[pve-devel] [PATCH 1/3] add check_connection in storage_check_enabled.

[Alexandre Derumier]

Currently, if a storage have an infinite timeout (nfs hard option or iscsi with high timeout for failover),
and if the storage is down or is doing ha failover with dual controller

- pvestatd hang on storage_info. (So pve-manager see the host as down, and hang on differents parts, gui,rrds display)

- pvedaemon hang on differents place (storage disk list,vm_status, hardware info,...)

- I also see a lot of qemu-img info process hanging in uninterruptible sleep state with iscsi storage.

This patch add a check to see if the storage is accessible.

For nfs and iscsi : ping with a timeout of 2 second.  (Maybe this can be polish with some kind of filesystem/storage read test with timeout...)

Default for other storage plugins is to check nothing. (not a problem with sheepdog,ceph,libiscsi... as commands have timeout)

Signed-off-by: Alexandre Derumier <aderumier at odiso.com>
---
 PVE/Storage.pm        |    6 ++++++
 PVE/Storage/Plugin.pm |    6 ++++++
 2 files changed, 12 insertions(+), 0 deletions(-)

diff --git a/PVE/Storage.pm b/PVE/Storage.pm
index c86ce0c..4b0805c 100755
--- a/PVE/Storage.pm
+++ b/PVE/Storage.pm
@@ -93,6 +93,12 @@ sub storage_check_enabled {
 	return undef;
     }
 
+    my $plugin = PVE::Storage::Plugin->lookup($scfg->{type});
+    if (!$plugin->check_connection($storeid, $scfg)) {
+	die "storage '$storeid' is not available\n" if !$noerr;
+	return undef;
+    }
+
     return storage_check_node($cfg, $storeid, $node, $noerr);
 }
 
diff --git a/PVE/Storage/Plugin.pm b/PVE/Storage/Plugin.pm
index 7c0e3e9..27f004c 100644
--- a/PVE/Storage/Plugin.pm
+++ b/PVE/Storage/Plugin.pm
@@ -579,4 +579,10 @@ sub deactivate_volume {
     # do nothing by default
 }
 
+sub check_connection {
+    my ($class, $storeid, $scfg) = @_;
+    # do nothing by default
+    return 1;
+}
+
 1;
-- 
1.7.2.5