Merge pull request ceph#51071 from ivancich/wip-versioned-reindex

rgw: reindex of objects in versioned buckets

Reviewed-by: Casey Bodley <[email protected]>

Author: ivancich

ceph.spec.in

@@ -1704,6 +1704,7 @@ exit 0
 %{_mandir}/man8/rbd-replay-many.8*
 %{_mandir}/man8/rbd-replay-prep.8*
 %{_mandir}/man8/rgw-orphan-list.8*
+%{_mandir}/man8/rgw-restore-bucket-index.8*
 %dir %{_datadir}/ceph/
 %{_datadir}/ceph/known_hosts_drop.ceph.com
 %{_datadir}/ceph/id_rsa_drop.ceph.com

debian/radosgw.install

@@ -7,3 +7,4 @@ usr/bin/radosgw-token
 usr/share/man/man8/ceph-diff-sorted.8
 usr/share/man/man8/radosgw.8
 usr/share/man/man8/rgw-orphan-list.8
+usr/share/man/man8/rgw-restore-bucket-index.8

doc/man/8/CMakeLists.txt

@@ -60,7 +60,8 @@ if(WITH_RADOSGW)
 	radosgw-admin.rst
 	rgw-orphan-list.rst
 	rgw-policy-check.rst
-	ceph-diff-sorted.rst)
+	ceph-diff-sorted.rst
+	rgw-restore-bucket-index.rst)
 endif()
 
 if(WITH_RBD)

doc/man/8/radosgw-admin.rst

@@ -146,6 +146,9 @@ as follows:
 :command:`object rewrite`
   Rewrite the specified object.
 
+:command:`object reindex`
+  Add an object to its bucket's index. Used rarely for emergency repairs.
+
 :command:`objects expire`
   Run expired objects cleanup.
 

doc/man/8/rgw-restore-bucket-index.rst

@@ -0,0 +1,106 @@
+:orphan:
+
+==================================================================================
+ rgw-restore-bucket-index -- try to restore a bucket's objects to its bucket index
+==================================================================================
+
+.. program:: rgw-restore-bucket-index
+
+Synopsis
+========
+
+| **rgw-restore-bucket-index**
+
+Description
+===========
+
+:program:`rgw-restore-bucket-index` is an *EXPERIMENTAL* RADOS gateway
+user administration utility. It scans the data pool for objects that
+belong to a given bucket and tries to add those objects back to the
+bucket index. It's intended as a **last resort** after a
+**catastrophic** loss of a bucket index. Please thorougly review the
+*Warnings* listed below.
+
+The utility works with regular (i.e., un-versioned) buckets, versioned
+buckets, and buckets were versioning has been suspended.
+
+Warnings
+========
+
+This utility is currently considered *EXPERIMENTAL*.
+
+The results are unpredictable if the bucket is in
+active use while this utility is running.
+
+The results are unpredictable if only some bucket's objects are
+missing from the bucket index. In such a case, consider using the
+"object reindex" subcommand of `radosgw-admin` to restore object's to
+the bucket index one-by-one.
+
+For objects in versioned buckets, if the latest version is a delete
+marker, it will be restored. If a delete marker has been written over
+with a new version, then that delete marker will not be restored. This
+should have minimal impact on results in that the it recovers the
+latest version and previous versions are all accessible.
+
+Command-Line Arguments
+======================
+
+.. option:: -b <bucket>
+
+   Specify the bucket to be reindexed.
+
+.. option:: -p <pool>
+
+   Optional, specify the data pool containing head objects for the
+   bucket. If omitted the utility will try to determine the data pool
+   on its own.
+
+.. option:: -r <realm-name>
+
+   Optional, specify the realm if the restoration is not being applied
+   to the default realm.
+
+.. option:: -g <zonegroup-name>
+
+   Optional, specify the zonegroup if the restoration is not being applied
+   to the default zonegroup.
+
+.. option:: -z <zone-name>
+
+   Optional, specify the zone if the restoration is not being applied
+   to the default zone.
+
+.. option:: -l <rados-ls-output-file>
+
+   Optional, specify a file containing the output of a rados listing
+   of the data pool. Since listing the data pool can be an expensive
+   and time-consuming operation, if trying to recover the indices for
+   multiple buckets, it could be more efficient to re-use the same
+   listing.
+
+.. option:: -y
+
+   Optional, proceed without further prompting. Without this option
+   the utility will display some information and prompt the user as to
+   whether to proceed. When provided, the utility will simply
+   proceed. Please use caution when using this option.
+
+Examples
+========
+
+Attempt to restore the index for a bucket named *summer-2023-photos*::
+
+        $ rgw-restore-bucket-index -b summer-2023-photos
+
+Availability
+============
+
+:program:`rgw-restore-bucket-index` is part of Ceph, a massively
+scalable, open-source, distributed storage system.  Please refer to
+the Ceph documentation at https://docs.ceph.com for more information.
+
+See also
+========
+
+:doc:`radosgw-admin <radosgw-admin>`\(8)

doc/man_index.rst

@@ -48,3 +48,4 @@
    man/8/ceph-immutable-object-cache
    man/8/ceph-diff-sorted
    man/8/rgw-policy-check
+   man/8/rgw-restore-bucket-index

src/cls/rgw/cls_rgw_client.cc

@@ -286,7 +286,8 @@ void cls_rgw_bucket_complete_op(ObjectWriteOperation& o, RGWModifyOp op, const s
                                 const rgw_bucket_dir_entry_meta& dir_meta,
 				const list<cls_rgw_obj_key> *remove_objs, bool log_op,
                                 uint16_t bilog_flags,
-                                const rgw_zone_set *zones_trace)
+                                const rgw_zone_set *zones_trace,
+				const std::string& obj_locator)
 {
 
   bufferlist in;
@@ -295,6 +296,7 @@ void cls_rgw_bucket_complete_op(ObjectWriteOperation& o, RGWModifyOp op, const s
   call.tag = tag;
   call.key = key;
   call.ver = ver;
+  call.locator = obj_locator;
   call.meta = dir_meta;
   call.log_op = log_op;
   call.bilog_flags = bilog_flags;

src/cls/rgw/cls_rgw_client.h

@@ -355,7 +355,8 @@ void cls_rgw_bucket_complete_op(librados::ObjectWriteOperation& o, RGWModifyOp o
                                 const cls_rgw_obj_key& key,
                                 const rgw_bucket_dir_entry_meta& dir_meta,
 				const std::list<cls_rgw_obj_key> *remove_objs, bool log_op,
-                                uint16_t bilog_op, const rgw_zone_set *zones_trace);
+                                uint16_t bilog_op, const rgw_zone_set *zones_trace,
+				const std::string& obj_locator = ""); // ignored if it's the empty string
 
 void cls_rgw_remove_obj(librados::ObjectWriteOperation& o, std::list<std::string>& keep_attr_prefixes);
 void cls_rgw_obj_store_pg_ver(librados::ObjectWriteOperation& o, const std::string& attr);