From: Patrick Donnelly <pdonnell@ibm.com>
Date: Fri, 27 Jun 2025 18:38:17 +0000 (-0400)
Subject: mds: allow disabling batch ops
X-Git-Tag: testing/wip-vshankar-testing-20250721.082855-tentacle-debug~2^2~2
X-Git-Url: http://git.apps.os.sepia.ceph.com/?a=commitdiff_plain;h=dcc9aa2414aab9a50f58231f83d574af33b6e83c;p=ceph-ci.git

mds: allow disabling batch ops

To address a bug and future ones where batching lookup/getattr does not help
"kick" the MDS in switching state more quickly (e.g. flushing the MDS journal).

Signed-off-by: Patrick Donnelly <pdonnell@ibm.com>
(cherry picked from commit 0201f86e6939a3d787bea755a48cb4b4254d2f9c)
---

diff --git a/src/common/options/mds.yaml.in b/src/common/options/mds.yaml.in
index 06c12ea671d..80be6be4d6d 100644
--- a/src/common/options/mds.yaml.in
+++ b/src/common/options/mds.yaml.in
@@ -793,6 +793,21 @@ options:
   services:
   - mds
   with_legacy: true
+- name: mds_allow_batched_ops
+  type: bool
+  level: advanced
+  desc: allow MDS to batch lookup/getattr RPCs
+  long_desc: >
+    The MDS will batch a lookup or getattr RPC on the same inode when
+    possible to avoid repetitive locks on metadata and to bypass other
+    requests acquiring write locks. Generally, this should only
+    improve performance but this switch exists to provide a means to
+    turn this behavior off for comparison.
+  default: true
+  services:
+  - mds
+  flags:
+  - runtime
 # multiple of size_max that triggers immediate split
 - name: mds_bal_fragment_fast_factor
   type: float
diff --git a/src/mds/MDSRank.cc b/src/mds/MDSRank.cc
index 6f4adb81be4..59f695a8c04 100644
--- a/src/mds/MDSRank.cc
+++ b/src/mds/MDSRank.cc
@@ -4070,6 +4070,7 @@ std::vector<std::string> MDSRankDispatcher::get_tracked_keys()
     "fsid",
     "host",
     "mds_allow_async_dirops",
+    "mds_allow_batched_ops",
     "mds_alternate_name_max",
     "mds_bal_export_pin",
     "mds_bal_fragment_dirs",
diff --git a/src/mds/Server.cc b/src/mds/Server.cc
index 6586defce91..bf353f37122 100644
--- a/src/mds/Server.cc
+++ b/src/mds/Server.cc
@@ -280,6 +280,7 @@ Server::Server(MDSRank *m, MetricsHandler *metrics_handler) :
 {
   forward_all_requests_to_auth = g_conf().get_val<bool>("mds_forward_all_requests_to_auth");
   replay_unsafe_with_closed_session = g_conf().get_val<bool>("mds_replay_unsafe_with_closed_session");
+  allow_batched_ops = g_conf().get_val<bool>("mds_allow_batched_ops");
   cap_revoke_eviction_timeout = g_conf().get_val<double>("mds_cap_revoke_eviction_timeout");
   max_snaps_per_dir = g_conf().get_val<uint64_t>("mds_max_snaps_per_dir");
   delegate_inos_pct = g_conf().get_val<uint64_t>("mds_client_delegate_inos_pct");
@@ -1378,6 +1379,9 @@ void Server::handle_conf_change(const std::set<std::string>& changed) {
   if (changed.count("mds_forward_all_requests_to_auth")){
     forward_all_requests_to_auth = g_conf().get_val<bool>("mds_forward_all_requests_to_auth");
   }
+  if (changed.count("mds_allow_batched_ops")) {
+    allow_batched_ops = g_conf().get_val<bool>("mds_allow_batched_ops");
+  }
   if (changed.count("mds_cap_revoke_eviction_timeout")) {
     cap_revoke_eviction_timeout = g_conf().get_val<double>("mds_cap_revoke_eviction_timeout");
     dout(20) << __func__ << " cap revoke eviction timeout changed to "
@@ -4185,7 +4189,7 @@ void Server::handle_client_getattr(const MDRequestRef& mdr, bool is_lookup)
   if (mask & CEPH_STAT_RSTAT)
     want_auth = true; // set want_auth for CEPH_STAT_RSTAT mask
 
-  if (!mdr->is_batch_head() && mdr->can_batch()) {
+  if (!mdr->is_batch_head() && allow_batched_ops && mdr->can_batch()) {
     CF_MDS_RetryRequestFactory cf(mdcache, mdr, false);
     int r = mdcache->path_traverse(mdr, cf, mdr->get_filepath(),
 				   (want_auth ? MDS_TRAVERSE_WANT_AUTH : 0),
diff --git a/src/mds/Server.h b/src/mds/Server.h
index 87a2ac9bb05..f994514bbce 100644
--- a/src/mds/Server.h
+++ b/src/mds/Server.h
@@ -603,6 +603,7 @@ private:
   unsigned delegate_inos_pct = 0;
   uint64_t dir_max_entries = 0;
   int64_t bal_fragment_size_max = 0;
+  bool allow_batched_ops = true;
 
   double inject_rename_corrupt_dentry_first = 0.0;