+++ This bug was initially created as a clone of Bug #2168346 +++ -------------------------------------------------- Description of problem: -------------------------------------------------- as a part of an OOM investigation, I deliberately attempted hitting OOM on a VM by straining the memory while using a heavy IO workload,but something unexpected occurred when instead of just OOM which causes QEMU reboot the VM failed to run again: NAME AGE STATUS READY rhel82-vm0001 26h CrashLoopBackOff False rhel82-vm0002 26h Running True rhel82-vm0003 26h Stopped False pod logs : {"component":"virt-launcher","kind":"","level":"error","msg":"Failed to sync vmi","name":"rhel82-vm0001","namespace":"default","pos":"server.go:184","reason":"virError(Code=1, Domain=10, Message='internal error: UNIX socket path '/var/run/kubevirt-private/libvirt/qemu/channel/target/domain-212-default_rhel82-vm000/org.qemu.guest_agent.0' too long')","timestamp":"2023-02-08T18:36:49.591084Z","uid":"0a6404c5-2ba7-4cc0-ad2a-307018174023"} {"component":"virt-launcher","level":"info","msg":"Still missing PID for default_rhel82-vm0001, open /run/libvirt/qemu/run/default_rhel82-vm0001.pid: no such file or directory","pos":"monitor.go:125","timestamp":"2023-02-08T18:36:49.664614Z"} {"component":"virt-launcher","level":"error","msg":"internal error: UNIX socket path '/var/run/kubevirt-private/libvirt/qemu/channel/target/domain-213-default_rhel82-vm000/org.qemu.guest_agent.0' too long","pos":"qemuOpenChrChardevUNIXSocket:5223","subcomponent":"libvirt","thread":"30","timestamp":"2023-02-08T18:36:50.627000Z"} {"component":"virt-launcher","kind":"","level":"error","msg":"Failed to start VirtualMachineInstance with flags 0.","name":"rhel82-vm0001","namespace":"default","pos":"manager.go:880","reason":"virError(Code=1, Domain=10, Message='internal error: UNIX socket path '/var/run/kubevirt-private/libvirt/qemu/channel/target/domain-213-default_rhel82-vm000/org.qemu.guest_agent.0' too long')","timestamp":"2023-02-08T18:36:50.628244Z","uid":"0a6404c5-2ba7-4cc0-ad2a-307018174023"} {"component":"virt-launcher","kind":"","level":"error","msg":"Failed to sync vmi","name":"rhel82-vm0001","namespace":"default","pos":"server.go:184","reason":"virError(Code=1, Domain=10, Message='internal error: UNIX socket path '/var/run/kubevirt-private/libvirt/qemu/channel/target/domain-213-default_rhel82-vm000/org.qemu.guest_agent.0' too long')","timestamp":"2023-02-08T18:36:50.628304Z","uid":"0a6404c5-2ba7-4cc0-ad2a-307018174023"} {"component":"virt-launcher","level":"info","msg":"Still missing PID for default_rhel82-vm0001, open /run/libvirt/qemu/run/default_rhel82-vm0001.pid: no such file or directory","pos":"monitor.go:125","timestamp":"2023-02-08T18:36:50.663725Z"} {"component":"virt-launcher","level":"info","msg":"Still missing PID for default_rhel82-vm0001, open /run/libvirt/qemu/run/default_rhel82-vm0001.pid: no such file or directory","pos":"monitor.go:125","timestamp":"2023-02-08T18:36:51.663684Z"} {"component":"virt-launcher","level":"error","msg":"internal error: UNIX socket path '/var/run/kubevirt-private/libvirt/qemu/channel/target/domain-214-default_rhel82-vm000/org.qemu.guest_agent.0' too long","pos":"qemuOpenChrChardevUNIXSocket:5223","subcomponent":"libvirt","thread":"29","timestamp":"2023-02-08T18:36:51.663000Z"} {"component":"virt-launcher","kind":"","level":"error","msg":"Failed to start VirtualMachineInstance with flags 0.","name":"rhel82-vm0001","namespace":"default","pos":"manager.go:880","reason":"virError(Code=1, Domain=10, Message='internal error: UNIX socket path '/var/run/kubevirt-private/libvirt/qemu/channel/target/domain-214-default_rhel82-vm000/org.qemu.guest_agent.0' too long')","timestamp":"2023-02-08T18:36:51.664818Z","uid":"0a6404c5-2ba7-4cc0-ad2a-307018174023"} {"component":"virt-launcher","kind":"","level":"error","msg":"Failed to sync vmi","name":"rhel82-vm0001","namespace":"default","pos":"server.go:184","reason":"virError(Code=1, Domain=10, Message='internal error: UNIX socket path '/var/run/kubevirt-private/libvirt/qemu/channel/target/domain-214-default_rhel82-vm00 0/org.qemu.guest_agent.0' too long')","timestamp":"2023-02-08T18:36:51.664884Z","uid":"0a6404c5-2ba7-4cc0-ad2a-307018174023"} OOM record: [Wed Feb 8 12:19:15 2023] worker invoked oom-killer: gfp_mask=0x620100(GFP_NOIO|__GFP_HARDWALL|__GFP_WRITE), order=0, oom_score_adj=979 [Wed Feb 8 12:19:15 2023] oom_kill_process.cold.32+0xb/0x10 [Wed Feb 8 12:19:15 2023] [ pid ] uid tgid total_vm rss pgtables_bytes swapents oom_score_adj name [Wed Feb 8 12:19:15 2023] oom-kill:constraint=CONSTRAINT_MEMCG,nodemask=(null),cpuset=crio-a2021e5dd93338ba5e39cef21c773838a294ab95a466c7887054e9e24f72e8e4.scope,mems_allowed=0-1,oom_memcg=/kubepods.slice/kubepods-burstable.slice/kubepods-burstable-pod6553054d_e923_4628_b36c_c6754eb6e0b1.slice,task_memcg=/kubepods.slice/kubepods-burstable.slice/kubepods-burstable-pod6553054d_e923_4628_b36c_c6754eb6e0b1.slice/crio-a2021e5dd93338ba5e39cef21c773838a294ab95a466c7887054e9e24f72e8e4.scope,task=qemu-kvm,pid=3196344,uid=107 [Wed Feb 8 12:19:15 2023] Memory cgroup out of memory: Killed process 3196344 (qemu-kvm) total-vm:64560756kB, anon-rss:58285188kB, file-rss:17672kB, shmem-rss:4kB, UID:107 pgtables:115428kB oom_score_adj:979 [Wed Feb 8 12:19:15 2023] oom_reaper: reaped process 3196344 (qemu-kvm), now anon-rss:0kB, file-rss:68kB, shmem-rss:4kB -------------------------------------------------- Version-Release number of selected component (if applicable): -------------------------------------------------- kubevirt-hyperconverged-operator.v4.11.3 local-storage-operator.v4.12.0-202301042354 mcg-operator.v4.11.4 ocs-operator.v4.11.4 -------------------------------------------------- How reproducible: -------------------------------------------------- no idea but the current state is persistent throughout. -------------------------------------------------- Steps to Reproduce: -------------------------------------------------- 1. strain the VM using a heavy-duty workload 2. reach OOM 3. repeat -------------------------------------------------- Actual results: -------------------------------------------------- VM no longer boot. -------------------------------------------------- Expected results: -------------------------------------------------- VM reboots and starts normally -------------------------------------------------- logs: -------------------------------------------------- I collect both must gather and the SOS report from the specific node that ran the VM http://perf148h.perf.lab.eng.bos.redhat.com/share/BZ_logs/vm_doesnt_boot_after_oom.tar.gz --- Additional comment from Yan Du on 2023-02-15 13:14:50 UTC --- It doesn't look like storage component, move to Virt component. --- Additional comment from Jed Lejosne on 2023-02-27 18:55:41 UTC --- There's an interesting error there: internal error: UNIX socket path '/var/run/kubevirt-private/libvirt/qemu/channel/target/domain-214-default_rhel82-vm000/org.qemu.guest_agent.0' too long This indeed 108 characters long, 1 more than the 107 allowed by Linux. I think "214" here is the number of times the VM rebooted. This means VMs can only be rebooted 98 times. We need to address that. I don't see why, as far as libvirt is concerned, VMs couldn't just be called "vm" instead of "<namespace>_<VMI name>". --- Additional comment from Jed Lejosne on 2023-02-27 19:44:51 UTC --- (In reply to Jed Lejosne from comment #2) > [...] > This means VMs can only be rebooted 98 times. We need to address that. This is actually incorrect, VMs need to actually crash for that number to increase, so that's not such a big deal. However, @bbenshab, please give more information on how you managed to trigger the OOM killer. If that was solely by doing things from inside the guest, then that's a problem. No matter what guests do, that should cause virt-launcher to run out of memory... --- Additional comment from Boaz on 2023-02-28 10:29:42 UTC --- @jlejosne this was an investigation of a customer case that is described here: https://docs.google.com/document/d/1bMWAkw7Scp98XgXmtVH-vRD_YOjB2xUW7KUcs8SeUSg
The maximum socket path in Linux is 107. In KubeVirt, we can end up with a 108 characters path for guest-agent, like: /var/run/kubevirt-private/libvirt/qemu/channel/target/domain-214-default_rhel82-vm000/org.qemu.guest_agent.0 We need to find a way to either shorten the path or replace it with something else, like a file descriptor.
I see couple ways to fix this issue: 1) make the cfg->channelDir configurable so that mgmt apps can skip "/libvirt/qemu/channel/target/" part, or 2) KubeVirt may use shorter VM name, 3) Libvirt may generate even shorter "shortened names" (currently 20 chars), since we're guaranteed to be unique regardless of short name length (due to domain ID in the path). It was also suggested that libvirt could just pass FD to QEMU. I did not realize it back then, but libvirt's already doing that, e.g.: -chardev socket,id=charchannel0,fd=32,server=on,wait=off \ -device {"driver":"virtserialport","bus":"virtio-serial0.0","nr":1,"chardev":"charchannel0","id":"channel0","name":"org.qemu.guest_agent.0"} but that won't really fly when libvirtd restarts and needs to reconnect back to the socket. Unless there's a way in QEMU to provide a new fd, in which case libvirt could just use an unnamed socket and tell QEMU the new FD on reconnect.
Reproduce this bug with: libvirt-9.0.0-7.el9.x86_64 1. Create directory for guest agent socket source: # mkdir /var/run/kubevirt-private/libvirt/qemu/channel/target/domain-212-default_rhel82-vm000/ -p 2. Change the owner and selinux context of the directory: # chown qemu:qemu /var/run/kubevirt-private/libvirt/qemu/channel/target/domain-212-default_rhel82-vm000/ # chcon system_u:object_r:qemu_var_run_t:s0 /var/run/kubevirt-private/libvirt/qemu/channel/target/domain-212-default_rhel82-vm000/ 3. prepare a guest with the following xml snippet <channel type='unix'> <source mode='bind' path='/var/run/kubevirt-private/libvirt/qemu/channel/target/domain-212-default_rhel82-vm000/org.qemu.guest_agent.1'/> <target type='virtio' name='org.qemu.guest_agent.0'/> <address type='virtio-serial' controller='0' bus='0' port='2'/> </channel> 4. start the guest # virsh start vm1 error: Failed to start domain 'vm1' error: internal error: UNIX socket path '/var/run/kubevirt-private/libvirt/qemu/channel/target/domain-212-default_rhel82-vm000/org.qemu.guest_agent.1' too long Additional info: If I change the last character of the path to '0', i.e., /var/run/kubevirt-private/libvirt/qemu/channel/target/domain-212-default_rhel82-vm000/org.qemu.guest_agent.0 <channel type='unix'> <source mode='bind' path='/var/run/kubevirt-private/libvirt/qemu/channel/target/domain-212-default_rhel82-vm000/org.qemu.guest_agent.0'/> <target type='virtio' name='org.qemu.guest_agent.0'/> <address type='virtio-serial' controller='0' bus='0' port='2'/> </channel> libvirt will remove the specified path, and use the default socket path, <channel type='unix'> <source mode='bind' path='/var/lib/libvirt/qemu/channel/target/domain-1-vm1/org.qemu.guest_agent.0'/> <target type='virtio' name='org.qemu.guest_agent.0' state='disconnected'/> <alias name='channel0'/> <address type='virtio-serial' controller='0' bus='0' port='2'/> </channel> Hi, Michal I am curious why the above path "/var/run/kubevirt-private/libvirt/qemu/channel/target/domain-212-default_rhel82-vm000/org.qemu.guest_agent.0" was not removed in cnv. Please help to check if you are available. Thanks very much.
(In reply to Lili Zhu from comment #3) > Hi, Michal > > I am curious why the above path > "/var/run/kubevirt-private/libvirt/qemu/channel/target/domain-212- > default_rhel82-vm000/org.qemu.guest_agent.0" was not removed in cnv. Please > help to check if you are available. Thanks very much. I believe they don't specify the path. For the guest agent, libvirt puts the socket into so called channelTargetDir, which is derived as follows: virQEMUDriverConfig * virQEMUDriverConfigNew(bool privileged, const char *root) { if (root != NULL) { cfg->channelTargetDir = g_strdup_printf("%s/channel/target", cfg->libDir); } else if (privileged) { cfg->channelTargetDir = g_strdup_printf("%s/channel/target", cfg->libDir); } else { cfg->configBaseDir = virGetUserConfigDirectory(); cfg->channelTargetDir = g_strdup_printf("%s/qemu/channel/target", cfg->configBaseDir); } } and since CNV uses qemu:///session the third option is used. In here, virGetUserConfigDirectory() basically returns "$XDG_CONFIG_HOME/libvirt". In other words, this path is autogenerated by libvirt and as the domain ID grows, it also prolongs the socket path. And since the path was long enough to begin with, after couple of iterations, the domain ID is large enough to make the generated path not fit into the limit.
Hi Michal, Lili, Can you please update on the bug status? would you pick it for upcoming release? TIA, Igor
Patches proposed on the list: https://listman.redhat.com/archives/libvir-list/2023-July/240680.html
v2: https://listman.redhat.com/archives/libvir-list/2023-July/240904.html
*** Bug 2225204 has been marked as a duplicate of this bug. ***
resend: https://listman.redhat.com/archives/libvir-list/2023-August/241235.html
Merged upstream as: 8abc979bb0 qemu: Move channelTargetDir into stateDir d3759d3674 qemu: Generate shorter channel target paths v9.6.0-47-g8abc979bb0