From: Matan Breizman <mbreizma@redhat.com>
Date: Wed, 30 Aug 2023 08:57:18 +0000 (+0000)
Subject: osd/OSD: introduce trim_stale_maps
X-Git-Tag: testing/wip-batrick-testing-20240411.154038~768^2~1
X-Git-Url: http://git.apps.os.sepia.ceph.com/?a=commitdiff_plain;h=171da5b5cb75022a27af6e77a74c752f9247ed9c;p=ceph-ci.git

osd/OSD: introduce trim_stale_maps

```
  /**
   * trim_stale_maps
   *
   * trim_maps had a possible (rare) leak which resulted in stale osdmaps.
   * This method will cleanup any existing osdmaps from the store
   * with an epoch earlier than the superblock's oldest_map epoch.
   * See: https://tracker.ceph.com/issues/61962
   */
```

Signed-off-by: Matan Breizman <mbreizma@redhat.com>
---

diff --git a/src/osd/OSD.cc b/src/osd/OSD.cc
index a79b388deca..a14d70605f8 100644
--- a/src/osd/OSD.cc
+++ b/src/osd/OSD.cc
@@ -3078,6 +3078,19 @@ will start to track new ops received afterwards.";
       f, false, logger, counter);
   }
 
+  else if (prefix == "trim stale osdmaps") {
+    // osd_lock is not taken since trimming won't go past
+    // the superblock's oldest_map. The superblock won't
+    // be updated. Only the stored stale (no longer referenced)
+    // osdmaps are removed.
+    int ret = trim_stale_maps();
+    if (ret < 0) {
+     ss << " Error trimming stale osdmaps: " << cpp_strerror(ret);
+     goto out;
+    }
+    dout(20) << fmt::format("Trimmed {} osdmaps", ret) << dendl;
+  }
+
   else if (prefix == "cache drop") {
     lock_guard l(osd_lock);
     dout(20) << "clearing all caches" << dendl;
@@ -4303,6 +4316,12 @@ void OSD::final_init()
     asok_hook,
     "reset pg recovery statistics");
   ceph_assert(r == 0);
+  r = admin_socket->register_command(
+    "trim stale osdmaps",
+    asok_hook,
+    "cleanup any existing osdmap from the store "
+    "in the range of 0 up to the superblock's oldest_map.");
+  ceph_assert(r == 0);
   r = admin_socket->register_command(
     "cache drop",
     asok_hook,
@@ -7877,6 +7896,57 @@ void OSD::trim_maps(epoch_t oldest)
   ceph_assert(min <= service.map_cache.cached_key_lower_bound());
 }
 
+std::optional<epoch_t> OSD::get_epoch_from_osdmap_object(const ghobject_t& osdmap) {
+  const auto& name = osdmap.hobj.oid.name;
+  string osdmap_prefix = "osdmap.";
+  auto osdmap_pos = name.find(osdmap_prefix);
+  if (osdmap_pos == string::npos) {
+    return std::nullopt;
+  }
+  auto osdmap_string = name.substr(osdmap_pos + osdmap_prefix.size());
+  return stoul(osdmap_string);
+}
+
+int OSD::trim_stale_maps()
+{
+  int num_removed = 0;
+  vector<ghobject_t> objects;
+  int r = store->collection_list(
+      service.meta_ch,
+      ghobject_t{},
+      ghobject_t::get_max(),
+      INT_MAX,
+      &objects,
+      NULL);
+  if (r != 0) {
+    derr << __func__ << " list collection " << service.meta_ch
+         << " got: " << cpp_strerror(r) << dendl;
+    return r;
+  }
+  ObjectStore::Transaction t;
+  for (const auto& osdmap_obj : objects) {
+    if (auto epoch = get_epoch_from_osdmap_object(osdmap_obj);
+        epoch.has_value() && epoch < superblock.get_oldest_map()) {
+      dout(20) << __func__ << " removing stale osdmap epoch "
+               << epoch << dendl;
+      t.remove(coll_t::meta(), osdmap_obj);
+    }
+    if (t.get_num_ops() > cct->_conf->osd_target_transaction_size) {
+      num_removed += t.get_num_ops();
+      int tr = store->queue_transaction(service.meta_ch, t.claim_and_reset(), nullptr);
+      ceph_assert(tr == 0);
+    }
+  }
+
+  if (t.get_num_ops() > 0) {
+    num_removed += t.get_num_ops();
+    int tr = store->queue_transaction(service.meta_ch, t.claim_and_reset(), nullptr);
+    ceph_assert(tr == 0);
+  }
+
+  return num_removed;
+}
+
 void OSD::handle_osd_map(MOSDMap *m)
 {
   // wait for pgs to catch up
diff --git a/src/osd/OSD.h b/src/osd/OSD.h
index a6b7a3c31ed..83aed8a410b 100644
--- a/src/osd/OSD.h
+++ b/src/osd/OSD.h
@@ -1672,6 +1672,18 @@ protected:
   void note_up_osd(int osd);
   friend struct C_OnMapCommit;
 
+  std::optional<epoch_t> get_epoch_from_osdmap_object(const ghobject_t& osdmap);
+  /**
+   * trim_stale_maps
+   *
+   * trim_maps had a possible (rare) leak which resulted in stale osdmaps.
+   * This method will cleanup any existing osdmap from the store
+   * in the range of 0 up to the superblock's oldest_map.
+   * @return number of stale osdmaps which were removed.
+   * See: https://tracker.ceph.com/issues/61962
+   */
+  int trim_stale_maps();
+
   bool advance_pg(
     epoch_t advance_to,
     PG *pg,