Re: Question about compatibility between major kernel upgrades
Andrei Vagin <[email protected]> Tue, 14 Apr 2026 09:10:37 -0700
| Newsgroups | dev.linux.lists.criu |
|---|---|
| Message-ID | <CANaxB-wi=H=iReKKzBOS8_1a9xHE6iL+zNphU4fa3dyR0nnCXQ@mail.gmail.com> |
Hi Tom, Thank you for the kind words. Regarding your question, CRIU is designed to be backward compatible; restoring a checkpoint created on an older kernel onto a newer one is a supported and expected use case. However, a 5=E2=80=9310 year gap is longer than average, and such a large jump increases the risk of encountering edge cases or deprecated kernel features. I also want to ensure we are on the same page regarding the userspace: CRIU does not allow upgrading userspace libraries of checkpointed processes. This means all checkpointed containers must be restored using the same root file systems they were dumped with. You cannot dump a process and then expect it to run against a newer version of libc or other shared libraries upon restore. However, because the application's dependencies are isolated within a container, you can safely upgrade the host userspace, container engines, and other host-level components. Regarding the fallback, it should reduce the compatibility risk and it can be used as Plan B. Thanks, Andrei On Tue, Apr 14, 2026 at 5:27=E2=80=AFAM Tom Aviv <[email protected]> wrot= e: > > Hi, > > First, I'd like to thank you for the amazing work all of the supporters d= id for this important project > > I have a question about checkpoint/restore compatibility across major ker= nel upgrades. > > My company is evaluating CRIU as part of an OS upgrade flow. > The idea is to run a container with the target userspace on the old host = OS (so still on the old kernel), prepare our main process in the desired po= st-upgrade state, and dump it to disk with CRIU. > After upgrading the host OS and kernel, we would restore that process on = the upgraded system. > > Our main concern is long upgrade gaps. Some users may upgrade only once e= very 5-10 years. > In that case, how much can we rely on a CRIU dump created on the old syst= em being restorable on the newer kernel? > Is restore across major kernel upgrades something that is expected to wor= k, or is there no compatibility guarantee for that? > > As a fallback, we are considering running a VM with the target OS on the = old system, creating the CRIU dump there, and then restoring after the host= upgrade. > Would that reduce the compatibility risk compared to using a container wi= th the target userspace on the old host? > > For context, the host is Ubuntu LTS, and the container image or guest OS = would also be based on Ubuntu LTS. > > Thank you, > Tom