[pve-devel] [PATCH v4 pve-container 2/5] Propagate prestart-hook warnings to task-log.

[Robert Obkircher]

On 11/17/25 16:32, Fiona Ebner wrote:
> Am 17.11.25 um 12:31 PM schrieb Robert Obkircher:
>> On 11/13/25 17:35, Fiona Ebner wrote:
>>> Am 13.11.25 um 4:04 PM schrieb Robert Obkircher:
>>>> diff --git a/src/lxc-pve-prestart-hook b/src/lxc-pve-prestart-hook
>>>> index 73125e1..f5dd728 100755
>>>> --- a/src/lxc-pve-prestart-hook
>>>> +++ b/src/lxc-pve-prestart-hook
>>>> @@ -28,17 +28,6 @@ eval {
>>>>        $have_sdn = 1;
>>>>    };
>>>>    -my $WARNFD;
>>>> -
>>>> -sub log_warn {
>>>> -    my ($vmid, $message) = @_;
>>>> -
>>>> -    if (!defined($WARNFD)) {
>>>> -        open($WARNFD, '>', "/run/pve/ct-${vmid}.warnings");
>>>> -    }
>>>> -    print $WARNFD "$message\n";
>>>> -}
>>>> -
>>>>    PVE::LXC::Tools::lxc_hook(
>>>>        'pre-start',
>>>>        'lxc',
>>>> @@ -53,6 +42,15 @@ PVE::LXC::Tools::lxc_hook(
>>>>              PVE::RESTEnvironment->setup_default_cli_env();
>>>>    +        my $warn_file = "/run/pve/ct-${vmid}.warnings";
>>>> +        # open eagerly so logging works inside the protected_call
>>>> chroot
>>> Nit: Maybe "early" instead of "eagerly"?
>> I was specifically thinking about eager/lazy evaluation.
> Personally, I wouldn't call having code inside a CODE ref/sub variable
> "lazily evaluated". The code hasn't been called yet, not just not
> evaluated. Or do you mean something else?

What I meant was that $WARNFD itself used to be lazily initialized on 
first use (like a LazyLock in Rust) and I moved that initialization to 
the definition to ensure that it happens outside of a potential chroot.

I will send a v6 with a better comment and a small change to the path 
concatenation in the fourth commit.