From c33b23b5b4f3f74e00354ba964d35c903e9a086f Mon Sep 17 00:00:00 2001
From: Casey Bodley <cbodley@redhat.com>
Date: Tue, 28 Sep 2021 10:45:27 -0400
Subject: [PATCH] rgw: prevent reshard from creating too many log generations

Signed-off-by: Casey Bodley <cbodley@redhat.com>
---
 src/rgw/rgw_admin.cc   | 11 +++++++++++
 src/rgw/rgw_reshard.cc | 14 ++++++++++++++
 src/rgw/rgw_reshard.h  |  8 ++++++++
 3 files changed, 33 insertions(+)

diff --git a/src/rgw/rgw_admin.cc b/src/rgw/rgw_admin.cc
index 4994f67d143d7..07812f43f20ca 100644
--- a/src/rgw/rgw_admin.cc
+++ b/src/rgw/rgw_admin.cc
@@ -7673,6 +7673,17 @@ next:
       return ret;
     }
 
+    auto zone_svc = static_cast<rgw::sal::RadosStore*>(store)->svc()->zone;
+    if (!RGWBucketReshard::can_reshard(bucket->get_info(), zone_svc) &&
+        !yes_i_really_mean_it) {
+      std::cerr << "Bucket '" << bucket->get_name() << "' already has too many "
+          "log generations (" << bucket->get_info().layout.logs.size() << ") "
+          "from previous reshards that peer zones haven't finished syncing. "
+          "Resharding is not recommended until the old generations sync, but "
+          "you can force a reshard with --yes-i-really-mean-it." << std::endl;
+      return -EINVAL;
+    }
+
     RGWBucketReshard br(static_cast<rgw::sal::RadosStore*>(store), bucket->get_info(), nullptr /* no callback */);
 
 #define DEFAULT_RESHARD_MAX_ENTRIES 1000
diff --git a/src/rgw/rgw_reshard.cc b/src/rgw/rgw_reshard.cc
index 672b4b9c8e2c1..d2de54271a0fe 100644
--- a/src/rgw/rgw_reshard.cc
+++ b/src/rgw/rgw_reshard.cc
@@ -816,6 +816,13 @@ int RGWBucketReshard::execute(int num_shards,
   return 0;
 } // execute
 
+bool RGWBucketReshard::can_reshard(const RGWBucketInfo& bucket,
+                                   const RGWSI_Zone* zone_svc)
+{
+  return !zone_svc->need_to_log_data() ||
+      bucket.layout.logs.size() < max_bilog_history;
+}
+
 
 RGWReshard::RGWReshard(rgw::sal::RadosStore* _store, bool _verbose, ostream *_out,
                        Formatter *_formatter) :
@@ -1052,6 +1059,13 @@ int RGWReshard::process_entry(const cls_rgw_reshard_entry& entry,
     return 0;
   }
 
+  if (!RGWBucketReshard::can_reshard(bucket_info, store->svc()->zone)) {
+    ldpp_dout(dpp, 1) << "Bucket " << bucket_info.bucket << " is not "
+        "eligible for resharding until peer zones finish syncing one "
+        "or more of its old log generations" << dendl;
+    return remove(dpp, entry);
+  }
+
   RGWBucketReshard br(store, bucket_info, nullptr);
 
   ReshardFaultInjector f; // no fault injected
diff --git a/src/rgw/rgw_reshard.h b/src/rgw/rgw_reshard.h
index a8613d66467a3..09c89bb973006 100644
--- a/src/rgw/rgw_reshard.h
+++ b/src/rgw/rgw_reshard.h
@@ -163,6 +163,14 @@ public:
 
     return final_num_shards;
   }
+
+  // for multisite, the RGWBucketInfo keeps a history of old log generations
+  // until all peers are done with them. prevent this log history from growing
+  // too large by refusing to reshard the bucket until the old logs get trimmed
+  static constexpr size_t max_bilog_history = 4;
+
+  static bool can_reshard(const RGWBucketInfo& bucket,
+                          const RGWSI_Zone* zone_svc);
 }; // RGWBucketReshard
 
 
-- 
2.39.5