rgw: data/bilogs are trimmed when no peers are reading them

if no peers zones are configured to sync from us, we can safely trim
our entire log. start with the marker '99999999' (which compares larger
than any key based on a timestamp) instead of an empty marker, then take
the minimum of that with any peer markers

Fixes: http://tracker.ceph.com/issues/39487
Signed-off-by: Casey Bodley <cbodley@redhat.com>
(cherry picked from commit da3803f59fcf904ec32e6e03d2dedc23d0a127f8)

Conflicts:
	src/rgw/rgw_trim_datalog.cc
- file does not exist in mimic; made the changes manually in
  src/rgw/rgw_data_sync.cc

diff --git a/src/rgw/rgw_data_sync.cc b/src/rgw/rgw_data_sync.cc
index 07197507b722d175697ef0c16c1ab53c65b4942a..24dc908e47d9f360e4fb1dfc24d06ed4cd0c4d6b 100644 (file)
--- a/src/rgw/rgw_data_sync.cc
+++ b/src/rgw/rgw_data_sync.cc
@@ -3563,7 +3563,7 @@ class DataLogTrimCR : public RGWCoroutine {
       num_shards(num_shards),
       zone_id(store->get_zone().id),
       peer_status(store->zone_conn_map.size()),
-      min_shard_markers(num_shards),
+      min_shard_markers(num_shards, "99999999"),
       last_trim(last_trim)
   {}
 

diff --git a/src/rgw/rgw_sync_log_trim.cc b/src/rgw/rgw_sync_log_trim.cc
index f1e56c4262fec683f81a9314b04f1bf67b02f774..7d3820af460fe0fd447dbe0d7d3ac334a25de6f5 100644 (file)
--- a/src/rgw/rgw_sync_log_trim.cc
+++ b/src/rgw/rgw_sync_log_trim.cc
@@ -472,7 +472,9 @@ int BucketTrimInstanceCR::operate()
       }
     }
 
-    min_markers.resize(std::max(1u, bucket_info.num_shards));
+    // initialize each shard with the maximum marker, which is only used when
+    // there are no peers syncing from us
+    min_markers.assign(std::max(1u, bucket_info.num_shards), "99999999");
 
     // determine the minimum marker for each shard
     retcode = take_min_status(cct, peer_status.begin(), peer_status.end(),