This page is very important (I believe). It describes what can change in a tasks' environment after a C/R cycle. If a software you're using depends on either item from the list below, it may break after C/R. The list is incomplete, but we do our best to make it contain relevant data.
- Per-task statistics
- Various counters, that can be obtained via
/proc/$pid/status
orgetrusage()
- Namespaces' IDs
- Numbers you see in the
/proc/$pid/ns/*
links' targets
- Process start time
- It's in the 22nd field of the
/proc/$pid/stat
file
- Mount points IDs
- The numbers from first 2 columns of
/proc/$pid/mountinfo
- Sockets IDs
- If you stat() a file descriptor with a socket the st_ino value can be used as unique socket ID. This value does changes after C/R as these IDs are global and we have no API to restore one.
- VDSO
- The VDSO is a shared library linked into app by kernel. If the C/R cycle implies changing the kernel in between (e.g. live migration or seamless kernel upgrade, the VDSO contents may change. However, its linkage with the application is preserved.