Re: [PATCH proxmox-backup 3/5] bin: proxy: periodically schedule fstrim on datastore's filesystems

[Christian Ebner <c.ebner@proxmox.com>]

On 3/19/26 3:36 PM, Christian Ebner wrote:
> Run the fstrim command as scheduled job on datastores having a
> schedule defined in the config.
> 
> This is done since by default the systemd service to execute fstrim
> invokes the command with:
> 
> `/sbin/fstrim --listed-in /etc/fstab:/proc/self/mountinfo ...`
> 
> which however does not cover datastores created on top of a disk,
> since these are mounted via systemd mount units. fstrim however only
> evaluates the given list above up to the first non-empty file as
> stated in the man page [0].
> 
> [0] https://www.man7.org/linux/man-pages/man8/fstrim.8.html
> 
> Fixes: https://forum.proxmox.com/threads/181764/
> Signed-off-by: Christian Ebner <c.ebner@proxmox.com>
> ---
>   src/bin/proxmox-backup-proxy.rs | 59 +++++++++++++++++++++++++++++++++
>   1 file changed, 59 insertions(+)
> 
> diff --git a/src/bin/proxmox-backup-proxy.rs b/src/bin/proxmox-backup-proxy.rs
> index c1fe3ac15..1d375e69d 100644
> --- a/src/bin/proxmox-backup-proxy.rs
> +++ b/src/bin/proxmox-backup-proxy.rs
> @@ -508,6 +508,7 @@ async fn schedule_tasks() -> Result<(), Error> {
>       schedule_datastore_verify_jobs().await;
>       schedule_tape_backup_jobs().await;
>       schedule_task_log_rotate().await;
> +    schedule_fstrim().await;
>   
>       Ok(())
>   }
> @@ -879,6 +880,64 @@ async fn schedule_task_log_rotate() {
>       }
>   }
>   
> +async fn schedule_fstrim() {
> +    let config = match pbs_config::datastore::config() {
> +        Err(err) => {
> +            eprintln!("unable to read datastore config - {err}");
> +            return;
> +        }
> +        Ok((config, _digest)) => config,
> +    };
> +
> +    for (store, (_, store_config)) in config.sections {
> +        let store_config: DataStoreConfig = match serde_json::from_value(store_config) {
> +            Ok(c) => c,
> +            Err(err) => {
> +                eprintln!("datastore config from_value failed - {err}");
> +                continue;
> +            }
> +        };
> +
> +        let event_schedule = match &store_config.fstrim_schedule {
> +            Some(event_schedule) => event_schedule,
> +            None => continue,
> +        };
> +
> +        let worker_type = "fstrim";
> +        if check_schedule(worker_type, event_schedule, &store) {
> +            let mut job = match Job::new(worker_type, &store) {
> +                Ok(job) => job,
> +                Err(_) => continue, // could not get lock
> +            };
> +
> +            if let Err(err) = WorkerTask::new_thread(
> +                worker_type,
> +                None,
> +                Authid::root_auth_id().to_string(),
> +                false,
> +                move |worker| {
> +                    job.start(&worker.upid().to_string())?;
> +                    info!("executing fstrim on filesystem for {store}");
> +
> +                    let path = store_config.absolute_path();
> +                    let result = proxmox_backup::tools::disks::fstrim(Path::new(&path))
> +                        .map(|output| log::info!("{output}"));

Only realized now that this does not work since unprivileged, but this 
needs to be run as root. So this needs additional rework...

Other open questions still remain though.

> +
> +                    let status = worker.create_state(&result);
> +
> +                    if let Err(err) = job.finish(status) {
> +                        eprintln!("could not finish job state for {worker_type}: {err}");
> +                    }
> +
> +                    result
> +                },
> +            ) {
> +                eprintln!("unable to start fstrim task: {err}");
> +            }
> +        }
> +    }
> +}
> +
>   async fn command_reopen_access_logfiles() -> Result<(), Error> {
>       // only care about the most recent daemon instance for each, proxy & api, as other older ones
>       // should not respond to new requests anyway, but only finish their current one and then exit.