From: Igor Fedotov <ifed@suse.com>
Date: Fri, 5 Nov 2021 09:38:51 +0000 (+0300)
Subject: os/bluestore: do not select absent device in volume selector
X-Git-Tag: v16.2.7~24^2
X-Git-Url: http://git-server-git.apps.pok.os.sepia.ceph.com/?a=commitdiff_plain;h=edbdfd5002d4cf5895adbb9f87b7817b14272e19;p=ceph.git

os/bluestore: do not select absent device in volume selector

Fixes: ttps://tracker.ceph.com/issues/53139
Signed-off-by: Igor Fedotov <igor.fedotov@croit.io>
(cherry picked from commit 39132b0e460d2d667f02019335dced50d5db641d)
---

diff --git a/src/os/bluestore/BlueFS.cc b/src/os/bluestore/BlueFS.cc
index 16b64636b891..37a000ac6ef6 100644
--- a/src/os/bluestore/BlueFS.cc
+++ b/src/os/bluestore/BlueFS.cc
@@ -3819,10 +3819,9 @@ void* OriginalVolumeSelector::get_hint_by_dir(std::string_view dirname) const {
     // match up with bluestore.  the slow device is always the second
     // one (when a dedicated block.db device is present and used at
     // bdev 0).  the wal device is always last.
-    if (boost::algorithm::ends_with(dirname, ".slow")) {
+    if (boost::algorithm::ends_with(dirname, ".slow") && slow_total) {
       res = BlueFS::BDEV_SLOW;
-    }
-    else if (boost::algorithm::ends_with(dirname, ".wal")) {
+    } else if (boost::algorithm::ends_with(dirname, ".wal") && wal_total) {
       res = BlueFS::BDEV_WAL;
     }
   }
diff --git a/src/os/bluestore/BlueStore.cc b/src/os/bluestore/BlueStore.cc
index 04d67cabe479..77a12bd51009 100644
--- a/src/os/bluestore/BlueStore.cc
+++ b/src/os/bluestore/BlueStore.cc
@@ -9176,6 +9176,23 @@ void BlueStore::inject_zombie_spanning_blob(coll_t cid, ghobject_t oid,
   db->submit_transaction_sync(txn);
 }
 
+void BlueStore::inject_bluefs_file(std::string_view dir, std::string_view name, size_t new_size)
+{
+  ceph_assert(bluefs);
+
+  BlueFS::FileWriter* p_handle = nullptr;
+  auto ret = bluefs->open_for_write(dir, name, &p_handle, false);
+  ceph_assert(ret == 0);
+
+  std::string s('0', new_size);
+  bufferlist bl;
+  bl.append(s);
+  p_handle->append(bl);
+
+  bluefs->fsync(p_handle);
+  bluefs->close_writer(p_handle);
+}
+
 void BlueStore::collect_metadata(map<string,string> *pm)
 {
   dout(10) << __func__ << dendl;
@@ -16534,8 +16551,13 @@ uint8_t RocksDBBlueFSVolumeSelector::select_prefer_bdev(void* h) {
 
 void RocksDBBlueFSVolumeSelector::get_paths(const std::string& base, paths& res) const
 {
-  res.emplace_back(base, l_totals[LEVEL_DB - LEVEL_FIRST]);
-  res.emplace_back(base + ".slow", l_totals[LEVEL_SLOW - LEVEL_FIRST]);
+  auto db_size = l_totals[LEVEL_DB - LEVEL_FIRST];
+  res.emplace_back(base, db_size);
+  auto slow_size = l_totals[LEVEL_SLOW - LEVEL_FIRST];
+  if (slow_size == 0) {
+    slow_size = db_size;
+  }
+  res.emplace_back(base + ".slow", slow_size);
 }
 
 void* RocksDBBlueFSVolumeSelector::get_hint_by_dir(std::string_view dirname) const {
diff --git a/src/os/bluestore/BlueStore.h b/src/os/bluestore/BlueStore.h
index 3783f12c928d..0a5c451e5602 100644
--- a/src/os/bluestore/BlueStore.h
+++ b/src/os/bluestore/BlueStore.h
@@ -3024,6 +3024,10 @@ public:
   // resets per_pool_omap | pgmeta_omap for onode
   void inject_legacy_omap(coll_t cid, ghobject_t oid);
 
+  void inject_bluefs_file(std::string_view dir,
+			  std::string_view name,
+			  size_t new_size);
+
   void compact() override {
     ceph_assert(db);
     db->compact();
diff --git a/src/test/objectstore/store_test.cc b/src/test/objectstore/store_test.cc
index 069d72101b5d..c3926a3bf4a4 100644
--- a/src/test/objectstore/store_test.cc
+++ b/src/test/objectstore/store_test.cc
@@ -9525,6 +9525,69 @@ TEST_P(StoreTestSpecificAUSize, OmapUpgradeTest) {
   }
 }
 
+TEST_P(StoreTestSpecificAUSize, BluefsWriteInSingleDiskEnvTest) {
+  if (string(GetParam()) != "bluestore")
+    return;
+
+  g_conf().apply_changes(nullptr);
+
+  StartDeferred(0x1000);
+
+  BlueStore* bstore = dynamic_cast<BlueStore*> (store.get());
+  ceph_assert(bstore);
+  bstore->inject_bluefs_file("db.slow", "store_test_injection_slow", 1 << 20ul);
+  bstore->inject_bluefs_file("db.wal", "store_test_injection_wal", 1 << 20ul);
+  bstore->inject_bluefs_file("db", "store_test_injection_wal", 1 << 20ul);
+
+  AdminSocket* admin_socket = g_ceph_context->get_admin_socket();
+  ceph_assert(admin_socket);
+
+  ceph::bufferlist in, out;
+  ostringstream err;
+  auto r = admin_socket->execute_command(
+    { "{\"prefix\": \"bluefs stats\"}" },
+    in, err, &out);
+  if (r != 0) {
+    cerr << "failure querying: " << cpp_strerror(r) << std::endl;
+  }  else {
+    std::cout << std::string(out.c_str(), out.length()) << std::endl;
+  }
+}
+
+TEST_P(StoreTestSpecificAUSize, BluefsWriteInNoWalDiskEnvTest) {
+  if (string(GetParam()) != "bluestore")
+    return;
+
+  SetVal(g_conf(), "bluestore_block_db_path", "db");
+  SetVal(g_conf(), "bluestore_block_db_size", stringify(1ull << 31).c_str());
+  SetVal(g_conf(), "bluestore_block_db_create", "true");
+
+  g_conf().apply_changes(nullptr);
+
+  StartDeferred(0x1000);
+
+  BlueStore* bstore = dynamic_cast<BlueStore*> (store.get());
+  ceph_assert(bstore);
+  bstore->inject_bluefs_file("db.slow", "store_test_injection_slow", 1 << 20ul);
+  bstore->inject_bluefs_file("db.wal", "store_test_injection_wal", 1 << 20ul);
+  bstore->inject_bluefs_file("db", "store_test_injection_wal", 1 << 20ul);
+
+  AdminSocket* admin_socket = g_ceph_context->get_admin_socket();
+  ceph_assert(admin_socket);
+
+  ceph::bufferlist in, out;
+  ostringstream err;
+  auto r = admin_socket->execute_command(
+    { "{\"prefix\": \"bluefs stats\"}" },
+    in, err, &out);
+  if (r != 0) {
+    cerr << "failure querying: " << cpp_strerror(r) << std::endl;
+  }
+  else {
+    std::cout << std::string(out.c_str(), out.length()) << std::endl;
+  }
+}
+
 #endif  // WITH_BLUESTORE
 
 int main(int argc, char **argv) {