Andrew Clayton

The normal way to do this is as zh1029 says and use AF_UNSPEC with a single call to getaddrinfo() Underneath, getaddrinfo() will race to find working IPv4 and IPv6 addresses,...

One open question is how to handle failures to create or add processes to cgroups, I see three options. a) Ignore failures and carry on b) Fail hard c) Carry...

On Mon, 18 Jul 2022 15:39:29 -0700 Alejandro Colomar ***@***.***> wrote: > If a client can trigger the failure, we shouldn't fail, but I don't expect that to be possible....

On Mon, 18 Jul 2022 15:49:00 -0700 Alejandro Colomar ***@***.***> wrote: > @alejandro-colomar commented on this pull request. > > > > > +static nxt_int_t > +nxt_cgroup_create(const nxt_task_t *task, nxt_process_t...

This updated commit should address all the above feedback (except the error handling). Notable changes are - The removal of nxt_make_cgpaths(), it was introduced for a problem that no longer...

Apart from most the above, the biggest change here is to unconditionally include the cgroup_cleanup member in nxt_process_isolation_t, that way we can get rid of the #if/#endif from nxt_main_process_cleanup().

Changes this time 1) cgroup name validation 2) mark the cgroup name as required 3) check the return value of the fprintf(3) in nxt_cgroup_proc_add()

I'll give it some thought, but on the face of it, that sounds reasonable.

- Implemented @i4ki's suggestion. - Updated the commit message to reflect this change. So essentially it means that specifying an absolute path will retain the old behaviour of creating the...

This has the following changes - Don't bother forbidding '/./' in the cgroup path - Forbid an empty cgroup path - Use getline(3) instead of fscanf(3) for reading /proc/self/cgroup -...