Merge pull request #13888 from liewegas/wip-bluestore-dw

os/bluestore: fix deferred writes; improve flush Reviewed-by: Igor Fedotov <ifedotov@mirantis.com>
ceph · Mar 21, 2017 · 66b42be · 66b42be
2 parents 5444c9d + def1760
commit 66b42be
Show file tree

Hide file tree

Showing 14 changed files with 1,234 additions and 971 deletions.
diff --git a/src/common/Throttle.h b/src/common/Throttle.h
@@ -66,6 +66,13 @@ class Throttle {
    */
   int64_t get_max() const { return max.read(); }
 
+  /**
+   * return true if past midpoint
+   */
+  bool past_midpoint() const {
+    return count.read() >= max.read() / 2;
+  }
+
   /**
    * set the new max number, and wait until the number of taken slots drains
    * and drops below this limit.

diff --git a/src/common/config_opts.h b/src/common/config_opts.h
@@ -1051,9 +1051,9 @@ OPTION(bluestore_min_alloc_size, OPT_U32, 0)
 OPTION(bluestore_min_alloc_size_hdd, OPT_U32, 64*1024)
 OPTION(bluestore_min_alloc_size_ssd, OPT_U32, 4*1024)
 OPTION(bluestore_max_alloc_size, OPT_U32, 0)
-OPTION(bluestore_prefer_wal_size, OPT_U32, 0)
-OPTION(bluestore_prefer_wal_size_hdd, OPT_U32, 32768)
-OPTION(bluestore_prefer_wal_size_ssd, OPT_U32, 0)
+OPTION(bluestore_prefer_deferred_size, OPT_U32, 0)
+OPTION(bluestore_prefer_deferred_size_hdd, OPT_U32, 32768)
+OPTION(bluestore_prefer_deferred_size_ssd, OPT_U32, 0)
 OPTION(bluestore_compression_mode, OPT_STR, "none")  // force|aggressive|passive|none
 OPTION(bluestore_compression_algorithm, OPT_STR, "snappy")
 OPTION(bluestore_compression_min_blob_size, OPT_U32, 128*1024)
@@ -1104,14 +1104,11 @@ OPTION(bluestore_fsck_on_umount_deep, OPT_BOOL, true)
 OPTION(bluestore_fsck_on_mkfs, OPT_BOOL, true)
 OPTION(bluestore_fsck_on_mkfs_deep, OPT_BOOL, false)
 OPTION(bluestore_sync_submit_transaction, OPT_BOOL, false) // submit kv txn in queueing thread (not kv_sync_thread)
-OPTION(bluestore_sync_wal_apply, OPT_BOOL, true)     // perform initial wal work synchronously (possibly in combination with aio so we only *queue* ios)
-OPTION(bluestore_wal_threads, OPT_INT, 4)
-OPTION(bluestore_wal_thread_timeout, OPT_INT, 30)
-OPTION(bluestore_wal_thread_suicide_timeout, OPT_INT, 120)
 OPTION(bluestore_max_ops, OPT_U64, 512)
 OPTION(bluestore_max_bytes, OPT_U64, 64*1024*1024)
-OPTION(bluestore_wal_max_ops, OPT_U64, 512)
-OPTION(bluestore_wal_max_bytes, OPT_U64, 128*1024*1024)
+OPTION(bluestore_deferred_max_ops, OPT_U64, 512)
+OPTION(bluestore_deferred_max_bytes, OPT_U64, 128*1024*1024)
+OPTION(bluestore_deferred_batch_ops, OPT_U64, 8)
 OPTION(bluestore_nid_prealloc, OPT_INT, 1024)
 OPTION(bluestore_blobid_prealloc, OPT_U64, 10240)
 OPTION(bluestore_clone_cow, OPT_BOOL, true)  // do copy-on-write for clones
@@ -1126,7 +1123,6 @@ OPTION(bluestore_debug_prefragment_max, OPT_INT, 1048576)
 OPTION(bluestore_debug_inject_read_err, OPT_BOOL, false)
 OPTION(bluestore_debug_randomize_serial_transaction, OPT_INT, 0)
 OPTION(bluestore_debug_omit_block_device_write, OPT_BOOL, false)
-OPTION(bluestore_inject_wal_apply_delay, OPT_FLOAT, 0)
 OPTION(bluestore_shard_finishers, OPT_BOOL, false)
 
 OPTION(kstore_max_ops, OPT_U64, 512)

diff --git a/src/os/ObjectStore.h b/src/os/ObjectStore.h
@@ -134,8 +134,15 @@ class ObjectStore {
    */
   struct Sequencer_impl : public RefCountedObject {
     CephContext* cct;
+
+    // block until any previous transactions are visible.  specifically,
+    // collection_list and collection_empty need to reflect prior operations.
     virtual void flush() = 0;
 
+    // called when we are done with the impl.  the impl may have a different
+    // (longer) lifecycle than the Sequencer.
+    virtual void discard() {}
+
     /**
      * Async flush_commit
      *
@@ -165,8 +172,11 @@ class ObjectStore {
     Sequencer_implRef p;
 
     explicit Sequencer(string n)
-      : name(n), shard_hint(spg_t()), p(NULL) {}
+      : name(n), shard_hint(spg_t()), p(NULL) {
+    }
     ~Sequencer() {
+      if (p)
+	p->discard();  // tell impl we are done with it
     }
 
     /// return a unique string identifier for this sequencer