[pbs-devel] [PATCH proxmox-backup v2] tools/daemon: improve reload behaviour

[Fabian Ebner]

Still works for me. Tested 250 installs with a few 'stress' workloads on 
the side.

Tested-By: Fabian Ebner <f.ebner at proxmox.com>

Am 17.12.20 um 17:17 schrieb Thomas Lamprecht:
> On 17/12/2020 15:50, Dominik Csapak wrote:
>> it seems that sometimes, the child process signal gets handled
>> before the parent process signal. Systemd then ignores the
>> childs signal (finished reloading) and only after going into
>> reloading state because of the parent. this will never finish.
>>
>> Instead, wait for the state to change to 'reloading' after sending
>> that signal in the parent, an only fork afterwards. This way
>> we ensure that systemd knows about the reloading before actually trying
>> to do it.
>>
>> Signed-off-by: Dominik Csapak <d.csapak at proxmox.com>
>> ---
>> changes from v1:
>> * introduce wait_service_is_(not_)state
>>      it is a bit more generic
>>      has a better name
>> * factor the common code out into get_service_state
>>
> 
> that seems much nicer, thanks!
> @Fabian can you please redo your testing (1.5k is maybe a bit much in light
> of the changes between v1 and v2 I'm happy with 100 ^^ too)
> 
> I'm quite confident in the patch, but this is not something I'd like to touch
> again soon, so better we check it a bit to much than to less :)
> 
>>   src/tools/daemon.rs | 45 ++++++++++++++++++++++++++++-----------------
>>   1 file changed, 28 insertions(+), 17 deletions(-)
>>
>> diff --git a/src/tools/daemon.rs b/src/tools/daemon.rs
>> index 6bb4a41b..0e3a174a 100644
>> --- a/src/tools/daemon.rs
>> +++ b/src/tools/daemon.rs
>> @@ -291,6 +291,7 @@ where
>>           if let Err(e) = systemd_notify(SystemdNotify::Reloading) {
>>               log::error!("failed to notify systemd about the state change: {}", e);
>>           }
>> +        wait_service_is_state(service_name, "reloading").await?;
>>           if let Err(e) = reloader.take().unwrap().fork_restart() {
>>               log::error!("error during reload: {}", e);
>>               let _ = systemd_notify(SystemdNotify::Status("error during reload".to_string()));
>> @@ -305,7 +306,7 @@ where
>>   
>>       // FIXME: this is a hack, replace with sd_notify_barrier when available
>>       if server::is_reload_request() {
>> -        wait_service_is_active(service_name).await?;
>> +        wait_service_is_not_state(service_name, "reloading").await?;
>>       }
>>   
>>       log::info!("daemon shut down...");
>> @@ -313,26 +314,36 @@ where
>>   }
>>   
>>   // hack, do not use if unsure!
>> -async fn wait_service_is_active(service: &str) -> Result<(), Error> {
>> +async fn get_service_state(service: &str) -> Result<String, Error> {
>> +    let text = match tokio::process::Command::new("systemctl")
>> +        .args(&["is-active", service])
>> +        .output()
>> +        .await
>> +    {
>> +        Ok(output) => match String::from_utf8(output.stdout) {
>> +            Ok(text) => text,
>> +            Err(err) => bail!("output of 'systemctl is-active' not valid UTF-8 - {}", err),
>> +        },
>> +        Err(err) => bail!("executing 'systemctl is-active' failed - {}", err),
>> +    };
>> +
>> +    Ok(text.trim().trim_start().to_string())
>> +}
>> +
>> +async fn wait_service_is_state(service: &str, state: &str) -> Result<(), Error> {
>>       tokio::time::delay_for(std::time::Duration::new(1, 0)).await;
>> -    loop {
>> -        let text = match tokio::process::Command::new("systemctl")
>> -            .args(&["is-active", service])
>> -            .output()
>> -            .await
>> -        {
>> -            Ok(output) => match String::from_utf8(output.stdout) {
>> -                Ok(text) => text,
>> -                Err(err) => bail!("output of 'systemctl is-active' not valid UTF-8 - {}", err),
>> -            },
>> -            Err(err) => bail!("executing 'systemctl is-active' failed - {}", err),
>> -        };
>> +    while get_service_state(service).await? != state {
>> +        tokio::time::delay_for(std::time::Duration::new(5, 0)).await;
>> +    }
>> +    Ok(())
>> +}
>>   
>> -        if text.trim().trim_start() != "reloading" {
>> -            return Ok(());
>> -        }
>> +async fn wait_service_is_not_state(service: &str, state: &str) -> Result<(), Error> {
>> +    tokio::time::delay_for(std::time::Duration::new(1, 0)).await;
>> +    while get_service_state(service).await? == state {
>>           tokio::time::delay_for(std::time::Duration::new(5, 0)).await;
>>       }
>> +    Ok(())
>>   }
>>   
>>   #[link(name = "systemd")]
>>
>