osd: give op threads hdd/ssd specific defaults

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

diff --git a/src/common/config_opts.h b/src/common/config_opts.h
index 0a67ef5ea99f6e9179dfba75339bd06eb1ffffba..7e303707ae9f6e36d962b839f20bb88d3d2900fd 100644 (file)
--- a/src/common/config_opts.h
+++ b/src/common/config_opts.h
@@ -757,8 +757,12 @@ OPTION(osd_disk_threads, OPT_INT, 1)
 OPTION(osd_disk_thread_ioprio_class, OPT_STR, "") // rt realtime be best effort idle
 OPTION(osd_disk_thread_ioprio_priority, OPT_INT, -1) // 0-7
 OPTION(osd_recover_clone_overlap, OPT_BOOL, true)   // preserve clone_overlap during recovery/migration
-OPTION(osd_op_num_threads_per_shard, OPT_INT, 2)
-OPTION(osd_op_num_shards, OPT_INT, 5)
+OPTION(osd_op_num_threads_per_shard, OPT_INT, 0)
+OPTION(osd_op_num_threads_per_shard_hdd, OPT_INT, 1)
+OPTION(osd_op_num_threads_per_shard_ssd, OPT_INT, 2)
+OPTION(osd_op_num_shards, OPT_INT, 0)
+OPTION(osd_op_num_shards_hdd, OPT_INT, 5)
+OPTION(osd_op_num_shards_ssd, OPT_INT, 8)
 OPTION(osd_op_queue, OPT_STR, "wpq") // PrioritzedQueue (prio), Weighted Priority Queue (wpq), or debug_random
 OPTION(osd_op_queue_cut_off, OPT_STR, "low") // Min priority to go to strict queue. (low, high, debug_random)
 

diff --git a/src/osd/OSD.cc b/src/osd/OSD.cc
index 3babbe3f0114e0e3a7bb8ff3d7d8541d66f9d1b5..0558ca7d48215fc4d8b4128461e2d430f5cc4cd8 100644 (file)
--- a/src/osd/OSD.cc
+++ b/src/osd/OSD.cc
@@ -1706,7 +1706,7 @@ int OSD::mkfs(CephContext *cct, ObjectStore *store, const string &dev,
     goto free_store;
   }
 
-  store->set_cache_shards(cct->_conf->osd_op_num_shards);
+  store->set_cache_shards(1);  // doesn't matter for mkfs!
 
   ret = store->mount();
   if (ret) {
@@ -1874,6 +1874,7 @@ OSD::OSD(CephContext *cct_, ObjectStore *store_,
   clog(log_client.create_channel()),
   whoami(id),
   dev_path(dev), journal_path(jdev),
+  store_is_rotational(store->is_rotational()),
   trace_endpoint("0.0.0.0", 0, "osd"),
   asok_hook(NULL),
   osd_compat(get_osd_compat_set()),
@@ -1881,7 +1882,7 @@ OSD::OSD(CephContext *cct_, ObjectStore *store_,
             cct->_conf->osd_peering_wq_threads,
             "osd_peering_tp_threads"),
   osd_op_tp(cct, "OSD::osd_op_tp", "tp_osd_tp",
-    cct->_conf->osd_op_num_threads_per_shard * cct->_conf->osd_op_num_shards),
+           get_num_op_threads()),
   disk_tp(cct, "OSD::disk_tp", "tp_osd_disk", cct->_conf->osd_disk_threads, "osd_disk_threads"),
   command_tp(cct, "OSD::command_tp", "tp_osd_cmd",  1),
   session_waiting_lock("OSD::session_waiting_lock"),
@@ -1901,7 +1902,7 @@ OSD::OSD(CephContext *cct_, ObjectStore *store_,
   op_queue(get_io_queue()),
   op_prio_cutoff(get_io_prio_cut()),
   op_shardedwq(
-    cct->_conf->osd_op_num_shards,
+    get_num_op_shards(),
     this,
     cct->_conf->osd_op_thread_timeout,
     cct->_conf->osd_op_thread_suicide_timeout,
@@ -2278,6 +2279,26 @@ int OSD::enable_disable_fuse(bool stop)
   return 0;
 }
 
+int OSD::get_num_op_shards()
+{
+  if (cct->_conf->osd_op_num_shards)
+    return cct->_conf->osd_op_num_shards;
+  if (store_is_rotational)
+    return cct->_conf->osd_op_num_shards_hdd;
+  else
+    return cct->_conf->osd_op_num_shards_ssd;
+}
+
+int OSD::get_num_op_threads()
+{
+  if (cct->_conf->osd_op_num_threads_per_shard)
+    return get_num_op_shards() * cct->_conf->osd_op_num_threads_per_shard;
+  if (store_is_rotational)
+    return get_num_op_shards() * cct->_conf->osd_op_num_threads_per_shard_hdd;
+  else
+    return get_num_op_shards() * cct->_conf->osd_op_num_threads_per_shard_ssd;
+}
+
 int OSD::init()
 {
   CompatSet initial, diff;
@@ -2291,11 +2312,12 @@ int OSD::init()
   service.recovery_sleep_timer.init();
 
   // mount.
-  dout(2) << "mounting " << dev_path << " "
-         << (journal_path.empty() ? "(no journal)" : journal_path) << dendl;
+  dout(2) << "init " << dev_path
+         << " (looks like " << (store_is_rotational ? "hdd" : "ssd") << ")"
+         << dendl;
   assert(store);  // call pre_init() first!
 
-  store->set_cache_shards(cct->_conf->osd_op_num_shards);
+  store->set_cache_shards(get_num_op_shards());
 
   int r = store->mount();
   if (r < 0) {

diff --git a/src/osd/OSD.h b/src/osd/OSD.h
index 49f17fdf413519c84fcf9b12c7a298f5bc54295a..889450d25a01ddeba0ed3f035fe73e164b2d2141 100644 (file)
--- a/src/osd/OSD.h
+++ b/src/osd/OSD.h
@@ -1312,6 +1312,8 @@ protected:
   int whoami;
   std::string dev_path, journal_path;
 
+  bool store_is_rotational = true;
+
   ZTracer::Endpoint trace_endpoint;
   void create_logger();
   void create_recoverystate_perf();
@@ -2457,6 +2459,9 @@ private:
 
   int init_op_flags(OpRequestRef& op);
 
+  int get_num_op_shards();
+  int get_num_op_threads();
+
 public:
   static int peek_meta(ObjectStore *store, string& magic,
                       uuid_d& cluster_fsid, uuid_d& osd_fsid, int& whoami);