osd: send health-checks to mgr

Signed-off-by: Kefu Chai <kchai@redhat.com>

diff --git a/src/osd/OSD.cc b/src/osd/OSD.cc
index 97b829a53339e14546507d6a6aaaf3f3f02a9438..b9afa353d6e97d8d4d496f21e484ce89bcfdcfbb 100644 (file)
--- a/src/osd/OSD.cc
+++ b/src/osd/OSD.cc
@@ -5019,23 +5019,11 @@ void OSD::tick_without_osd_lock()
     }
   }
 
-  check_ops_in_flight();
+  mgrc.update_osd_health(get_health_metrics());
   service.kick_recovery_queue();
   tick_timer_without_osd_lock.add_event_after(OSD_TICK_INTERVAL, new C_Tick_WithoutOSDLock(this));
 }
 
-void OSD::check_ops_in_flight()
-{
-  string summary;
-  vector<string> warnings;
-  if (op_tracker.check_ops_in_flight(&summary, warnings)) {
-    clog->warn() << summary;
-    for (const auto& warning : warnings) {
-      clog->warn() << warning;
-    }
-  }
-}
-
 // Usage:
 //   setomapval <pool-id> [namespace/]<obj-name> <key> <val>
 //   rmomapkey <pool-id> [namespace/]<obj-name> <key>
@@ -7096,6 +7084,42 @@ MPGStats* OSD::collect_pg_stats()
   return m;
 }
 
+vector<OSDHealthMetric> OSD::get_health_metrics()
+{
+  vector<OSDHealthMetric> metrics;
+  {
+    utime_t oldest_secs;
+    const utime_t now = ceph_clock_now();
+    auto too_old = now;
+    too_old -= cct->_conf->get_val<double>("osd_op_complaint_time");
+    int slow = 0;
+    auto count_slow_ops = [&](TrackedOp& op) {
+      if (op.get_initiated() < too_old) {
+       slow++;
+       return true;
+      } else {
+       return false;
+      }
+    };
+    if (op_tracker.visit_ops_in_flight(&oldest_secs, count_slow_ops)) {
+      metrics.emplace_back(osd_metric::SLOW_OPS, slow, oldest_secs);
+    } else {
+      // no news is not good news.
+      metrics.emplace_back(osd_metric::SLOW_OPS, 0, 0);
+    }
+  }
+  with_unique_lock(pending_creates_lock, [&]() {
+      auto n_primaries = pending_creates_from_mon;
+      for (const auto& create : pending_creates_from_osd) {
+       if (create.second) {
+         n_primaries++;
+       }
+      }
+      metrics.emplace_back(osd_metric::PENDING_CREATING_PGS, n_primaries);
+    });
+  return metrics;
+}
+
 // =====================================================
 // MAP
 

diff --git a/src/osd/OSD.h b/src/osd/OSD.h
index c0591a4d1ffc14bb25f34fc6ad187401886b0e62..bc20926a590d1ca831c7b19fe98630fdaec9ee8c 100644 (file)
--- a/src/osd/OSD.h
+++ b/src/osd/OSD.h
@@ -1554,7 +1554,6 @@ private:
   
   // -- op tracking --
   OpTracker op_tracker;
-  void check_ops_in_flight();
   void test_ops(std::string command, std::string args, ostream& ss);
   friend class TestOpsSocketHook;
   TestOpsSocketHook *test_ops_hook;
@@ -2233,6 +2232,7 @@ protected:
 
   // -- status reporting --
   MPGStats *collect_pg_stats();
+  std::vector<OSDHealthMetric> get_health_metrics();
 
 private:
   bool ms_can_fast_dispatch_any() const override { return true; }

diff --git a/src/osd/OSDHealthMetric.h b/src/osd/OSDHealthMetric.h
new file mode 100644 (file)
index 0000000..ec16cff
--- /dev/null
+++ b/src/osd/OSDHealthMetric.h
@@ -0,0 +1,61 @@
+// -*- mode:C++; tab-width:8; c-basic-offset:2; indent-tabs-mode:t -*-
+// vim: ts=8 sw=2 smarttab
+
+#pragma once
+
+#include <cstdint>
+#include "include/denc.h"
+
+enum class osd_metric : uint8_t {
+  SLOW_OPS,
+  PENDING_CREATING_PGS,
+  NONE,
+};
+
+union osd_metric_t {
+  struct {
+    uint32_t n1;
+    uint32_t n2;
+  };
+  uint64_t n;
+  osd_metric_t(uint32_t x, uint32_t y)
+    : n1(x), n2(y)
+  {}
+  osd_metric_t(uint64_t x = 0)
+    : n(x)
+  {}
+};
+
+class OSDHealthMetric
+{
+public:
+  OSDHealthMetric() = default;
+  OSDHealthMetric(osd_metric type_, uint64_t n)
+    : type(type_), value(n)
+  {}
+  OSDHealthMetric(osd_metric type_, uint32_t n1, uint32_t n2)
+    : type(type_), value(n1, n2)
+  {}
+  osd_metric get_type() const {
+    return type;
+  }
+  uint64_t get_n() const {
+    return value.n;
+  }
+  uint32_t get_n1() const {
+    return value.n1;
+  }
+  uint32_t get_n2() const {
+    return value.n2;
+  }
+  DENC(OSDHealthMetric, v, p) {
+    DENC_START(1, 1, p);
+    denc(v.type, p);
+    denc(v.value.n, p);
+    DENC_FINISH(p);
+  }
+private:
+  osd_metric type = osd_metric::NONE;
+  osd_metric_t value;
+};
+WRITE_CLASS_DENC(OSDHealthMetric)