os/bluestore: ExtentMap: request reshard range from update() intelligently

Split ourselves, or merge with our immediate predecessor or
successor.

Signed-off-by: Sage Weil <sage@redhat.com>

diff --git a/src/os/bluestore/BlueStore.cc b/src/os/bluestore/BlueStore.cc
index a9929acc6d8a32a60a8cb5b5e28cd2d0cc9d822e..e1cb30946f0d62794a5be4c0e8dc869a11761ba9 100644 (file)
--- a/src/os/bluestore/BlueStore.cc
+++ b/src/os/bluestore/BlueStore.cc
@@ -1793,6 +1793,7 @@ void BlueStore::ExtentMap::update(KeyValueDB::Transaction t,
 
     auto p = shards.begin();
     unsigned pos = 0;
+    auto prev_p = p;
     while (p != shards.end()) {
       auto n = p;
       ++n;
@@ -1815,19 +1816,38 @@ void BlueStore::ExtentMap::update(KeyValueDB::Transaction t,
 
        dout(20) << __func__ << " shard 0x" << std::hex
                 << p->offset << std::dec << " is " << len
-                << " bytes (was " << p->shard_info->bytes << ") from " << nn
-                << " extents" << dendl;
-
-        // indicate need for reshard if force mode selected, len >
-        // shard_max size OR non-last shard size is below the min
-        // threshold. The last check is to avoid potential unneeded
-        // reshardings since this might happen permanently.
-        if (!force &&
-             (len > cct->_conf->bluestore_extent_map_shard_max_size ||
-               (n != shards.end() &&
-               len < g_conf->bluestore_extent_map_shard_min_size))) {
-         request_reshard(p->offset, endoff);
-          return;
+                << " bytes (was " << p->shard_info->bytes << ") from "
+                << p->extents << " extents" << dendl;
+
+        if (!force) {
+         if (len > cct->_conf->bluestore_extent_map_shard_max_size) {
+           // we are big; reshard ourselves
+           request_reshard(p->offset, endoff);
+           return;
+         }
+         // avoid resharding the trailing shard, even if it is small
+         if (n != shards.end() &&
+             len < g_conf->bluestore_extent_map_shard_min_size) {
+           // we are small; combine with a neighbor
+           if (p == shards.begin() && endoff == OBJECT_MAX_SIZE) {
+             // we are an only shard
+             request_reshard(0, OBJECT_MAX_SIZE);
+           } else if (p == shards.begin()) {
+             // combine with next shard
+             request_reshard(p->offset, endoff + 1);
+           } else if (endoff == OBJECT_MAX_SIZE) {
+             // combine with previous shard
+             request_reshard(prev_p->offset, endoff);
+           } else {
+             // combine with the smaller of the two
+             if (prev_p->shard_info->bytes > n->shard_info->bytes) {
+               request_reshard(p->offset, endoff + 1);
+             } else {
+               request_reshard(prev_p->offset, endoff);
+             }
+           }
+           return;
+         }
         }
        assert(p->shard_info->offset == p->offset);
        p->shard_info->bytes = len;
@@ -1836,6 +1856,7 @@ void BlueStore::ExtentMap::update(KeyValueDB::Transaction t,
         dirty_shards.push_back(encoded_shards[pos]);
        p->dirty = false;
       }
+      prev_p = p;
       p = n;
       ++pos;
     }