From: Yanhu Cao <gmayyyha@gmail.com>
Date: Tue, 11 Aug 2020 08:10:26 +0000 (+0800)
Subject: client: choose a random replica mds to send the request
X-Git-Tag: wip-pdonnell-testing-20200918.022351~296^2
X-Git-Url: http://git.apps.os.sepia.ceph.com/?a=commitdiff_plain;h=504c8cbc2cfcfd0822a70b07b07330be2abb8212;p=ceph-ci.git

client: choose a random replica mds to send the request

Fixes: https://tracker.ceph.com/issues/46894
Signed-off-by: Yanhu Cao <gmayyyha@gmail.com>
---

diff --git a/src/client/Client.cc b/src/client/Client.cc
index e7eb2c23ef7..bf414edbfda 100755
--- a/src/client/Client.cc
+++ b/src/client/Client.cc
@@ -84,6 +84,7 @@
 #include "include/lru.h"
 #include "include/compat.h"
 #include "include/stringify.h"
+#include "include/random.h"
 
 #include "Client.h"
 #include "Inode.h"
@@ -1073,7 +1074,11 @@ void Client::update_dir_dist(Inode *in, DirStat *dst)
   }
 
   // replicated
-  in->dir_replicated = !dst->dist.empty();  // FIXME that's just one frag!
+  in->dir_replicated = !dst->dist.empty();
+  if (!dst->dist.empty())
+    in->frag_repmap[dst->frag].assign(dst->dist.begin(), dst->dist.end()) ;
+  else
+    in->frag_repmap.erase(dst->frag);
 }
 
 void Client::clear_dir_complete_and_ordered(Inode *diri, bool complete)
@@ -1498,9 +1503,16 @@ mds_rank_t Client::choose_target_mds(MetaRequest *req, Inode** phash_diri)
     ldout(cct, 20) << __func__ << " " << *in << " is_hash=" << is_hash
              << " hash=" << hash << dendl;
   
-    if (is_hash && S_ISDIR(in->mode) && !in->fragmap.empty()) {
+    if (is_hash && S_ISDIR(in->mode) && (!in->fragmap.empty() || !in->frag_repmap.empty())) {
       frag_t fg = in->dirfragtree[hash];
-      if (in->fragmap.count(fg)) {
+      if (!req->auth_is_best()) {
+        auto repmapit = in->frag_repmap.find(fg);
+        if (repmapit != in->frag_repmap.end()) {
+          auto& repmap = repmapit->second;
+          auto r = ceph::util::generate_random_number<uint64_t>(0, repmap.size()-1);
+          mds = repmap.at(r);
+        }
+      } else if (in->fragmap.count(fg)) {
 	mds = in->fragmap[fg];
 	if (phash_diri)
 	  *phash_diri = in;
diff --git a/src/client/Inode.h b/src/client/Inode.h
index afaf64e0515..dce9be84569 100644
--- a/src/client/Inode.h
+++ b/src/client/Inode.h
@@ -227,6 +227,7 @@ struct Inode {
   string    symlink;  // symlink content, if it's a symlink
   map<string,bufferptr> xattrs;
   map<frag_t,int> fragmap;  // known frag -> mds mappings
+  map<frag_t, std::vector<mds_rank_t>> frag_repmap; // non-auth mds mappings
 
   std::list<ceph::condition_variable*> waitfor_caps;
   std::list<ceph::condition_variable*> waitfor_commit;