From 3d3ecb2a1eef8fd900a56034ce90d4b90ba1e633 Mon Sep 17 00:00:00 2001
From: Shilpa Jagannath <smanjara@redhat.com>
Date: Tue, 14 Jul 2020 17:56:05 +0530
Subject: [PATCH] rgw: failing to reshard, restore old indexes

Signed-off-by: Shilpa Jagannath <smanjara@redhat.com>
---
 src/rgw/rgw_reshard.cc | 41 ++++++++++++++++++++++++++++-------------
 src/rgw/rgw_reshard.h  |  1 +
 2 files changed, 29 insertions(+), 13 deletions(-)

diff --git a/src/rgw/rgw_reshard.cc b/src/rgw/rgw_reshard.cc
index 5c1e28018a0..a809d791b21 100644
--- a/src/rgw/rgw_reshard.cc
+++ b/src/rgw/rgw_reshard.cc
@@ -659,7 +659,6 @@ int RGWBucketReshard::do_reshard(int num_shards,
   }
 
   //overwrite current_index for the next reshard process
-  const auto prev_index = bucket_info.layout.current_index;
   bucket_info.layout.current_index = *bucket_info.layout.target_index;
   bucket_info.layout.target_index = std::nullopt; // target_layout doesn't need to exist after reshard
   bucket_info.layout.resharding = rgw::BucketReshardState::None;
@@ -674,18 +673,6 @@ int RGWBucketReshard::do_reshard(int num_shards,
       return ret;
   }
 
-  // resharding successful, so remove old bucket index shards; use
-  // best effort and don't report out an error; the lock isn't needed
-  // at this point since all we're using a best effor to to remove old
-  // shard objects
-
-  ret = store->svc()->bi->clean_index(dpp, bucket_info, prev_index);
-  if (ret < 0) {
-    ldpp_dout(dpp, -1) << "Error: " << __func__ <<
-    " failed to clean up old shards; " <<
-    "RGWRados::clean_bucket_index returned " << ret << dendl;
-  }
-
   return 0;
   // NB: some error clean-up is done by ~BucketInfoReshardUpdate
 } // RGWBucketReshard::do_reshard
@@ -719,6 +706,9 @@ int RGWBucketReshard::execute(int num_shards, int max_op_entries,
     }
   }
 
+  // keep a copy of old index layout
+  prev_index = bucket_info.layout.current_index;
+
   ret = do_reshard(num_shards,
 		   max_op_entries,
                    verbose, out, formatter, dpp);
@@ -731,6 +721,18 @@ int RGWBucketReshard::execute(int num_shards, int max_op_entries,
 
   reshard_lock.unlock();
 
+  // resharding successful, so remove old bucket index shards; use
+  // best effort and don't report out an error; the lock isn't needed
+  // at this point since all we're using a best effor to to remove old
+  // shard objects
+
+  ret = store->svc()->bi->clean_index(dpp, bucket_info, prev_index);
+  if (ret < 0) {
+    ldpp_dout(dpp, -1) << __func__ << "Error: " << __func__ <<
+    " failed to clean up old shards; " <<
+    "RGWRados::clean_bucket_index returned " << ret << dendl;
+  }
+
   ldpp_dout(dpp, 1) << __func__ <<
     " INFO: reshard of bucket \"" << bucket_info.bucket.name << "\" completed successfully" << dendl;
 
@@ -752,6 +754,19 @@ error_out:
       "RGWRados::clean_bucket_index returned " << ret2 << dendl;
   }
 
+  // restore old index if reshard fails
+  bucket_info.layout.current_index = prev_index;
+  ret = store->getRados()->put_bucket_instance_info(bucket_info, false, real_time(), &bucket_attrs, dpp);
+  if (ret < 0) {
+    lderr(store->ctx()) << "ERROR: failed writing bucket instance info: " << dendl;
+      return ret;
+  }
+
+  ret = store->svc()->bi->init_index(dpp, bucket_info, bucket_info.layout.current_index);
+  if (ret < 0) {
+      return ret;
+  }
+  
   return ret;
 } // execute
 
diff --git a/src/rgw/rgw_reshard.h b/src/rgw/rgw_reshard.h
index c75dd169f87..ed1325d5adf 100644
--- a/src/rgw/rgw_reshard.h
+++ b/src/rgw/rgw_reshard.h
@@ -76,6 +76,7 @@ private:
   rgw::sal::RadosStore* store;
   RGWBucketInfo bucket_info;
   std::map<std::string, bufferlist> bucket_attrs;
+  rgw::bucket_index_layout_generation prev_index;
 
   RGWBucketReshardLock reshard_lock;
   RGWBucketReshardLock* outer_reshard_lock;