulon last edited by

                  @nebulon running apps are not the problem, what I see is high CPU usage at the / root level of the box 1b17bca0-95e0-4eec-8b4c-577b5f29a827-image.png

                  1 Reply Last reply Reply Quote 0
                  • nebulon
                    nebulon Staff last edited by

                    Ok this is interesting, according to https://linux-mm.org/Drop_Caches this means the system tries to free up memory pages. And this essentially from time to time.
                    From Cloudron side we do not actively do this at all, so this is likely Ubuntu behavior. If this takes up a lot of cpu and for longer time periods, this may be related to the system I/O performance as such.

                    I have to read up further though, for example is this is swap related and then it will indeed be painful if the swap is on some slow disk.

                    1 Reply Last reply Reply Quote 1
                    • rmdes
                      rmdes last edited by rmdes

                      After last reboot,the process is now gone and CPU is more stable, quite crazy how this process completely freeze the cloudron box, Swap is being used at 338M over 4GB available and seem to increase very slowly, CPU is now around 50% more or less, still with Highs of 89%

                      i guess it will be better when on Ubuntu 20.x ? and moving to a new VPS or VDS soon probably 🙂

                      1 Reply Last reply Reply Quote 0
                      • nebulon
                        nebulon Staff last edited by

                        Not sure if Ubuntu 20.04 will change anything here, this seems to be a pretty normal thing since a long time for linux distros.

                        Also this may not indicate that you need more main memory or so, it might simply mean that the process of discarding and finding those resources takes long on your system. Thus my concern about system I/O speeds.

                        Which VPS provider are you currently using and is this an SSD system?

                        rmdes 1 Reply Last reply Reply Quote 0
                        • rmdes
                          rmdes @nebulon last edited by

                          @nebulon This is the VPS specs I'm using : https://contabo.com/?show=configurator&vserver_id=237
                          8 CPU cores
                          30 GB RAM (guaranteed)
                          800 GB disk space (100% SSD)
                          100% SSD disk space

                          1 Reply Last reply Reply Quote 0
                          • nebulon
                            nebulon Staff last edited by

                            This looks ok and good enough to me. I am always a bit unsure about disk I/O measurements and also if this is the root cause here at all, but anyways maybe you can do some sanity check with hdparm -t /dev/vda1 (replace /dev/vda1 with you main disk partition)

                            It should be somewhere above 500MB/s at least on provider with SSD in my experience.

                            1 Reply Last reply Reply Quote 2
                            • ruihildt
                              ruihildt last edited by ruihildt

                              FYI, CPU usage is now stable and around 13%, more or less since Contabo did, and I quote their email, a "technical adjustement".

                              You can clearly see on that monthly CPU chart, how it evolved (granted it's early days, but still):

                              89046ec8-7334-4c16-89cf-8504a749fb92-image.png

                              Not only this, but the dashboard now loads nearly instantly whereas beforehand, it could take up to 30 seconds.

                              Taking into account previous random issues with cheaper Contabo VPS in the past, I tend to think the issue was from Contabo from the start.

                              robi 1 Reply Last reply Reply Quote 4
                              • robi
                                robi @ruihildt last edited by

                                @ruihildt
                                nice to see it resolved outside of Cloudron.

                                If you want a similarly sized and less expensive system, I can share an invite link for ssd nodes that's been working well for us.

                                ruihildt 1 Reply Last reply Reply Quote 1
                                • ruihildt
                                  ruihildt @robi last edited by

                                  @robi Yes please. 🙂

                                  I also think next time I encounter a stability issue, I'll explore the provider angle sooner, maybe by replicating it elsewhere the same stack.

                                  1 Reply Last reply Reply Quote 1
                                  • First post
                                    Last post