mon/OSDMonitor: implement MOSDMarkMeDead handling

Allow updates to the dead_epoch in osd_xinfo, so that the OSD can inform
us that they know they are down (as of a particular epoch).

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

diff --git a/src/messages/MOSDMarkMeDead.h b/src/messages/MOSDMarkMeDead.h
new file mode 100644 (file)
index 0000000..9e8b306
--- /dev/null
+++ b/src/messages/MOSDMarkMeDead.h
@@ -0,0 +1,62 @@
+// -*- mode:C++; tab-width:8; c-basic-offset:2; indent-tabs-mode:t -*-
+// vim: ts=8 sw=2 smarttab
+
+#pragma once
+
+#include "messages/PaxosServiceMessage.h"
+
+class MOSDMarkMeDead : public PaxosServiceMessage {
+private:
+  static constexpr int HEAD_VERSION = 1;
+  static constexpr int COMPAT_VERSION = 1;
+
+ public:
+  uuid_d fsid;
+  int32_t target_osd;
+  epoch_t epoch = 0;
+
+  MOSDMarkMeDead()
+    : PaxosServiceMessage{MSG_OSD_MARK_ME_DEAD, 0,
+                         HEAD_VERSION, COMPAT_VERSION} { }
+  MOSDMarkMeDead(const uuid_d &fs, int osd,
+                epoch_t e)
+    : PaxosServiceMessage{MSG_OSD_MARK_ME_DEAD, e,
+                         HEAD_VERSION, COMPAT_VERSION},
+      fsid(fs), target_osd(osd),
+      epoch(e) {}
+ private:
+  ~MOSDMarkMeDead() override {}
+
+public:
+  epoch_t get_epoch() const { return epoch; }
+
+  void decode_payload() override {
+    auto p = payload.cbegin();
+    paxos_decode(p);
+    decode(fsid, p);
+    decode(target_osd, p);
+    decode(epoch, p);
+  }
+
+  void encode_payload(uint64_t features) override {
+    using ceph::encode;
+    paxos_encode();
+    header.version = HEAD_VERSION;
+    header.compat_version = COMPAT_VERSION;
+    encode(fsid, payload);
+    encode(target_osd, payload, features);
+    encode(epoch, payload);
+  }
+
+  std::string_view get_type_name() const override { return "MOSDMarkMeDead"; }
+  void print(ostream& out) const override {
+    out << "MOSDMarkMeDead("
+       << "osd." << target_osd
+       << ", epoch " << epoch
+       << ", fsid=" << fsid
+       << ")";
+  }
+private:
+  template<class T, typename... Args>
+  friend boost::intrusive_ptr<T> ceph::make_message(Args&&... args);
+};

diff --git a/src/mon/Monitor.cc b/src/mon/Monitor.cc
index 85185adf7a40a5861dedbf0840f531584f71be37..d33b0d89d4c02c0d2cc4bf19baf52c33e3f30490 100644 (file)
--- a/src/mon/Monitor.cc
+++ b/src/mon/Monitor.cc
@@ -4481,6 +4481,7 @@ void Monitor::dispatch_op(MonOpRequestRef op)
     case CEPH_MSG_POOLOP:
     case MSG_OSD_BEACON:
     case MSG_OSD_MARK_ME_DOWN:
+    case MSG_OSD_MARK_ME_DEAD:
     case MSG_OSD_FULL:
     case MSG_OSD_FAILURE:
     case MSG_OSD_BOOT:

diff --git a/src/mon/OSDMonitor.cc b/src/mon/OSDMonitor.cc
index f69fdf0a5841e3bdb7be699fb520207ea6d0b5f3..0944f8164cf052a29af3e9e42b1254458cd56e70 100644 (file)
--- a/src/mon/OSDMonitor.cc
+++ b/src/mon/OSDMonitor.cc
@@ -39,6 +39,7 @@
 #include "messages/MOSDBeacon.h"
 #include "messages/MOSDFailure.h"
 #include "messages/MOSDMarkMeDown.h"
+#include "messages/MOSDMarkMeDead.h"
 #include "messages/MOSDFull.h"
 #include "messages/MOSDMap.h"
 #include "messages/MMonGetOSDMap.h"
@@ -2287,6 +2288,8 @@ bool OSDMonitor::preprocess_query(MonOpRequestRef op)
     // damp updates
   case MSG_OSD_MARK_ME_DOWN:
     return preprocess_mark_me_down(op);
+  case MSG_OSD_MARK_ME_DEAD:
+    return preprocess_mark_me_dead(op);
   case MSG_OSD_FULL:
     return preprocess_full(op);
   case MSG_OSD_FAILURE:
@@ -2329,6 +2332,8 @@ bool OSDMonitor::prepare_update(MonOpRequestRef op)
     // damp updates
   case MSG_OSD_MARK_ME_DOWN:
     return prepare_mark_me_down(op);
+  case MSG_OSD_MARK_ME_DEAD:
+    return prepare_mark_me_dead(op);
   case MSG_OSD_FULL:
     return prepare_full(op);
   case MSG_OSD_FAILURE:
@@ -2607,6 +2612,62 @@ bool OSDMonitor::prepare_mark_me_down(MonOpRequestRef op)
   return true;
 }
 
+bool OSDMonitor::preprocess_mark_me_dead(MonOpRequestRef op)
+{
+  op->mark_osdmon_event(__func__);
+  MOSDMarkMeDead *m = static_cast<MOSDMarkMeDead*>(op->get_req());
+  int from = m->target_osd;
+
+  // check permissions
+  if (check_source(op, m->fsid)) {
+    mon->no_reply(op);
+    return true;
+  }
+
+  // first, verify the reporting host is valid
+  if (!m->get_orig_source().is_osd()) {
+    mon->no_reply(op);
+    return true;
+  }
+
+  if (!osdmap.exists(from) ||
+      !osdmap.is_down(from)) {
+    dout(5) << __func__ << " from nonexistent or up osd." << from
+           << ", ignoring" << dendl;
+    send_incremental(op, m->get_epoch()+1);
+    mon->no_reply(op);
+    return true;
+  }
+
+  return false;
+}
+
+bool OSDMonitor::prepare_mark_me_dead(MonOpRequestRef op)
+{
+  op->mark_osdmon_event(__func__);
+  MOSDMarkMeDead *m = static_cast<MOSDMarkMeDead*>(op->get_req());
+  int target_osd = m->target_osd;
+
+  ceph_assert(osdmap.is_down(target_osd));
+
+  mon->clog->info() << "osd." << target_osd << " marked itself dead as of e"
+                   << m->get_epoch();
+  if (!pending_inc.new_xinfo.count(target_osd)) {
+    pending_inc.new_xinfo[target_osd] = osdmap.osd_xinfo[target_osd];
+  }
+  pending_inc.new_xinfo[target_osd].dead_epoch = m->get_epoch();
+  wait_for_finished_proposal(
+    op,
+    new FunctionContext(
+      [op, this] (int r) {
+       if (r >= 0) {
+         mon->no_reply(op);      // ignore on success
+       }
+      }
+      ));
+  return true;
+}
+
 bool OSDMonitor::can_mark_down(int i)
 {
   if (osdmap.is_nodown(i)) {

diff --git a/src/mon/OSDMonitor.h b/src/mon/OSDMonitor.h
index 896f4ee7f63c2a09d9c1bb26e85d723cccdff61d..f856f0161ef97fa411714c9f065689af018b649f 100644 (file)
--- a/src/mon/OSDMonitor.h
+++ b/src/mon/OSDMonitor.h
@@ -403,6 +403,9 @@ private:
   void process_failures();
   void take_all_failures(list<MonOpRequestRef>& ls);
 
+  bool preprocess_mark_me_dead(MonOpRequestRef op);
+  bool prepare_mark_me_dead(MonOpRequestRef op);
+
   bool preprocess_full(MonOpRequestRef op);
   bool prepare_full(MonOpRequestRef op);
 

diff --git a/src/msg/Message.cc b/src/msg/Message.cc
index 802c73d7a6314f2329aa97b84aa1115c6d7d8d26..de20575287deef379583d142052e1f2cf50d41be 100644 (file)
--- a/src/msg/Message.cc
+++ b/src/msg/Message.cc
@@ -62,6 +62,7 @@
 #include "messages/MOSDPGTemp.h"
 #include "messages/MOSDFailure.h"
 #include "messages/MOSDMarkMeDown.h"
+#include "messages/MOSDMarkMeDead.h"
 #include "messages/MOSDFull.h"
 #include "messages/MOSDPing.h"
 #include "messages/MOSDOp.h"
@@ -475,6 +476,9 @@ Message *decode_message(CephContext *cct, int crcflags,
   case MSG_OSD_MARK_ME_DOWN:
     m = make_message<MOSDMarkMeDown>();
     break;
+  case MSG_OSD_MARK_ME_DEAD:
+    m = make_message<MOSDMarkMeDead>();
+    break;
   case MSG_OSD_FULL:
     m = make_message<MOSDFull>();
     break;

diff --git a/src/msg/Message.h b/src/msg/Message.h
index 87b4b263b686de3ab83857a32010ae191c4e7158..2a66b4a4b3523203bec140a0fbc347a70aaead05 100644 (file)
--- a/src/msg/Message.h
+++ b/src/msg/Message.h
@@ -71,6 +71,7 @@
 #define MSG_OSD_ALIVE        73
 #define MSG_OSD_MARK_ME_DOWN 74
 #define MSG_OSD_FULL         75
+#define MSG_OSD_MARK_ME_DEAD 123
 
 // removed right after luminous
 //#define MSG_OSD_SUBOP        76