Merge pull request #2054 from ceph/wip-fs-cmds-oops

Apply some accidentally-dropped commits from the wip-fs-cmds branch to be a little friendlier and pass tests.

Reviewed-by: Greg Farnum <greg@inktank.com>

qa/workunits/cephtool/test.sh

@@ -296,7 +296,11 @@ function test_mon_mds()
 {
   ceph osd pool create fs_data 10
   ceph osd pool create fs_metadata 10
-  ceph fs new default fs_metadata fs_data
+  ceph fs new cephfs fs_metadata fs_data
+
+  # Check for default crash_replay_interval set automatically in 'fs new'
+  ceph osd dump | grep fs_data > $TMPFILE
+  check_response "crash_replay_interval 45 "
 
   ceph mds cluster_down
   ceph mds cluster_up
@@ -356,13 +360,24 @@ function test_mon_mds()
   ceph mds add_data_pool mds-ec-pool 2>$TMPFILE
   check_response 'erasure-code' $? 22
   set -e
-  poolnum=$(ceph osd dump | grep 'pool.*mds-ec-pool' | awk '{print $2;}')
+  ec_poolnum=$(ceph osd dump | grep 'pool.*mds-ec-pool' | awk '{print $2;}')
+  data_poolnum=$(ceph osd dump | grep 'pool.*fs_data' | awk '{print $2;}')
+  metadata_poolnum=$(ceph osd dump | grep 'pool.*fs_metadata' | awk '{print $2;}')
   set +e
-  ceph mds newfs 0 $poolnum --yes-i-really-mean-it 2>$TMPFILE
+
+  ceph fs rm cephfs --yes-i-really-mean-it
+
+  ceph mds newfs $metadata_poolnum $ec_poolnum --yes-i-really-mean-it 2>$TMPFILE
   check_response 'erasure-code' $? 22
-  ceph mds newfs $poolnum 1 --yes-i-really-mean-it 2>$TMPFILE
+  ceph mds newfs $ec_poolnum $data_poolnum --yes-i-really-mean-it 2>$TMPFILE
   check_response 'erasure-code' $? 22
-  ceph mds newfs $poolnum $poolnum --yes-i-really-mean-it 2>$TMPFILE
+  ceph mds newfs $ec_poolnum $ec_poolnum --yes-i-really-mean-it 2>$TMPFILE
+  check_response 'erasure-code' $? 22
+  ceph fs new cephfs fs_metadata mds-ec-pool 2>$TMPFILE
+  check_response 'erasure-code' $? 22
+  ceph fs new cephfs mds-ec-pool fs_data 2>$TMPFILE
+  check_response 'erasure-code' $? 22
+  ceph fs new cephfs mds-ec-pool mds-ec-pool 2>$TMPFILE
   check_response 'erasure-code' $? 22
   set -e
   ceph osd pool delete mds-ec-pool mds-ec-pool --yes-i-really-really-mean-it
@@ -374,7 +389,6 @@ function test_mon_mds()
   # ceph mds set_state
   # ceph mds stop
 
-  ceph fs rm default --yes-i-really-mean-it
   ceph osd pool delete fs_data fs_data --yes-i-really-really-mean-it
   ceph osd pool delete fs_metadata fs_metadata --yes-i-really-really-mean-it
 }

qa/workunits/fs/misc/layout_vxattrs.sh

@@ -22,9 +22,13 @@ getfattr -n ceph.dir.layout file    2>&1 | grep -q 'No such attribute'
 setfattr -n ceph.file.layout.stripe_unit -v 1048576 file2
 setfattr -n ceph.file.layout.stripe_count -v 8 file2
 setfattr -n ceph.file.layout.object_size -v 10485760 file2
+
+# Assumption: that the data pool is called 'data' and has ID '2'
+# (see teuthology.task.ceph.cephfs_setup)
 setfattr -n ceph.file.layout.pool -v data file2
-setfattr -n ceph.file.layout.pool -v 0 file2
+setfattr -n ceph.file.layout.pool -v 2 file2
 getfattr -n ceph.file.layout.pool file2 | grep -q data
+
 getfattr -n ceph.file.layout.stripe_unit file2 | grep -q 1048576
 getfattr -n ceph.file.layout.stripe_count file2 | grep -q 8
 getfattr -n ceph.file.layout.object_size file2 | grep -q 10485760
@@ -52,8 +56,12 @@ getfattr -d -m - dir | grep -q ceph.file.layout      && exit 1 || true
 setfattr -n ceph.dir.layout.stripe_unit -v 1048576 dir
 setfattr -n ceph.dir.layout.stripe_count -v 8 dir
 setfattr -n ceph.dir.layout.object_size -v 10485760 dir
+
+# Assumption: that the data pool is called 'data' and has ID '2'
+# (see teuthology.task.ceph.cephfs_setup)
 setfattr -n ceph.dir.layout.pool -v data dir
-setfattr -n ceph.dir.layout.pool -v 0 dir
+setfattr -n ceph.dir.layout.pool -v 2 dir
+
 getfattr -n ceph.dir.layout dir
 getfattr -n ceph.dir.layout dir | grep -q object_size=10485760
 getfattr -n ceph.dir.layout dir | grep -q stripe_count=8

src/mds/MDSMap.h

@@ -71,6 +71,8 @@ extern CompatSet get_mdsmap_compat_set_base(); // pre v0.20
 #define MDS_FEATURE_INCOMPAT_INLINE CompatSet::Feature(7, "mds uses inline data")
 #define MDS_FEATURE_INCOMPAT_NOANCHOR CompatSet::Feature(8, "no anchor table")
 
+#define MDS_FS_NAME_DEFAULT "cephfs"
+
 class MDSMap {
 public:
   // mds states
@@ -193,7 +195,7 @@ public:
 
 public:
   MDSMap() 
-    : epoch(0), enabled(false), fs_name("default"),
+    : epoch(0), enabled(false), fs_name(MDS_FS_NAME_DEFAULT),
       flags(0), last_failure(0),
       last_failure_osd_epoch(0),
       tableserver(0), root(0),

src/mon/MDSMonitor.cc

@@ -908,8 +908,8 @@ bool MDSMonitor::management_command(
 {
   if (prefix == "mds newfs") {
     /* Legacy `newfs` command, takes pool numbers instead of
-     * names, assumes fs name to be 'default', and can
-     * overwrite existing filesystem settings */
+     * names, assumes fs name to be MDS_FS_NAME_DEFAULT, and
+     * can overwrite existing filesystem settings */
     MDSMap newmap;
     int64_t metadata, data;
 
@@ -966,7 +966,7 @@ bool MDSMonitor::management_command(
       newmap.inc = pending_mdsmap.inc;
       pending_mdsmap = newmap;
       pending_mdsmap.epoch = mdsmap.epoch + 1;
-      create_new_fs(pending_mdsmap, "default", metadata, data);
+      create_new_fs(pending_mdsmap, MDS_FS_NAME_DEFAULT, metadata, data);
       ss << "new fs with metadata pool " << metadata << " and data pool " << data;
       r = 0;
       return true;
@@ -989,15 +989,7 @@ bool MDSMonitor::management_command(
       ss << "pool '" << data_name << "' does not exist";
       return true;
     }
-
-    // Warn if crash_replay_interval is not set on the data pool
-    //  (on creation should have done pools[pool].crash_replay_interval =
-    //  cct->_conf->osd_default_data_pool_replay_window;)
-    pg_pool_t const *data_pool = mon->osdmon()->osdmap.get_pg_pool(data);
-    if (data_pool->get_crash_replay_interval() == 0) {
-      ss << "warning: crash_replay_interval not set on data pool '" << data << "', ";
-    }
-    
+   
     string fs_name;
     cmd_getval(g_ceph_context, cmdmap, "fs_name", fs_name);
     if (fs_name.empty()) {
@@ -1023,6 +1015,31 @@ bool MDSMonitor::management_command(
       r = -EINVAL;
     }
 
+    pg_pool_t const *data_pool = mon->osdmon()->osdmap.get_pg_pool(data);
+    assert(data_pool != NULL);  // Checked it existed above
+    pg_pool_t const *metadata_pool = mon->osdmon()->osdmap.get_pg_pool(metadata);
+    assert(metadata_pool != NULL);  // Checked it existed above
+
+    // Automatically set crash_replay_interval on data pool if it
+    // isn't already set.
+    if (data_pool->get_crash_replay_interval() == 0) {
+      r = mon->osdmon()->set_crash_replay_interval(data, g_conf->osd_default_data_pool_replay_window);
+      assert(r == 0);  // We just did get_pg_pool so it must exist and be settable
+      request_proposal(mon->osdmon());
+    }
+
+    if (data_pool->is_erasure()) {
+      ss << "data pool '" << data_name << " is an erasure-code pool";
+      r = -EINVAL;
+      return true;
+    }
+
+    if (metadata_pool->is_erasure()) {
+      ss << "metadata pool '" << metadata_name << " is an erasure-code pool";
+      r = -EINVAL;
+      return true;
+    }
+
     // All checks passed, go ahead and create.
     MDSMap newmap;
     newmap.inc = pending_mdsmap.inc;

src/mon/OSDMonitor.cc

@@ -3309,6 +3309,35 @@ int OSDMonitor::parse_osd_id(const char *s, stringstream *pss)
   return id;
 }
 
+
+/**
+ * Special setter for crash_replay_interval on a pool.  Equivalent to
+ * using prepare_command_pool_set, but in a form convenient for use
+ * from MDSMonitor rather than from an administrative command.
+ */
+int OSDMonitor::set_crash_replay_interval(const int64_t pool_id, const uint32_t cri)
+{
+  pg_pool_t p;
+  if (pending_inc.new_pools.count(pool_id)) {
+    p = pending_inc.new_pools[pool_id];
+  } else {
+    const pg_pool_t *p_ptr = osdmap.get_pg_pool(pool_id);
+    if (p_ptr == NULL) {
+      return -ENOENT;
+    } else {
+      p = *p_ptr;
+    }
+  }
+
+  dout(10) << "Set pool " << pool_id << " crash_replay_interval=" << cri << dendl;
+  p.crash_replay_interval = cri;
+  p.last_change = pending_inc.epoch;
+  pending_inc.new_pools[pool_id] = p;
+
+  return 0;
+}
+
+
 int OSDMonitor::prepare_command_pool_set(map<string,cmd_vartype> &cmdmap,
                                          stringstream& ss)
 {

src/mon/OSDMonitor.h

@@ -363,6 +363,7 @@ private:
   bool prepare_command(MMonCommand *m);
   bool prepare_command_impl(MMonCommand *m, map<string,cmd_vartype> &cmdmap);
 
+  int set_crash_replay_interval(const int64_t pool_id, const uint32_t cri);
   int prepare_command_pool_set(map<string,cmd_vartype> &cmdmap,
                                stringstream& ss);
 

src/vstart.sh

@@ -536,7 +536,7 @@ EOF
         echo $cmd
         $cmd
 
-        cmd="$CEPH_ADM fs new default cephfs_metadata cephfs_data"
+        cmd="$CEPH_ADM fs new cephfs cephfs_metadata cephfs_data"
         echo $cmd
         $cmd
 	fi