From: Yin Congmin <congmin.yin@intel.com>
Date: Mon, 8 Mar 2021 16:26:04 +0000 (+0800)
Subject: librbd/cache/pwl: set max size of continuous data
X-Git-Tag: v16.2.0~92^2~3
X-Git-Url: http://git-server-git.apps.pok.os.sepia.ceph.com/?a=commitdiff_plain;h=d34cb367742bc4c86197fd9293924ae001b61aaa;p=ceph.git

librbd/cache/pwl: set max size of continuous data

Signed-off-by: Yin Congmin <congmin.yin@intel.com>
(cherry picked from commit bcad92c126526be7ba249322ac3ead0d83b4d188)
---

diff --git a/src/librbd/cache/pwl/AbstractWriteLog.cc b/src/librbd/cache/pwl/AbstractWriteLog.cc
index 640fead2e189..b08d5df7962d 100644
--- a/src/librbd/cache/pwl/AbstractWriteLog.cc
+++ b/src/librbd/cache/pwl/AbstractWriteLog.cc
@@ -818,15 +818,44 @@ void AbstractWriteLog<I>::write(Extents &&image_extents,
 
   ceph_assert(m_initialized);
 
+  /* Split images because PMDK doesn't support allocating too big extent
+   * TODO: If the bluestore allocator is implemented as a library,
+   * the split operation is not necessary
+   */
+  Extents split_image_extents;
+  uint64_t max_extent_size = get_max_extent();
+  if (max_extent_size != 0) {
+    for (auto extent : image_extents) {
+      if (extent.second > max_extent_size) {
+        uint64_t off = extent.first;
+        uint64_t extent_bytes = extent.second;
+        for (int i = 0; extent_bytes != 0; ++i) {
+          Extent _ext;
+          _ext.first = off + i * max_extent_size;
+          _ext.second = std::min(max_extent_size, extent_bytes);
+          extent_bytes = extent_bytes - _ext.second ;
+          split_image_extents.emplace_back(_ext);
+        }
+      } else {
+        split_image_extents.emplace_back(extent);
+      }
+    }
+  } else {
+    split_image_extents = image_extents;
+  }
+
   C_WriteRequestT *write_req =
-    m_builder->create_write_request(*this, now, std::move(image_extents), std::move(bl),
-                                    fadvise_flags, m_lock, m_perfcounter, on_finish);
-  m_perfcounter->inc(l_librbd_pwl_wr_bytes, write_req->image_extents_summary.total_bytes);
+    m_builder->create_write_request(*this, now, std::move(split_image_extents),
+                                    std::move(bl), fadvise_flags, m_lock,
+                                    m_perfcounter, on_finish);
+  m_perfcounter->inc(l_librbd_pwl_wr_bytes,
+                      write_req->image_extents_summary.total_bytes);
 
   /* The lambda below will be called when the block guard for all
    * blocks affected by this write is obtained */
   GuardedRequestFunctionContext *guarded_ctx =
-    new GuardedRequestFunctionContext([this, write_req](GuardedRequestFunctionContext &guard_ctx) {
+    new GuardedRequestFunctionContext([this,
+      write_req](GuardedRequestFunctionContext &guard_ctx) {
       write_req->blockguard_acquired(guard_ctx);
       alloc_and_dispatch_io_req(write_req);
     });
diff --git a/src/librbd/cache/pwl/AbstractWriteLog.h b/src/librbd/cache/pwl/AbstractWriteLog.h
index 29b6c3b9c709..56ecd6c1c209 100644
--- a/src/librbd/cache/pwl/AbstractWriteLog.h
+++ b/src/librbd/cache/pwl/AbstractWriteLog.h
@@ -393,6 +393,10 @@ protected:
       const std::shared_ptr<pwl::GenericLogEntry> log_entry) {
     return nullptr;
   }
+  virtual uint64_t get_max_extent() {
+    return 0;
+  }
+
 };
 
 } // namespace pwl
diff --git a/src/librbd/cache/pwl/Types.h b/src/librbd/cache/pwl/Types.h
index e6d0e08a922f..5d438b2f11d0 100644
--- a/src/librbd/cache/pwl/Types.h
+++ b/src/librbd/cache/pwl/Types.h
@@ -169,7 +169,7 @@ const uint32_t LOG_STATS_INTERVAL_SECONDS = 5;
 /**** Write log entries ****/
 const unsigned long int MAX_ALLOC_PER_TRANSACTION = 8;
 const unsigned long int MAX_FREE_PER_TRANSACTION = 1;
-const unsigned int MAX_CONCURRENT_WRITES = 256;
+const unsigned int MAX_CONCURRENT_WRITES = (1024 * 1024);
 
 const uint64_t DEFAULT_POOL_SIZE = 1u<<30;
 const uint64_t MIN_POOL_SIZE = DEFAULT_POOL_SIZE;
@@ -182,7 +182,7 @@ const double RETIRE_HIGH_WATER = 0.50;
 const double RETIRE_LOW_WATER = 0.40;
 const int RETIRE_BATCH_TIME_LIMIT_MS = 250;
 const uint64_t CONTROL_BLOCK_MAX_LOG_ENTRIES = 32;
-const uint64_t SPAN_MAX_DATA_LEN = (16*1024*1024);
+const uint64_t SPAN_MAX_DATA_LEN = (16 * 1024 * 1024);
 
 /* offset of ring on SSD */
 const uint64_t DATA_RING_BUFFER_OFFSET = 8192;
diff --git a/src/librbd/cache/pwl/rwl/WriteLog.h b/src/librbd/cache/pwl/rwl/WriteLog.h
index dcdd9880aa89..7aea8573c917 100644
--- a/src/librbd/cache/pwl/rwl/WriteLog.h
+++ b/src/librbd/cache/pwl/rwl/WriteLog.h
@@ -58,6 +58,7 @@ private:
   PMEMobjpool *m_log_pool = nullptr;
   Builder<This> *m_builderobj;
   const char* m_pwl_pool_layout_name;
+  const uint64_t MAX_EXTENT_SIZE = 1048576;
 
   Builder<This>* create_builder();
   void remove_pool_file();
@@ -106,6 +107,9 @@ protected:
   void write_data_to_buffer(
       std::shared_ptr<pwl::WriteLogEntry> ws_entry,
       pwl::WriteLogCacheEntry *pmem_entry) override;
+  uint64_t get_max_extent() override {
+    return MAX_EXTENT_SIZE;
+  }
 };
 
 } // namespace rwl