Merge pull request #8085 from tangledbytes/utkarsh/backport/5.15.3/ta…

…pecloud-fix-1

[Backport 5.15] [NC | NSFS | Glacier] filter out failed recalls

config.js

@@ -755,6 +755,11 @@ config.NSFS_GLACIER_BACKEND = 'TAPECLOUD';
 // TAPECLOUD Glacier backend specific configs
 config.NSFS_GLACIER_TAPECLOUD_BIN_DIR = '/opt/ibm/tapecloud/bin';
 
+// If set to true will disable cleanup of the task show output
+// Should be used only for debugging or else will keep filling
+// up the disk space.
+config.NSFS_GLACIER_TAPECLOUD_PRESERVE_TASK_SHOW_OUTPUT = false;
+
 // NSFS_GLACIER_MIGRATE_INTERVAL indicates the interval between runs
 // of `manage_nsfs glacier migrate`
 config.NSFS_GLACIER_MIGRATE_INTERVAL = 15 * 60 * 1000;

src/sdk/nsfs_glacier_backend/backend.js

@@ -263,6 +263,14 @@ class GlacierBackend {
         const restore_status = GlacierBackend.get_restore_status(stat.xattr, new Date(), file);
         if (!restore_status) return false;
 
+        // We don't check for pre-existing expiry here, it can happen in 2 cases
+        // 1. A restore is already going and someone somehow initiated this second
+        // call. In that case we might see partial extended attributes such that
+        // both request as well a future expiry time exists.
+        // 2. A restore request was partially processed and then failed before
+        // removing the request extended attribute. In such case, NSFS would still
+        // report the object restore status to be `ONGOING` and we are going
+        // to allow a retry of that entry.
         return restore_status.state === GlacierBackend.RESTORE_STATUS_ONGOING;
     }
 }