add complete support for tape reclaim

Signed-off-by: Utkarsh Srivastava <srivastavautkarsh8097@gmail.com>

Author: tangledbytes

config.js

@@ -928,6 +928,9 @@ config.NSFS_GLACIER_DMAPI_PMIG_DAYS = config.S3_RESTORE_REQUEST_MAX_DAYS;
 // accidental blocking reads from happening.
 config.NSFS_GLACIER_DMAPI_FINALIZE_RESTORE_ENABLE = false;
 
+config.NSFS_GLACIER_DMAPI_ENABLE_TAPE_RECLAIM = false;
+config.NSFS_GLACIER_RECLAIM_INTERVAL = 15 * 60 * 1000;
+
 config.NSFS_STATFS_CACHE_SIZE = 10000;
 config.NSFS_STATFS_CACHE_EXPIRY_MS = 1 * 1000;
 
@@ -971,7 +974,7 @@ config.NSFS_GLACIER_MIGRATE_LOG_THRESHOLD = 50 * 1024;
 config.NSFS_GLACIER_METRICS_STATFS_PATHS = [];
 config.NSFS_GLACIER_METRICS_STATFS_INTERVAL = 60 * 1000; // Refresh statfs value every minute
 
-/** 
+/**
  * NSFS_GLACIER_RESERVED_BUCKET_TAGS defines an object of bucket tags which will be reserved
  * by the system and PUT operations for them via S3 API would be limited - as in they would be
  * mutable only if specified and only under certain conditions.
@@ -982,7 +985,7 @@ config.NSFS_GLACIER_METRICS_STATFS_INTERVAL = 60 * 1000; // Refresh statfs value
  *  default: any,
  *  event: boolean
  * }>}
- * 
+ *
  * @example
  * {
     'deep-archive-copies': {

src/cmd/manage_nsfs.js

@@ -879,6 +879,9 @@ async function manage_glacier_operations(action, argv) {
         case GLACIER_ACTIONS.EXPIRY:
             await manage_nsfs_glacier.process_expiry();
             break;
+        case GLACIER_ACTIONS.RECLAIM:
+            await manage_nsfs_glacier.process_reclaim();
+            break;
         default:
             throw_cli_error(ManageCLIError.InvalidGlacierOperation);
     }

src/manage_nsfs/manage_nsfs_constants.js

@@ -26,6 +26,7 @@ const GLACIER_ACTIONS = Object.freeze({
     MIGRATE: 'migrate',
     RESTORE: 'restore',
     EXPIRY: 'expiry',
+    RECLAIM: 'reclaim',
 });
 
 const DIAGNOSE_ACTIONS = Object.freeze({
@@ -72,6 +73,7 @@ const VALID_OPTIONS_GLACIER = {
     'migrate': new Set([ CONFIG_ROOT_FLAG]),
     'restore': new Set([ CONFIG_ROOT_FLAG]),
     'expiry': new Set([ CONFIG_ROOT_FLAG]),
+    'reclaim': new Set([ CONFIG_ROOT_FLAG]),
 };
 
 const VALID_OPTIONS_DIAGNOSE = {

src/manage_nsfs/manage_nsfs_glacier.js

@@ -58,6 +58,19 @@ async function process_expiry() {
     }
 }
 
+async function process_reclaim() {
+    const fs_context = native_fs_utils.get_process_fs_context();
+    const backend = Glacier.getBackend();
+
+    if (
+        await backend.low_free_space() ||
+        !(await time_exceeded(fs_context, config.NSFS_GLACIER_RECLAIM_INTERVAL, Glacier.RECLAIM_TIMESTAMP_FILE))
+    ) return;
+
+    await backend.perform(prepare_galcier_fs_context(fs_context), "RECLAIM");
+    const timestamp_file_path = path.join(config.NSFS_GLACIER_LOGS_DIR, Glacier.RECLAIM_TIMESTAMP_FILE);
+    await record_current_time(fs_context, timestamp_file_path);
+}
 
 /**
  * time_exceeded returns true if the time between last run recorded in the given
@@ -129,3 +142,4 @@ function prepare_galcier_fs_context(fs_context) {
 exports.process_migrations = process_migrations;
 exports.process_restores = process_restores;
 exports.process_expiry = process_expiry;
+exports.process_reclaim = process_reclaim;

src/native/fs/fs_napi.cpp

@@ -50,9 +50,11 @@
 #define GPFS_DMAPI_DOT_IBMOBJ_EA "IBMObj"
 #define GPFS_DMAPI_DOT_IBMPMIG_EA "IBMPMig"
 #define GPFS_DMAPI_DOT_IBMTPS_EA "IBMTPS"
+#define GPFS_DMAPI_DOT_IBMUID_EA "IBMUID"
 #define GPFS_DMAPI_XATTR_TAPE_INDICATOR GPFS_DMAPI_XATTR_PREFIX "." GPFS_DMAPI_DOT_IBMOBJ_EA
 #define GPFS_DMAPI_XATTR_TAPE_PREMIG GPFS_DMAPI_XATTR_PREFIX "." GPFS_DMAPI_DOT_IBMPMIG_EA
 #define GPFS_DMAPI_XATTR_TAPE_TPS GPFS_DMAPI_XATTR_PREFIX "." GPFS_DMAPI_DOT_IBMTPS_EA
+#define GPFS_DMAPI_XATTR_TAPE_UID GPFS_DMAPI_XATTR_PREFIX "." GPFS_DMAPI_DOT_IBMUID_EA
 
 // This macro should be used after openning a file
 // it will autoclose the file using AutoCloser and will throw an error in case of failures
@@ -255,6 +257,7 @@ const static std::vector<std::string> GPFS_DMAPI_XATTRS{
     GPFS_DMAPI_XATTR_TAPE_INDICATOR,
     GPFS_DMAPI_XATTR_TAPE_PREMIG,
     GPFS_DMAPI_XATTR_TAPE_TPS,
+    GPFS_DMAPI_XATTR_TAPE_UID,
 };
 const static std::vector<std::string> USER_XATTRS{
     "user.content_type",

src/sdk/glacier.js

@@ -21,6 +21,7 @@ class Glacier {
     static MIGRATE_TIMESTAMP_FILE = 'migrate.timestamp';
     static RESTORE_TIMESTAMP_FILE = 'restore.timestamp';
     static EXPIRY_TIMESTAMP_FILE = 'expiry.timestamp';
+    static RECLAIM_TIMESTAMP_FILE = 'reclaim.timestamp';
 
     /**
      * XATTR_RESTORE_REQUEST is set to a NUMBER (expiry days) by `restore_object` when
@@ -71,10 +72,21 @@ class Glacier {
      */
     static GPFS_DMAPI_XATTR_TAPE_TPS = 'dmapi.IBMTPS';
 
+    /**
+     * GPFS_DMAPI_XATTR_TAPE_UID xattr contains UID which contains the unique ID of the UID
+     * 
+     * Example: `1284427297506873931-5499940123615166566-1799306066-279655-0` (here 279655 is
+     * the inode number)
+     * 
+     * NOTE: If IBMUID EA exists, that means the file is either migrated or premigrated.
+     */
+    static GPFS_DMAPI_XATTR_TAPE_UID = 'dmapi.IBMUID';
+
     static MIGRATE_WAL_NAME = 'migrate';
     static MIGRATE_STAGE_WAL_NAME = 'stage.migrate';
     static RESTORE_WAL_NAME = 'restore';
     static RESTORE_STAGE_WAL_NAME = 'stage.restore';
+    static RECLAIM_WAL_NAME = 'reclaim';
 
     /** @type {nb.RestoreState} */
     static RESTORE_STATUS_CAN_RESTORE = 'CAN_RESTORE';
@@ -86,6 +98,7 @@ class Glacier {
     static GLACIER_CLUSTER_LOCK = 'glacier.cluster.lock';
     static GLACIER_MIGRATE_CLUSTER_LOCK = 'glacier.cluster.migrate.lock';
     static GLACIER_RESTORE_CLUSTER_LOCK = 'glacier.cluster.restore.lock';
+    static GLACIER_RECLAIM_CLUSTER_LOCK = 'glacier.cluster.reclaim.lock';
     static GLACIER_SCAN_LOCK = 'glacier.scan.lock';
 
     /**
@@ -181,6 +194,20 @@ class Glacier {
         throw new Error('Unimplementented');
     }
 
+    /**
+     * reclaim cleans up inindexed items in the underlying
+     * glacier storage
+     * 
+     * NOTE: This needs to be implemented by each backend.
+     * @param {nb.NativeFSContext} fs_context
+     * @param {LogFile} log_file log filename
+     * @param {(entry: string) => Promise<void>} failure_recorder
+     * @returns {Promise<boolean>}
+     */
+    async reclaim(fs_context, log_file, failure_recorder) {
+        throw new Error('Unimplementented');
+    }
+
     /**
      * low_free_space must return true if the backend has
      * low free space.
@@ -199,7 +226,7 @@ class Glacier {
 
     /**
      * @param {nb.NativeFSContext} fs_context 
-     * @param {"MIGRATION" | "RESTORE" | "EXPIRY"} type 
+     * @param {"MIGRATION" | "RESTORE" | "EXPIRY" | "RECLAIM"} type 
      */
     async perform(fs_context, type) {
         const lock_path = lock_file => path.join(config.NSFS_GLACIER_LOGS_DIR, lock_file);
@@ -217,8 +244,8 @@ class Glacier {
          * ) => Promise<boolean>} log_cb */
 
         /**
-         * @param {string} namespace 
-         * @param {log_cb} cb 
+         * @param {string} namespace
+         * @param {log_cb} cb
          */
         const process_glacier_logs = async (namespace, cb) => {
             const logs = new PersistentLogger(
@@ -266,6 +293,10 @@ class Glacier {
                 this.restore.bind(this),
                 Glacier.GLACIER_RESTORE_CLUSTER_LOCK,
             );
+        } else if (type === 'RECLAIM') {
+            await native_fs_utils.lock_and_run(fs_context, lock_path(Glacier.GLACIER_RECLAIM_CLUSTER_LOCK), async () => {
+                await process_glacier_logs(Glacier.RECLAIM_WAL_NAME, this.reclaim.bind(this));
+            });
         }
     }
 

src/sdk/glacier_tapecloud.js

@@ -24,6 +24,7 @@ function get_bin_path(bin_name) {
 class TapeCloudUtils {
     static MIGRATE_SCRIPT = 'migrate';
     static RECALL_SCRIPT = 'recall';
+    static RECLAIM_SCRIPT = 'reclaim';
     static TASK_SHOW_SCRIPT = 'task_show';
     static PROCESS_EXPIRED_SCRIPT = 'process_expired';
     static LOW_FREE_SPACE_SCRIPT = 'low_free_space';
@@ -182,6 +183,29 @@ class TapeCloudUtils {
         }
     }
 
+    /**
+     * reclaim takes name of a file which contains the list
+     * of the files to be reclaimed.
+     *
+     * reclaim doesn't perform any failure handling and expects the
+     * underlying scripts to take care of retries.
+     *
+     * @param {string} file filename
+     * @returns {Promise<boolean>} Indicates success if true
+     */
+    static async reclaim(file) {
+        try {
+            dbg.log1("Starting reclaim for file", file);
+            const out = await exec(`${get_bin_path(TapeCloudUtils.RECLAIM_SCRIPT)} ${file}`, { return_stdout: true });
+            dbg.log4("reclaim finished with:", out);
+            dbg.log1("Finished reclaim for file", file);
+        } catch (error) {
+            dbg.error("Failed to run TapeCloudUtils.reclaim for file:", file, "due to error:", error);
+        }
+
+        return true;
+    }
+
     static async process_expired() {
         dbg.log1("Starting process_expired");
         const out = await exec(`${get_bin_path(TapeCloudUtils.PROCESS_EXPIRED_SCRIPT)}`, { return_stdout: true });
@@ -416,6 +440,21 @@ class TapeCloudGlacier extends Glacier {
         }
     }
 
+    /**
+     *
+     * @param {nb.NativeFSContext} fs_context
+     * @param {LogFile} log_file log filename
+     * @param {(entry: string) => Promise<void>} failure_recorder
+     * @returns {Promise<boolean>}
+     */
+    async reclaim(fs_context, log_file, failure_recorder) {
+        try {
+            return this._process_reclaimed(log_file.log_path);
+        } catch (error) {
+            dbg.error('unexpected error occured while running tapecloud.reclaim:', error);
+        }
+    }
+
     async low_free_space() {
         const result = await exec(get_bin_path(TapeCloudUtils.LOW_FREE_SPACE_SCRIPT), { return_stdout: true });
         return result.toLowerCase().trim() === 'true';
@@ -457,6 +496,17 @@ class TapeCloudGlacier extends Glacier {
         return TapeCloudUtils.process_expired();
     }
 
+    /**
+     * _process_reclaimed should perform reclaimed
+     *
+     * NOTE: Must be overwritten for tests
+     * @param {string} file
+	 * @returns {Promise<boolean>}
+     */
+    async _process_reclaimed(file) {
+        return TapeCloudUtils.reclaim(file);
+    }
+
     /**
      * finalizes the restore by setting the required EAs
      *

src/sdk/namespace_fs.js

@@ -1405,6 +1405,8 @@ class NamespaceFS {
         const is_disabled_dir_content = this._is_directory_content(file_path, params.key) && this._is_versioning_disabled();
 
         const stat = await target_file.stat(fs_context);
+        const file_path_stat = config.NSFS_GLACIER_DMAPI_ENABLE_TAPE_RECLAIM &&
+            await nb_native().fs.stat(fs_context, file_path).catch(_.noop);
         this._verify_encryption(params.encryption, this._get_encryption_info(stat));
 
         const copy_xattr = params.copy_source && params.xattr_copy;
@@ -1453,6 +1455,11 @@ class NamespaceFS {
         dbg.log1('NamespaceFS._finish_upload:', open_mode, file_path, upload_path, fs_xattr);
 
         if (!same_inode && !part_upload) {
+            // If the target file is already on tape then this is a candidate for tape reclaim
+            if (file_path_stat && file_path_stat.xattr[Glacier.GPFS_DMAPI_XATTR_TAPE_INDICATOR]) {
+                await this.append_to_reclaim_wal(fs_context, upload_path, file_path_stat);
+            }
+
             await this._move_to_dest(fs_context, upload_path, file_path, target_file, open_mode, params.key);
         }
 
@@ -2124,7 +2131,16 @@ class NamespaceFS {
                 if (files) await this._close_files(fs_context, files.delete_version, undefined, true);
             }
         } else {
-            await native_fs_utils.unlink_ignore_enoent(fs_context, file_path);
+            try {
+                const stat = config.NSFS_GLACIER_DMAPI_ENABLE_TAPE_RECLAIM &&
+                    await nb_native().fs.stat(fs_context, file_path).catch(dbg.warn.bind(this));
+                await nb_native().fs.unlink(fs_context, file_path);
+                if (stat) {
+                    await this.append_to_reclaim_wal(fs_context, file_path, stat);
+                }
+            } catch (err) {
+                if (err.code !== 'ENOENT' && err.code !== 'EISDIR') throw err;
+            }
         }
 
         await this._delete_path_dirs(file_path, fs_context);
@@ -3709,6 +3725,28 @@ class NamespaceFS {
         await NamespaceFS.restore_wal.append(entry);
     }
 
+    /**
+     *
+     * @param {nb.NativeFSContext} fs_context
+     * @param {string} file_path
+     * @param {nb.NativeFSStats} [stat]
+     * @returns
+     */
+    async append_to_reclaim_wal(fs_context, file_path, stat) {
+        if (!config.NSFS_GLACIER_LOGS_ENABLED || !config.NSFS_GLACIER_DMAPI_ENABLE_TAPE_RECLAIM) return;
+
+        if (!stat) {
+            stat = await nb_native().fs.stat(fs_context, file_path);
+        }
+
+        const data = JSON.stringify({
+            full_path: file_path,
+            logical_size: stat.size,
+            ea: stat.xattr,
+        });
+        await NamespaceFS.reclaim_wal.append(data);
+    }
+
     static get migrate_wal() {
         if (!NamespaceFS._migrate_wal) {
             NamespaceFS._migrate_wal = new PersistentLogger(config.NSFS_GLACIER_LOGS_DIR, Glacier.MIGRATE_WAL_NAME, {
@@ -3731,6 +3769,17 @@ class NamespaceFS {
         return NamespaceFS._restore_wal;
     }
 
+    static get reclaim_wal() {
+        if (!NamespaceFS._reclaim_wal) {
+            NamespaceFS._reclaim_wal = new PersistentLogger(config.NSFS_GLACIER_LOGS_DIR, Glacier.RECLAIM_WAL_NAME, {
+                poll_interval: config.NSFS_GLACIER_LOGS_POLL_INTERVAL,
+                locking: 'SHARED',
+            });
+        }
+
+        return NamespaceFS._reclaim_wal;
+    }
+
     ////////////////////////////
     //    LIFECYLE HELPERS    //
     ////////////////////////////
@@ -3757,6 +3806,9 @@ class NamespaceFS {
             this._check_lifecycle_filter_before_deletion(params, stat);
             const bucket_tmp_dir_path = this.get_bucket_tmpdir_full_path();
             await native_fs_utils.safe_unlink(fs_context, file_path, stat, { dir_file, src_file }, bucket_tmp_dir_path);
+            if (!is_empty_directory_content) {
+                await this.append_to_reclaim_wal(fs_context, file_path, src_stat).catch(dbg.warn.bind(this));
+            }
         } catch (err) {
             dbg.log0('_verify_lifecycle_filter_and_unlink err', err.code, err, file_path);
             if (err.code !== 'ENOENT' && err.code !== 'EISDIR') throw err;
@@ -3803,7 +3855,8 @@ NamespaceFS._migrate_wal = null;
 /** @type {PersistentLogger} */
 NamespaceFS._restore_wal = null;
 
+/** @type {PersistentLogger} */
+NamespaceFS._reclaim_wal = null;
+
 module.exports = NamespaceFS;
 module.exports.multi_buffer_pool = multi_buffer_pool;
-
-