From: Sage Weil <sage@redhat.com>
Date: Thu, 17 Nov 2016 22:45:53 +0000 (-0600)
Subject: osd/ReplicatedPG: limit omap request by bytes
X-Git-Tag: v11.1.0~195^2~1
X-Git-Url: http://git-server-git.apps.pok.os.sepia.ceph.com/?a=commitdiff_plain;h=41d46e4921c6cb7fc792c9c1eb3da51f0554b163;p=ceph.git

osd/ReplicatedPG: limit omap request by bytes

Signed-off-by: Sage Weil <sage@redhat.com>
---

diff --git a/src/common/config_opts.h b/src/common/config_opts.h
index d8936bfde4a1..448b401290c7 100644
--- a/src/common/config_opts.h
+++ b/src/common/config_opts.h
@@ -930,6 +930,7 @@ OPTION(osd_max_attr_name_len, OPT_U32, 100)    // max rados attr name len; canno
 OPTION(osd_max_attr_size, OPT_U64, 0)
 
 OPTION(osd_max_omap_entries_per_request, OPT_U64, 131072)
+OPTION(osd_max_omap_bytes_per_request, OPT_U64, 4<<20)
 
 OPTION(osd_objectstore, OPT_STR, "filestore")  // ObjectStore backend type
 OPTION(osd_objectstore_tracing, OPT_BOOL, false) // true if LTTng-UST tracepoints should be enabled
diff --git a/src/osd/ReplicatedPG.cc b/src/osd/ReplicatedPG.cc
index 7a0c351bf6da..0586f8a79e5c 100644
--- a/src/osd/ReplicatedPG.cc
+++ b/src/osd/ReplicatedPG.cc
@@ -5826,21 +5826,25 @@ int ReplicatedPG::do_osd_ops(OpContext *ctx, vector<OSDOp>& ops)
 	  max_return = g_conf->osd_max_omap_entries_per_request;
 	}
 	tracepoint(osd, do_osd_op_pre_omapgetkeys, soid.oid.name.c_str(), soid.snap.val, start_after.c_str(), max_return);
-	set<string> out_set;
 
+	bufferlist bl;
+	uint32_t num = 0;
 	if (oi.is_omap()) {
 	  ObjectMap::ObjectMapIterator iter = osd->store->get_omap_iterator(
 	    coll, ghobject_t(soid)
 	    );
 	  assert(iter);
 	  iter->upper_bound(start_after);
-	  for (uint64_t i = 0;
-	       i < max_return && iter->valid();
-	       ++i, iter->next(false)) {
-	    out_set.insert(iter->key());
+	  for (num = 0;
+	       num < max_return &&
+		 bl.length() < g_conf->osd_max_omap_bytes_per_request &&
+		 iter->valid();
+	       ++num, iter->next(false)) {
+	    ::encode(iter->key(), bl);
 	  }
 	} // else return empty out_set
-	::encode(out_set, osd_op.outdata);
+	::encode(num, osd_op.outdata);
+	osd_op.outdata.claim_append(bl);
 	ctx->delta_stats.num_rd_kb += SHIFT_ROUND_UP(osd_op.outdata.length(), 10);
 	ctx->delta_stats.num_rd++;
       }
@@ -5866,8 +5870,9 @@ int ReplicatedPG::do_osd_ops(OpContext *ctx, vector<OSDOp>& ops)
 	  max_return = g_conf->osd_max_omap_entries_per_request;
 	}
 	tracepoint(osd, do_osd_op_pre_omapgetvals, soid.oid.name.c_str(), soid.snap.val, start_after.c_str(), max_return, filter_prefix.c_str());
-	map<string, bufferlist> out_set;
 
+	uint32_t num = 0;
+	bufferlist bl;
 	if (oi.is_omap()) {
 	  ObjectMap::ObjectMapIterator iter = osd->store->get_omap_iterator(
 	    coll, ghobject_t(soid)
@@ -5878,15 +5883,19 @@ int ReplicatedPG::do_osd_ops(OpContext *ctx, vector<OSDOp>& ops)
           }
 	  iter->upper_bound(start_after);
 	  if (filter_prefix > start_after) iter->lower_bound(filter_prefix);
-	  for (uint64_t i = 0;
-	       i < max_return && iter->valid() &&
+	  for (num = 0;
+	       num < max_return &&
+		 bl.length() < g_conf->osd_max_omap_bytes_per_request &&
+		 iter->valid() &&
 		 iter->key().substr(0, filter_prefix.size()) == filter_prefix;
-	       ++i, iter->next(false)) {
+	       ++num, iter->next(false)) {
 	    dout(20) << "Found key " << iter->key() << dendl;
-	    out_set.insert(make_pair(iter->key(), iter->value()));
+	    ::encode(iter->key(), bl);
+	    ::encode(iter->value(), bl);
 	  }
 	} // else return empty out_set
-	::encode(out_set, osd_op.outdata);
+	::encode(num, osd_op.outdata);
+	osd_op.outdata.claim_append(bl);
 	ctx->delta_stats.num_rd_kb += SHIFT_ROUND_UP(osd_op.outdata.length(), 10);
 	ctx->delta_stats.num_rd++;
       }