mon/OSDMonitor: add 'osd [rm-]pg-remap[-items] ...' commands

Add commands to add and remove pg_remap mappings.

Require that a mon config option is set before this is allowed so that
users don't inadvertantly prevent older clients from interacting with the
cluster.

Signed-off-by: Sage Weil <sage@redhat.com>

diff --git a/src/common/config_opts.h b/src/common/config_opts.h
index 5dd068089009a77ac42f4ff8f81c9d1ae0875e59..29c44aca40bd038da39b2083e3ea2643d86f36ce 100644 (file)
--- a/src/common/config_opts.h
+++ b/src/common/config_opts.h
@@ -282,6 +282,7 @@ OPTION(mon_osd_max_op_age, OPT_DOUBLE, 32)     // max op age before we get conce
 OPTION(mon_osd_max_split_count, OPT_INT, 32) // largest number of PGs per "involved" OSD to let split create
 OPTION(mon_osd_allow_primary_temp, OPT_BOOL, false)  // allow primary_temp to be set in the osdmap
 OPTION(mon_osd_allow_primary_affinity, OPT_BOOL, false)  // allow primary_affinity to be set in the osdmap
+OPTION(mon_osd_allow_pg_remap, OPT_BOOL, false) // allow pg remap to be set in the osdmap
 OPTION(mon_osd_prime_pg_temp, OPT_BOOL, true)  // prime osdmap with pg mapping changes
 OPTION(mon_osd_prime_pg_temp_max_time, OPT_FLOAT, .5)  // max time to spend priming
 OPTION(mon_osd_prime_pg_temp_max_estimate, OPT_FLOAT, .25) // max estimate of pg total before we do all pgs in parallel

diff --git a/src/mon/MonCommands.h b/src/mon/MonCommands.h
index 7589054bfa996224aea7d9ef04fe27e9967d3c6a..dbe80a46753d0a4ecee0153d9ba8b2655b5c031f 100644 (file)
--- a/src/mon/MonCommands.h
+++ b/src/mon/MonCommands.h
@@ -676,6 +676,25 @@ COMMAND("osd pg-temp " \
        "name=id,type=CephOsdName,n=N,req=false", \
        "set pg_temp mapping pgid:[<id> [<id>...]] (developers only)", \
         "osd", "rw", "cli,rest")
+COMMAND("osd pg-remap " \
+       "name=pgid,type=CephPgid " \
+       "name=id,type=CephOsdName,n=N", \
+       "set pg_remap mapping <pgid>:[<id> [<id>...]] primary <primary> (developers only)", \
+        "osd", "rw", "cli,rest")
+COMMAND("osd rm-pg-remap " \
+       "name=pgid,type=CephPgid",                                      \
+       "clear pg_remap mapping for <pgid> (developers only)", \
+        "osd", "rw", "cli,rest")
+
+COMMAND("osd pg-remap-items " \
+       "name=pgid,type=CephPgid " \
+       "name=id,type=CephOsdName,n=N", \
+       "set pg_remap_items mapping <pgid>:{<id> to <id>, [...]} (developers only)", \
+        "osd", "rw", "cli,rest")
+COMMAND("osd rm-pg-remap-items " \
+       "name=pgid,type=CephPgid",                \
+       "clear pg_remap_items mapping for <pgid> (developers only)", \
+        "osd", "rw", "cli,rest")
 COMMAND("osd primary-temp " \
        "name=pgid,type=CephPgid " \
        "name=id,type=CephOsdName", \

diff --git a/src/mon/OSDMonitor.cc b/src/mon/OSDMonitor.cc
index 1f6024f7d9081cf2b7b7f6085156192766ac2448..d0f4cac5081f418121e541f676955570877fee4e 100644 (file)
--- a/src/mon/OSDMonitor.cc
+++ b/src/mon/OSDMonitor.cc
@@ -7050,6 +7050,206 @@ bool OSDMonitor::prepare_command_impl(MonOpRequestRef op,
     pending_inc.new_primary_temp[pgid] = osd;
     ss << "set " << pgid << " primary_temp mapping to " << osd;
     goto update;
+  } else if (prefix == "osd pg-remap") {
+    if (!g_conf->mon_osd_allow_pg_remap) {
+      ss << "you must enable 'mon osd allow pg remap = true' on the mons before you can adjust pg_remap.  note that pre-luminous clients will no longer be able to communicate with the cluster.";
+      err = -EPERM;
+      goto reply;
+    }
+    err = check_cluster_features(CEPH_FEATUREMASK_OSDMAP_REMAP, ss);
+    if (err == -EAGAIN)
+      goto wait;
+    if (err < 0)
+      goto reply;
+    string pgidstr;
+    if (!cmd_getval(g_ceph_context, cmdmap, "pgid", pgidstr)) {
+      ss << "unable to parse 'pgid' value '"
+         << cmd_vartype_stringify(cmdmap["pgid"]) << "'";
+      err = -EINVAL;
+      goto reply;
+    }
+    pg_t pgid;
+    if (!pgid.parse(pgidstr.c_str())) {
+      ss << "invalid pgid '" << pgidstr << "'";
+      err = -EINVAL;
+      goto reply;
+    }
+    if (!osdmap.pg_exists(pgid)) {
+      ss << "pg " << pgid << " does not exist";
+      err = -ENOENT;
+      goto reply;
+    }
+    if (pending_inc.new_pg_remap.count(pgid) ||
+       pending_inc.old_pg_remap.count(pgid)) {
+      dout(10) << __func__ << " waiting for pending update on " << pgid << dendl;
+      wait_for_finished_proposal(op, new C_RetryMessage(this, op));
+      return true;
+    }
+    vector<int64_t> id_vec;
+    if (!cmd_getval(g_ceph_context, cmdmap, "id", id_vec)) {
+      ss << "unable to parse 'id' value(s) '"
+         << cmd_vartype_stringify(cmdmap["id"]) << "'";
+      err = -EINVAL;
+      goto reply;
+    }
+    vector<int32_t> new_pg_remap;
+    for (auto osd : id_vec) {
+      if (osd != CRUSH_ITEM_NONE && !osdmap.exists(osd)) {
+        ss << "osd." << osd << " does not exist";
+        err = -ENOENT;
+        goto reply;
+      }
+      new_pg_remap.push_back(osd);
+    }
+
+    pending_inc.new_pg_remap[pgid] = new_pg_remap;
+    ss << "set " << pgid << " pg_remap mapping to " << new_pg_remap;
+    goto update;
+  } else if (prefix == "osd rm-pg-remap") {
+    if (!g_conf->mon_osd_allow_pg_remap) {
+      ss << "you must enable 'mon osd allow pg remap = true' on the mons before you can adjust pg_remap.  note that pre-luminous clients will no longer be able to communicate with the cluster.";
+      err = -EPERM;
+      goto reply;
+    }
+    err = check_cluster_features(CEPH_FEATUREMASK_OSDMAP_REMAP, ss);
+    if (err == -EAGAIN)
+      goto wait;
+    if (err < 0)
+      goto reply;
+    string pgidstr;
+    if (!cmd_getval(g_ceph_context, cmdmap, "pgid", pgidstr)) {
+      ss << "unable to parse 'pgid' value '"
+         << cmd_vartype_stringify(cmdmap["pgid"]) << "'";
+      err = -EINVAL;
+      goto reply;
+    }
+    pg_t pgid;
+    if (!pgid.parse(pgidstr.c_str())) {
+      ss << "invalid pgid '" << pgidstr << "'";
+      err = -EINVAL;
+      goto reply;
+    }
+    if (!osdmap.pg_exists(pgid)) {
+      ss << "pg " << pgid << " does not exist";
+      err = -ENOENT;
+      goto reply;
+    }
+    if (pending_inc.new_pg_remap.count(pgid) ||
+       pending_inc.old_pg_remap.count(pgid)) {
+      dout(10) << __func__ << " waiting for pending update on " << pgid << dendl;
+      wait_for_finished_proposal(op, new C_RetryMessage(this, op));
+      return true;
+    }
+
+    pending_inc.old_pg_remap.insert(pgid);
+    ss << "clear " << pgid << " pg_remap mapping";
+    goto update;
+  } else if (prefix == "osd pg-remap-items") {
+    if (!g_conf->mon_osd_allow_pg_remap) {
+      ss << "you must enable 'mon osd allow pg remap = true' on the mons before you can adjust pg_remap.  note that pre-luminous clients will no longer be able to communicate with the cluster.";
+      err = -EPERM;
+      goto reply;
+    }
+    err = check_cluster_features(CEPH_FEATUREMASK_OSDMAP_REMAP, ss);
+    if (err == -EAGAIN)
+      goto wait;
+    if (err < 0)
+      goto reply;
+    string pgidstr;
+    if (!cmd_getval(g_ceph_context, cmdmap, "pgid", pgidstr)) {
+      ss << "unable to parse 'pgid' value '"
+         << cmd_vartype_stringify(cmdmap["pgid"]) << "'";
+      err = -EINVAL;
+      goto reply;
+    }
+    pg_t pgid;
+    if (!pgid.parse(pgidstr.c_str())) {
+      ss << "invalid pgid '" << pgidstr << "'";
+      err = -EINVAL;
+      goto reply;
+    }
+    if (!osdmap.pg_exists(pgid)) {
+      ss << "pg " << pgid << " does not exist";
+      err = -ENOENT;
+      goto reply;
+    }
+    if (pending_inc.new_pg_remap_items.count(pgid) ||
+       pending_inc.old_pg_remap_items.count(pgid)) {
+      dout(10) << __func__ << " waiting for pending update on " << pgid << dendl;
+      wait_for_finished_proposal(op, new C_RetryMessage(this, op));
+      return true;
+    }
+    vector<int64_t> id_vec;
+    if (!cmd_getval(g_ceph_context, cmdmap, "id", id_vec)) {
+      ss << "unable to parse 'id' value(s) '"
+         << cmd_vartype_stringify(cmdmap["id"]) << "'";
+      err = -EINVAL;
+      goto reply;
+    }
+    if (id_vec.size() % 2) {
+      ss << "you must specify pairs of osd ids to be remapped";
+      err = -EINVAL;
+      goto reply;
+    }
+    vector<pair<int32_t,int32_t>> new_pg_remap_items;
+    for (auto p = id_vec.begin(); p != id_vec.end(); ++p) {
+      int from = *p++;
+      int to = *p;
+      if (!osdmap.exists(from)) {
+       ss << "osd." << from << " does not exist";
+       err = -ENOENT;
+       goto reply;
+      }
+      if (to != CRUSH_ITEM_NONE && !osdmap.exists(to)) {
+       ss << "osd." << to << " does not exist";
+       err = -ENOENT;
+       goto reply;
+      }
+      new_pg_remap_items.push_back(make_pair(from, to));
+    }
+
+    pending_inc.new_pg_remap_items[pgid] = new_pg_remap_items;
+    ss << "set " << pgid << " pg_remap_items mapping to " << new_pg_remap_items;
+    goto update;
+  } else if (prefix == "osd rm-pg-remap-items") {
+    if (!g_conf->mon_osd_allow_pg_remap) {
+      ss << "you must enable 'mon osd allow pg remap = true' on the mons before you can adjust pg_remap.  note that pre-luminous clients will no longer be able to communicate with the cluster.";
+      err = -EPERM;
+      goto reply;
+    }
+    err = check_cluster_features(CEPH_FEATUREMASK_OSDMAP_REMAP, ss);
+    if (err == -EAGAIN)
+      goto wait;
+    if (err < 0)
+      goto reply;
+    string pgidstr;
+    if (!cmd_getval(g_ceph_context, cmdmap, "pgid", pgidstr)) {
+      ss << "unable to parse 'pgid' value '"
+         << cmd_vartype_stringify(cmdmap["pgid"]) << "'";
+      err = -EINVAL;
+      goto reply;
+    }
+    pg_t pgid;
+    if (!pgid.parse(pgidstr.c_str())) {
+      ss << "invalid pgid '" << pgidstr << "'";
+      err = -EINVAL;
+      goto reply;
+    }
+    if (!osdmap.pg_exists(pgid)) {
+      ss << "pg " << pgid << " does not exist";
+      err = -ENOENT;
+      goto reply;
+    }
+    if (pending_inc.new_pg_remap_items.count(pgid) ||
+       pending_inc.old_pg_remap_items.count(pgid)) {
+      dout(10) << __func__ << " waiting for pending update on " << pgid << dendl;
+      wait_for_finished_proposal(op, new C_RetryMessage(this, op));
+      return true;
+    }
+
+    pending_inc.old_pg_remap_items.insert(pgid);
+    ss << "clear " << pgid << " pg_remap_items mapping";
+    goto update;
   } else if (prefix == "osd primary-affinity") {
     int64_t id;
     if (!cmd_getval(g_ceph_context, cmdmap, "id", id)) {

diff --git a/src/vstart.sh b/src/vstart.sh
index 015db3ad04fd26acd0e8703ab6de36150f65964f..06a804f16b07a10986453fb69c3905d94c6c7436 100755 (executable)
--- a/src/vstart.sh
+++ b/src/vstart.sh
@@ -472,6 +472,7 @@ $extra_conf
 [mon]
         mon pg warn min per osd = 3
         mon osd allow primary affinity = true
+        mon osd allow pg remap = true
         mon reweight min pgs per osd = 4
         mon osd prime pg temp = true
         crushtool = $CEPH_BIN/crushtool