import 15.2.5

[ceph.git] / ceph / src / mon / PGMap.cc

diff --git a/ceph/src/mon/PGMap.cc b/ceph/src/mon/PGMap.cc
index 636ab0c7156327814c384b2e1050a9a361226d66..4d5950fc48a0851e0b6e9851b429d28e33f4a688 100644 (file)
--- a/ceph/src/mon/PGMap.cc
+++ b/ceph/src/mon/PGMap.cc
@@ -958,6 +958,7 @@ void PGMapDigest::dump_object_stat_sum(
       f->dump_int("compress_under_bytes", statfs.data_compressed_original);
       // Stored by user amplified by replication
       f->dump_int("stored_raw", stored_raw);
+      f->dump_unsigned("avail_raw", avail);
     }
   } else {
     tbl << stringify(byte_u_t(stored_normalized));
@@ -1176,7 +1177,7 @@ void PGMap::apply_incremental(CephContext *cct, const Incremental& inc)
 
     auto pool_statfs_iter =
       pool_statfs.find(std::make_pair(update_pool, update_osd));
-    if (pg_pool_sum.count(update_pool)) { 
+    if (pg_pool_sum.count(update_pool)) {
       pool_stat_t &pool_sum_ref = pg_pool_sum[update_pool];
       if (pool_statfs_iter == pool_statfs.end()) {
         pool_statfs.emplace(std::make_pair(update_pool, update_osd), statfs_inc);
@@ -1212,6 +1213,13 @@ void PGMap::apply_incremental(CephContext *cct, const Incremental& inc)
     bool pool_erased = false;
     if (s != pg_stat.end()) {
       pool_erased = stat_pg_sub(removed_pg, s->second);
+
+      // decrease pool stats if pg was removed
+      auto pool_stats_it = pg_pool_sum.find(removed_pg.pool());
+      if (pool_stats_it != pg_pool_sum.end()) {
+        pool_stats_it->second.sub(s->second);
+      }
+
       pg_stat.erase(s);
       if (pool_erased) {
         deleted_pools.insert(removed_pg.pool());
@@ -2525,7 +2533,11 @@ void PGMap::get_health_checks(
         if (pg_response.stuck_since) {
           // Delayed response, check for stuckness
           utime_t last_whatever = pg_response.stuck_since(pg_info);
-          if (last_whatever >= cutoff) {
+          if (last_whatever.is_zero() &&
+            pg_info.last_change >= cutoff) {
+            // still moving, ignore
+            continue;
+          } else if (last_whatever >= cutoff) {
             // Not stuck enough, ignore.
             continue;
           } else {
@@ -2823,6 +2835,7 @@ void PGMap::get_health_checks(
 
     list<string> detail_back;
     list<string> detail_front;
+    list<string> detail;
     set<mon_ping_item_t> back_sorted, front_sorted;
     for (auto i : osd_stat) {
       for (auto j : i.second.hb_pingtime) {
@@ -2853,6 +2866,19 @@ void PGMap::get_health_checks(
          front_sorted.emplace(front);
        }
       }
+      if (i.second.num_shards_repaired >
+                     cct->_conf.get_val<uint64_t>("mon_osd_warn_num_repaired")) {
+        ostringstream ss;
+       ss << "osd." << i.first << " had " << i.second.num_shards_repaired << " reads repaired";
+        detail.push_back(ss.str());
+      }
+    }
+    if (!detail.empty()) {
+      ostringstream ss;
+      ss << "Too many repaired reads on " << detail.size() << " OSDs";
+      auto& d = checks->add("OSD_TOO_MANY_REPAIRS", HEALTH_WARN, ss.str(),
+                     detail.size());
+      d.detail.swap(detail);
     }
     int max_detail = 10;
     for (auto &sback : boost::adaptors::reverse(back_sorted)) {