osd/mon: fix upgrades for pg log hard limit

Signed-off-by: Neha Ojha <nojha@redhat.com>
(cherry picked from commit 6ceeab6e204286148a69eb147fbc4045beddef49)

Conflicts:
	src/include/rados.h
	src/mon/MonCommands.h
	src/mon/OSDMonitor.cc
	src/osd/OSDMap.cc
          Luminous does not have CEPH_OSDMAP_NOSNAPTRIM flag.
          In nautilus, CEPH_OSDMAP_PGLOG_HARDLIMIT is set by default,
          which is not the case in luminous.

diff --git a/src/include/rados.h b/src/include/rados.h
index e91236a011c47dab77d52767c7ac9cfb2f4f181b..040a534d9a4e5ce73f10b3e0d37d46ef7e8a0207 100644 (file)
--- a/src/include/rados.h
+++ b/src/include/rados.h
@@ -157,6 +157,7 @@ extern const char *ceph_osd_state_name(int s);
 #define CEPH_OSDMAP_REQUIRE_LUMINOUS (1<<18) /* require l for booting osds */
 #define CEPH_OSDMAP_RECOVERY_DELETES (1<<19) /* deletes performed during recovery instead of peering */
 #define CEPH_OSDMAP_PURGED_SNAPDIRS  (1<<20) /* osds have converted snapsets */
+#define CEPH_OSDMAP_PGLOG_HARDLIMIT  (1<<22) /* put a hard limit on pg log length */
 
 /* these are hidden in 'ceph status' view */
 #define CEPH_OSDMAP_SEMIHIDDEN_FLAGS (CEPH_OSDMAP_REQUIRE_JEWEL|       \

diff --git a/src/mon/MonCommands.h b/src/mon/MonCommands.h
index c147e3679d08d88601bddc1fca264fa27a264bb4..6890af25a0bc5b0e33ca4c74513e07499db0890d 100644 (file)
--- a/src/mon/MonCommands.h
+++ b/src/mon/MonCommands.h
@@ -742,7 +742,7 @@ COMMAND("osd erasure-code-profile ls", \
        "list all erasure code profiles", \
        "osd", "r", "cli,rest")
 COMMAND("osd set " \
-       "name=key,type=CephChoices,strings=full|pause|noup|nodown|noout|noin|nobackfill|norebalance|norecover|noscrub|nodeep-scrub|notieragent|sortbitwise|recovery_deletes|require_jewel_osds|require_kraken_osds " \
+       "name=key,type=CephChoices,strings=full|pause|noup|nodown|noout|noin|nobackfill|norebalance|norecover|noscrub|nodeep-scrub|notieragent|sortbitwise|recovery_deletes|require_jewel_osds|require_kraken_osds||pglog_hardlimit " \
        "name=sure,type=CephChoices,strings=--yes-i-really-mean-it,req=false", \
        "set <key>", "osd", "rw", "cli,rest")
 COMMAND("osd unset " \

diff --git a/src/mon/OSDMonitor.cc b/src/mon/OSDMonitor.cc
index d8806679bde8bf1a27e64a55bbe9892504818aee..939f5b26535b89afd01f4377c0fe557152f71a90 100644 (file)
--- a/src/mon/OSDMonitor.cc
+++ b/src/mon/OSDMonitor.cc
@@ -2515,6 +2515,13 @@ bool OSDMonitor::preprocess_boot(MonOpRequestRef op)
                        << " because all up OSDs are post-hammer";
       goto ignore;
     }
+
+  if (osdmap.test_flag(CEPH_OSDMAP_PGLOG_HARDLIMIT) &&
+      !(m->osd_features & CEPH_FEATURE_OSD_PGLOG_HARDLIMIT)) {
+    mon->clog->info() << "disallowing boot of OSD "
+                     << m->get_orig_source_inst()
+                     << " because 'pglog_hardlimit' osdmap flag is set and OSD lacks the OSD_PGLOG_HARDLIMIT feature";
+    goto ignore;
   }
 
   // already booted?
@@ -9367,6 +9374,21 @@ bool OSDMonitor::prepare_command_impl(MonOpRequestRef op,
        err = -EPERM;
        goto reply;
       }
+    } else if (key == "pglog_hardlimit") {
+      if (!osdmap.get_num_up_osds() && !sure) {
+        ss << "Not advisable to continue since no OSDs are up. Pass "
+           << "--yes-i-really-mean-it if you really wish to continue.";
+        err = -EPERM;
+        goto reply;
+      }
+      if (HAVE_FEATURE(osdmap.get_up_osd_features(), OSD_PGLOG_HARDLIMIT)
+          || sure) {
+       return prepare_set_flag(op, CEPH_OSDMAP_PGLOG_HARDLIMIT);
+      } else {
+       ss << "not all up OSDs have OSD_PGLOG_HARDLIMIT feature";
+       err = -EPERM;
+       goto reply;
+      }
     } else if (key == "require_jewel_osds") {
       if (!osdmap.get_num_up_osds() && sure != "--yes-i-really-mean-it") {
         ss << "Not advisable to continue since no OSDs are up. Pass "

diff --git a/src/osd/OSDMap.cc b/src/osd/OSDMap.cc
index 8d47205b229c17fac12f9d0b682428ddfb5a490e..3b7ddfa4a40fab9f6ffd0320ca0a5817d7abcb97 100644 (file)
--- a/src/osd/OSDMap.cc
+++ b/src/osd/OSDMap.cc
@@ -3153,6 +3153,8 @@ string OSDMap::get_flag_string(unsigned f)
     s += ",recovery_deletes";
   if (f & CEPH_OSDMAP_PURGED_SNAPDIRS)
     s += ",purged_snapdirs";
+  if (f & CEPH_OSDMAP_PGLOG_HARDLIMIT)
+    s += ",pglog_hardlimit";
   if (s.length())
     s.erase(0, 1);
   return s;

diff --git a/src/osd/PG.h b/src/osd/PG.h
index e160b770287de50d74a0c08d6183fe867dc5096b..d782d897fedffa8c4d73a1483e8969e3d6237886 100644 (file)
--- a/src/osd/PG.h
+++ b/src/osd/PG.h
@@ -2508,6 +2508,10 @@ protected:
     return !(get_osdmap()->test_flag(CEPH_OSDMAP_RECOVERY_DELETES));
   }
 
+  bool hard_limit_pglog() const {
+    return (get_osdmap()->test_flag(CEPH_OSDMAP_PGLOG_HARDLIMIT));
+  }
+
   void init_primary_up_acting(
     const vector<int> &newup,
     const vector<int> &newacting,

diff --git a/src/osd/PrimaryLogPG.cc b/src/osd/PrimaryLogPG.cc
index 466f9b2c1cd5489934a4181b419094e655d6b748..0074c7964bf62ad3752a7a4baea5340f1e5b54be 100644 (file)
--- a/src/osd/PrimaryLogPG.cc
+++ b/src/osd/PrimaryLogPG.cc
@@ -3375,7 +3375,10 @@ void PrimaryLogPG::execute_ctx(OpContext *ctx)
   assert(op->may_write() || op->may_cache());
 
   // trim log?
-  calc_trim_to();
+  if (hard_limit_pglog())
+    calc_trim_to_aggressive();
+  else
+    calc_trim_to();
 
   // verify that we are doing this in order?
   if (cct->_conf->osd_debug_op_order && m->get_source().is_client() &&
@@ -9635,7 +9638,10 @@ void PrimaryLogPG::simple_opc_submit(OpContextUPtr ctx)
   dout(20) << __func__ << " " << repop << dendl;
   issue_repop(repop, ctx.get());
   eval_repop(repop);
-  calc_trim_to();
+  if (hard_limit_pglog())
+    calc_trim_to_aggressive();
+  else
+    calc_trim_to();
   repop->put();
 }
 
@@ -9777,7 +9783,10 @@ void PrimaryLogPG::submit_log_entries(
       assert(r == 0);
     });
 
-  calc_trim_to();
+  if (hard_limit_pglog())
+    calc_trim_to_aggressive();
+  else
+    calc_trim_to();
 }
 
 void PrimaryLogPG::cancel_log_updates()