osd/mon: fix upgrades for pg log hard limit

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

Conflicts:
	src/mon/MonCommands.h: trivial resolution

diff --git a/src/include/rados.h b/src/include/rados.h
index 3c555a7032fd57f970dc57565294421287d15b8f..74dd86b29acbfd9d836b8807e08aad1e5d31b80d 100644 (file)
--- a/src/include/rados.h
+++ b/src/include/rados.h
@@ -159,6 +159,7 @@ extern const char *ceph_osd_state_name(int s);
 #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_NOSNAPTRIM       (1<<21) /* disable snap trimming */
+#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 11f430f156e06663ffad667776d44cff94f348c8..d3a9b997e2a5450960a33ab23a1a508817eb5856 100644 (file)
--- a/src/mon/MonCommands.h
+++ b/src/mon/MonCommands.h
@@ -739,7 +739,7 @@ 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|nosnaptrim|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|nosnaptrim|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 16a3f7bd638a09d82ebc69bb333f220403967282..d218b28e46070e902739f102f5f208f1a2ee553d 100644 (file)
--- a/src/mon/OSDMonitor.cc
+++ b/src/mon/OSDMonitor.cc
@@ -2734,6 +2734,14 @@ bool OSDMonitor::preprocess_boot(MonOpRequestRef op)
     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?
   if (osdmap.is_up(from) &&
       osdmap.get_inst(from) == m->get_orig_source_inst() &&
@@ -9570,6 +9578,24 @@ bool OSDMonitor::prepare_command_impl(MonOpRequestRef op,
        err = -EPERM;
        goto reply;
       }
+    } else if (key == "pglog_hardlimit") {
+      if (!osdmap.get_num_up_osds() && sure != "--yes-i-really-mean-it") {
+        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;
+      }
+      // The release check here is required because for OSD_PGLOG_HARDLIMIT,
+      // we are reusing a jewel feature bit that was retired in luminous.
+      if (osdmap.require_osd_release >= CEPH_RELEASE_LUMINOUS &&
+         (HAVE_FEATURE(osdmap.get_up_osd_features(), OSD_PGLOG_HARDLIMIT)
+          || sure == "--yes-i-really-mean-it")) {
+       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 65c7ea6295e414d1e5b1fbd382b2fed4492825bf..3018b53510ff49db3899318b2874979194bcf6b0 100644 (file)
--- a/src/osd/OSDMap.cc
+++ b/src/osd/OSDMap.cc
@@ -2015,6 +2015,12 @@ int OSDMap::apply_incremental(const Incremental &inc)
     }
   }
 
+  if (inc.new_require_osd_release >= 0) {
+    require_osd_release = inc.new_require_osd_release;
+    if (require_osd_release >= CEPH_RELEASE_NAUTILUS) {
+      flags |= CEPH_OSDMAP_PGLOG_HARDLIMIT;
+    }
+  }
   // do new crush map last (after up/down stuff)
   if (inc.crush.length()) {
     bufferlist bl(inc.crush);
@@ -2959,6 +2965,9 @@ void OSDMap::decode(bufferlist::iterator& bl)
     if (struct_v >= 5) {
       decode(require_min_compat_client, bl);
       decode(require_osd_release, bl);
+      if (require_osd_release >= CEPH_RELEASE_NAUTILUS) {
+       flags |= CEPH_OSDMAP_PGLOG_HARDLIMIT;
+      }
       if (require_osd_release >= CEPH_RELEASE_LUMINOUS) {
        flags &= ~(CEPH_OSDMAP_LEGACY_REQUIRE_FLAGS);
        flags |= CEPH_OSDMAP_RECOVERY_DELETES;
@@ -3277,6 +3286,8 @@ 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 36a8e5a1825bcee3d09db0464d73f12caeffbab9..91a2c9241e9fe8b944643087e38784efa908ad3f 100644 (file)
--- a/src/osd/PG.h
+++ b/src/osd/PG.h
@@ -2760,6 +2760,10 @@ 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 55301dd5d8a85ad8aeeb2c134c153124aa2ca8a3..44b9aca1e8d8514b890004b5bc659ea724d1a4a3 100644 (file)
--- a/src/osd/PrimaryLogPG.cc
+++ b/src/osd/PrimaryLogPG.cc
@@ -3975,7 +3975,10 @@ 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() &&
@@ -10500,7 +10503,10 @@ 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();
 }
 
@@ -10614,7 +10620,10 @@ void PrimaryLogPG::submit_log_entries(
       op_applied(info.last_update);
     });
 
-  calc_trim_to();
+  if (hard_limit_pglog())
+    calc_trim_to_aggressive();
+  else
+    calc_trim_to();
 }
 
 void PrimaryLogPG::cancel_log_updates()