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 moved from rgw_data_sync.cc

diff --git a/src/rgw/rgw_data_sync.cc b/src/rgw/rgw_data_sync.cc
index bd6c4b50a47fea89cf743de434d531c32ef8b4a3..e24b1206c498766e0997a2f6c30e0d6122e35c07 100644 (file)
--- a/src/rgw/rgw_data_sync.cc
+++ b/src/rgw/rgw_data_sync.cc
@@ -3584,7 +3584,7 @@ class DataLogTrimCR : public RGWCoroutine {
       num_shards(num_shards),
       zone_id(store->svc.zone->get_zone().id),
       peer_status(store->svc.zone->get_zone_data_notify_to_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 243fa65d0f6a4c27ebf0aba0293b58665a4626e2..d6d53cd535d5262c7f88d9fcd3038e5ad4b5ade3 100644 (file)
--- a/src/rgw/rgw_sync_log_trim.cc
+++ b/src/rgw/rgw_sync_log_trim.cc
@@ -477,7 +477,9 @@ 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(),