From: shenhang <shenhang@kuaishou.com>
Date: Fri, 10 Sep 2021 10:59:49 +0000 (+0800)
Subject: mds: add an option to decide whether prefetching entire dir or not.
X-Git-Tag: v18.0.0~1394^2
X-Git-Url: http://git-server-git.apps.pok.os.sepia.ceph.com/?a=commitdiff_plain;h=refs%2Fpull%2F39315%2Fhead;p=ceph.git

mds: add an option to decide whether prefetching entire dir or not.
Accessing one single dentry could be fastened by set this option to
false, when dir is not in the memory.
Signed-off-by: "Shen, Hang" <shenhang@kuaishou.com>
---

diff --git a/qa/cephfs/overrides/prefetch_entire_dirfrags/no.yaml b/qa/cephfs/overrides/prefetch_entire_dirfrags/no.yaml
new file mode 100644
index 000000000000..41903ef45fca
--- /dev/null
+++ b/qa/cephfs/overrides/prefetch_entire_dirfrags/no.yaml
@@ -0,0 +1,4 @@
+overrides:
+  ceph:
+    cephfs:
+      mds_dir_prefetch: false 
diff --git a/qa/cephfs/overrides/prefetch_entire_dirfrags/yes.yaml b/qa/cephfs/overrides/prefetch_entire_dirfrags/yes.yaml
new file mode 100644
index 000000000000..143865a33f1a
--- /dev/null
+++ b/qa/cephfs/overrides/prefetch_entire_dirfrags/yes.yaml
@@ -0,0 +1,4 @@
+overrides:
+  ceph:
+    cephfs:
+      mds_dir_prefetch: true
diff --git a/qa/suites/fs/thrash/workloads/overrides/prefetch_entire_dirfrags b/qa/suites/fs/thrash/workloads/overrides/prefetch_entire_dirfrags
new file mode 120000
index 000000000000..9b8024fbadd1
--- /dev/null
+++ b/qa/suites/fs/thrash/workloads/overrides/prefetch_entire_dirfrags
@@ -0,0 +1 @@
+.qa/cephfs/overrides/prefetch_entire_dirfrags
\ No newline at end of file
diff --git a/src/common/options/mds.yaml.in b/src/common/options/mds.yaml.in
index 697710484c8f..3a6464ebcdd7 100644
--- a/src/common/options/mds.yaml.in
+++ b/src/common/options/mds.yaml.in
@@ -395,6 +395,15 @@ options:
   - mds
   flags:
   - runtime
+- name: mds_dir_prefetch
+  type: bool
+  level: advanced
+  desc: flag to prefetch entire dir
+  default: true
+  services:
+  - mds
+  flags:
+  - runtime
 - name: mds_tick_interval
   type: float
   level: advanced
diff --git a/src/mds/CDir.cc b/src/mds/CDir.cc
index 5bd44be7f4c1..31b5eff331cb 100644
--- a/src/mds/CDir.cc
+++ b/src/mds/CDir.cc
@@ -1567,7 +1567,7 @@ void CDir::fetch(std::string_view dname, snapid_t last,
 
   // FIXME: to fetch a snap dentry, we need to get omap key in range
   //       [(name, last), (name, CEPH_NOSNAP))
-  if (!dname.empty() && last == CEPH_NOSNAP) {
+  if (!dname.empty() && last == CEPH_NOSNAP && !g_conf().get_val<bool>("mds_dir_prefetch")) {
     dentry_key_t key(last, dname, inode->hash_dentry_name(dname));
     fetch_keys({key}, c);
     return;
@@ -1594,7 +1594,7 @@ void CDir::fetch(std::string_view dname, snapid_t last,
 
 void CDir::fetch_keys(const std::vector<dentry_key_t>& keys, MDSContext *c)
 {
-  dout(10) << "fetch " << keys.size() << " keys on " << *this << dendl;
+  dout(10) << __func__ << " " << keys.size() << " keys on " << *this << dendl;
   ceph_assert(is_auth());
   ceph_assert(!is_complete());
 
@@ -2132,7 +2132,7 @@ void CDir::_omap_fetched(bufferlist& hdrbl, map<string, bufferlist>& omap,
   for (auto p = omap.rbegin(); p != omap.rend(); ++p, --pos) {
     string_snap_t key;
     dentry_key_t::decode_helper(p->first, key.name, key.snapid);
-    bool touch;
+    bool touch = false;
 
     if (key.snapid == CEPH_NOSNAP) {
       if (complete) {
@@ -2141,8 +2141,6 @@ void CDir::_omap_fetched(bufferlist& hdrbl, map<string, bufferlist>& omap,
 	touch = proc_nulls_and_waiters(p->first, key);
       }
       last_name = std::string_view(p->first.c_str(), key.name.length());
-    } else {
-      touch = false;
     }
 
     CDentry *dn = nullptr;
diff --git a/src/mds/MDCache.cc b/src/mds/MDCache.cc
index eb0bc7e645a0..8e97ee551990 100644
--- a/src/mds/MDCache.cc
+++ b/src/mds/MDCache.cc
@@ -5974,19 +5974,28 @@ bool MDCache::open_undef_inodes_dirfrags()
   map<CDir*, pair<bool, std::vector<dentry_key_t> > > fetch_queue;
   for (auto& dir : rejoin_undef_dirfrags) {
     ceph_assert(dir->get_version() == 0);
-    (void)fetch_queue[dir];
+    fetch_queue.emplace(std::piecewise_construct, std::make_tuple(dir), std::make_tuple());
   }
 
-  for (auto& in : rejoin_undef_inodes) {
-    assert(!in->is_base());
-    CDentry *dn = in->get_parent_dn();
-    auto& p = fetch_queue[dn->get_dir()];
-    if (dn->last != CEPH_NOSNAP) {
-      p.first = true;
-      p.second.clear();
-    } else if (!p.first) {
-      p.second.push_back(dn->key());
+  if (g_conf().get_val<bool>("mds_dir_prefetch")) {
+    for (auto& in : rejoin_undef_inodes) {
+      ceph_assert(!in->is_base());
+      ceph_assert(in->get_parent_dir());
+      fetch_queue.emplace(std::piecewise_construct, std::make_tuple(in->get_parent_dir()), std::make_tuple());
     }
+  } else {
+    for (auto& in : rejoin_undef_inodes) {
+      assert(!in->is_base());
+      CDentry *dn = in->get_parent_dn();
+      auto& p = fetch_queue[dn->get_dir()];
+
+      if (dn->last != CEPH_NOSNAP) {
+        p.first = true;
+        p.second.clear();
+      } else if (!p.first) {
+        p.second.push_back(dn->key());
+      }
+         }
   }
 
   if (fetch_queue.empty())