mon/OSDMonitor: warn when cache pools do not have hit_sets configured

Give users a clue when cache pools are enabled but the hit_set is not
configured.  Note that technically this will work, but not well, so for
now let's just steer them away.

Signed-off-by: Sage Weil <sage@redhat.com>
(cherry picked from commit 383536a2ef2506c5591b25c0fd61dc8c181fdab9)

NOTE: removed readforward state for backport

diff --git a/src/common/config_opts.h b/src/common/config_opts.h
index c0679b020c96b679b4faf70fb59232cbea08fac8..fdd35575dfe9a0dc5f5fadf72d3f2f257353847f 100644 (file)
--- a/src/common/config_opts.h
+++ b/src/common/config_opts.h
@@ -177,6 +177,7 @@ OPTION(mon_force_standby_active, OPT_BOOL, true) // should mons force standby-re
 OPTION(mon_warn_on_old_mons, OPT_BOOL, true) // should mons set health to WARN if part of quorum is old?
 OPTION(mon_warn_on_legacy_crush_tunables, OPT_BOOL, true) // warn if crush tunables are not optimal
 OPTION(mon_warn_on_osd_down_out_interval_zero, OPT_BOOL, true) // warn if 'mon_osd_down_out_interval == 0'
+OPTION(mon_warn_on_cache_pools_without_hit_sets, OPT_BOOL, true)
 OPTION(mon_min_osdmap_epochs, OPT_INT, 500)
 OPTION(mon_max_pgmap_epochs, OPT_INT, 500)
 OPTION(mon_max_log_epochs, OPT_INT, 500)

diff --git a/src/mon/OSDMonitor.cc b/src/mon/OSDMonitor.cc
index b58d1b938a8eb8fcb4292664a903ec6c2886de26..fbe03531837d76d1bba830212301cc6d2cd22a85 100644 (file)
--- a/src/mon/OSDMonitor.cc
+++ b/src/mon/OSDMonitor.cc
@@ -2067,6 +2067,32 @@ void OSDMonitor::get_health(list<pair<health_status_t,string> >& summary,
       }
     }
 
+    // hit_set-less cache_mode?
+    if (g_conf->mon_warn_on_cache_pools_without_hit_sets) {
+      int problem_cache_pools = 0;
+      for (map<int64_t, pg_pool_t>::const_iterator p = osdmap.pools.begin();
+          p != osdmap.pools.end();
+          ++p) {
+       const pg_pool_t& info = p->second;
+       if (info.cache_mode_requires_hit_set() &&
+           info.hit_set_params.get_type() == HitSet::TYPE_NONE) {
+         ++problem_cache_pools;
+         if (detail) {
+           ostringstream ss;
+           ss << "pool '" << osdmap.get_pool_name(p->first)
+              << "' with cache_mode " << info.get_cache_mode_name()
+              << " needs hit_set_type to be set but it is not";
+           detail->push_back(make_pair(HEALTH_WARN, ss.str()));
+         }
+       }
+      }
+      if (problem_cache_pools) {
+       ostringstream ss;
+       ss << problem_cache_pools << " cache pools are missing hit_sets";
+       summary.push_back(make_pair(HEALTH_WARN, ss.str()));
+      }
+    }
+
     // Warn if 'mon_osd_down_out_interval' is set to zero.
     // Having this option set to zero on the leader acts much like the
     // 'noout' flag.  It's hard to figure out what's going wrong with clusters

diff --git a/src/osd/osd_types.h b/src/osd/osd_types.h
index 086cc94f42bc5c57acdef20e3d30f9e606a47b94..b0ce28e2872d277ebccd82058ce364fd8d695d19 100644 (file)
--- a/src/osd/osd_types.h
+++ b/src/osd/osd_types.h
@@ -870,6 +870,18 @@ struct pg_pool_t {
   const char *get_cache_mode_name() const {
     return get_cache_mode_name(cache_mode);
   }
+  bool cache_mode_requires_hit_set() const {
+    switch (cache_mode) {
+    case CACHEMODE_NONE:
+    case CACHEMODE_FORWARD:
+    case CACHEMODE_READONLY:
+      return false;
+    case CACHEMODE_WRITEBACK:
+      return true;
+    default:
+      assert(0 == "implement me");
+    }
+  }
 
   uint64_t flags;           ///< FLAG_*
   __u8 type;                ///< TYPE_*