[pbs-devel] [PATCH proxmox-backup] proxy: rrd: skip update disk stats for offline datastores

[pbs-devel] [PATCH proxmox-backup] proxy: rrd: skip update disk stats for offline datastores

[Hannes Laimer]

RDD update did not use lookup_datastore() and therefore bypassed
the maintenance mode checks. This adds the needed check directly.

Signed-off-by: Hannes Laimer <h.laimer@proxmox.com>
---
 src/bin/proxmox-backup-proxy.rs | 6 ++++++
 1 file changed, 6 insertions(+)

diff --git a/src/bin/proxmox-backup-proxy.rs b/src/bin/proxmox-backup-proxy.rs
index 6a305b7c..c71d0009 100644
--- a/src/bin/proxmox-backup-proxy.rs
+++ b/src/bin/proxmox-backup-proxy.rs
@@ -1071,6 +1071,12 @@ fn generate_host_stats_sync() {
                 .unwrap_or_default();
 
             for config in datastore_list {
+                if config
+                    .get_maintenance_mode()
+                    .map_or(false, |mode| mode.check(Some(Operation::Read)).is_err())
+                {
+                    continue;
+                }
                 let rrd_prefix = format!("datastore/{}", config.name);
                 let path = std::path::Path::new(&config.path);
                 gather_disk_stats(disk_manager.clone(), path, &rrd_prefix);
-- 
2.30.2

[pbs-devel] applied: [PATCH proxmox-backup] proxy: rrd: skip update disk stats for offline datastores

[Thomas Lamprecht]

Am 5/4/22 um 07:49 schrieb Hannes Laimer:
> RDD update did not use lookup_datastore() and therefore bypassed
> the maintenance mode checks. This adds the needed check directly.
> 
> Signed-off-by: Hannes Laimer <h.laimer@proxmox.com>
> ---
>  src/bin/proxmox-backup-proxy.rs | 6 ++++++
>  1 file changed, 6 insertions(+)
> 
>

applied, thanks!

But there was a similar issue in the schedule_datastore_garbage_collection fn,
which tried to look up the Datastore with a Write operation at the very start,
before even checking if GC for that store is scheduled at all and if it would be
pending now, and the actual problematic check if GC is already running (which needs
a datastore instance, so just moving the lookup code downward couldn't be the single
solution).

For that I added a new Lookup operation in a limited scope, and then open the
datastore with the Write op only directly before actually wanting to start GC,
at which point a log entry is ok.