Merge pull request #396 from eBay/jkou-dev

Rebase & merge V4 back to main

Author: koujl

.gitignore

@@ -2,6 +2,7 @@
 .idea/**
 .cache
 .clangd
+.claude/*
 
 build
 test_package/build

conanfile.py

@@ -10,7 +10,7 @@
 
 class HomeObjectConan(ConanFile):
     name = "homeobject"
-    version = "3.0.20"
+    version = "4.1.0"
 
     homepage = "https://github.com/eBay/HomeObject"
     description = "Blob Store built on HomeStore"

src/include/homeobject/blob_manager.hpp

@@ -43,7 +43,7 @@ class BlobManager : public Manager< BlobError > {
 public:
     virtual AsyncResult< blob_id_t > put(shard_id_t shard, Blob&&, trace_id_t tid = 0) = 0;
     virtual AsyncResult< Blob > get(shard_id_t shard, blob_id_t const& blob, uint64_t off = 0, uint64_t len = 0,
-                                    trace_id_t tid = 0) const = 0;
+                                    bool allow_skip_verify = false, trace_id_t tid = 0) const = 0;
     virtual NullAsyncResult del(shard_id_t shard, blob_id_t const& blob, trace_id_t tid = 0) = 0;
 };
 

src/include/homeobject/shard_manager.hpp

@@ -25,6 +25,7 @@ struct ShardError {
 };
 
 struct ShardInfo {
+    static constexpr uint64_t meta_length = 1024 + 1;
     enum class State : uint8_t {
         OPEN = 0,
         SEALED = 1,
@@ -40,6 +41,7 @@ struct ShardInfo {
     uint64_t available_capacity_bytes;
     uint64_t total_capacity_bytes;
     std::optional< peer_id_t > current_leader{std::nullopt};
+    uint8_t meta[meta_length]{};
 
     auto operator<=>(ShardInfo const& rhs) const { return id <=> rhs.id; }
     auto operator==(ShardInfo const& rhs) const { return id == rhs.id; }
@@ -55,7 +57,7 @@ class ShardManager : public Manager< ShardError > {
 
     virtual AsyncResult< ShardInfo > get_shard(shard_id_t id, trace_id_t tid = 0) const = 0;
     virtual AsyncResult< InfoList > list_shards(pg_id_t id, trace_id_t tid = 0) const = 0;
-    virtual AsyncResult< ShardInfo > create_shard(pg_id_t pg_owner, uint64_t size_bytes, trace_id_t tid = 0) = 0;
+    virtual AsyncResult< ShardInfo > create_shard(pg_id_t pg_owner, uint64_t size_bytes, std::string meta, trace_id_t tid = 0) = 0;
     virtual AsyncResult< ShardInfo > seal_shard(shard_id_t id, trace_id_t tid = 0) = 0;
 };
 

src/lib/blob_manager.cpp

@@ -5,11 +5,11 @@ namespace homeobject {
 std::shared_ptr< BlobManager > HomeObjectImpl::blob_manager() { return shared_from_this(); }
 
 BlobManager::AsyncResult< Blob > HomeObjectImpl::get(shard_id_t shard, blob_id_t const& blob_id, uint64_t off,
-                                                     uint64_t len, trace_id_t tid) const {
+                                                     uint64_t len, bool allow_skip_verify, trace_id_t tid) const {
     return _get_shard(shard, tid)
-        .thenValue([this, blob_id, off, len, tid](auto const e) -> BlobManager::AsyncResult< Blob > {
+        .thenValue([this, blob_id, off, len, allow_skip_verify, tid](auto const e) -> BlobManager::AsyncResult< Blob > {
             if (!e) return folly::makeUnexpected(BlobError(BlobErrorCode::UNKNOWN_SHARD));
-            return _get_blob(e.value(), blob_id, off, len, tid);
+            return _get_blob(e.value(), blob_id, off, len, allow_skip_verify, tid);
         });
 }
 

src/lib/homeobject_impl.hpp

@@ -88,12 +88,12 @@ class HomeObjectImpl : public HomeObject,
                        public std::enable_shared_from_this< HomeObjectImpl > {
 
     /// Implementation defines these
-    virtual ShardManager::AsyncResult< ShardInfo > _create_shard(pg_id_t, uint64_t size_bytes, trace_id_t tid) = 0;
+    virtual ShardManager::AsyncResult< ShardInfo > _create_shard(pg_id_t, uint64_t size_bytes, std::string meta, trace_id_t tid) = 0;
     virtual ShardManager::AsyncResult< ShardInfo > _seal_shard(ShardInfo const&, trace_id_t tid) = 0;
 
     virtual BlobManager::AsyncResult< blob_id_t > _put_blob(ShardInfo const&, Blob&&, trace_id_t tid) = 0;
     virtual BlobManager::AsyncResult< Blob > _get_blob(ShardInfo const&, blob_id_t, uint64_t off, uint64_t len,
-                                                       trace_id_t tid) const = 0;
+                                                       bool allow_skip_verify, trace_id_t tid) const = 0;
     virtual BlobManager::NullAsyncResult _del_blob(ShardInfo const&, blob_id_t, trace_id_t tid) = 0;
     ///
 
@@ -189,15 +189,15 @@ class HomeObjectImpl : public HomeObject,
 
     /// ShardManager
     ShardManager::AsyncResult< ShardInfo > get_shard(shard_id_t id, trace_id_t tid) const final;
-    ShardManager::AsyncResult< ShardInfo > create_shard(pg_id_t pg_owner, uint64_t size_bytes, trace_id_t tid) final;
+    ShardManager::AsyncResult< ShardInfo > create_shard(pg_id_t pg_owner, uint64_t size_bytes, std::string meta, trace_id_t tid) final;
     ShardManager::AsyncResult< InfoList > list_shards(pg_id_t pg, trace_id_t tid) const final;
     ShardManager::AsyncResult< ShardInfo > seal_shard(shard_id_t id, trace_id_t tid) final;
     uint64_t get_current_timestamp();
 
     /// BlobManager
     BlobManager::AsyncResult< blob_id_t > put(shard_id_t shard, Blob&&, trace_id_t tid) final;
     BlobManager::AsyncResult< Blob > get(shard_id_t shard, blob_id_t const& blob, uint64_t off, uint64_t len,
-                                         trace_id_t tid) const final;
+                                         bool allow_skip_verify, trace_id_t tid) const final;
     BlobManager::NullAsyncResult del(shard_id_t shard, blob_id_t const& blob, trace_id_t tid) final;
 };
 

src/lib/homestore_backend/gc_manager.cpp

@@ -122,7 +122,10 @@ void GCManager::start() {
         gc_actor->start();
         LOGINFOMOD(gcmgr, "start gc actor for pdev={}", pdev_id);
     }
+    start_gc_scan_timer();
+}
 
+void GCManager::start_gc_scan_timer() {
     const auto gc_scan_interval_sec = HS_BACKEND_DYNAMIC_CONFIG(gc_scan_interval_sec);
 
     // the initial idea here is that we want gc timer to run in a reactor that not shared with other fibers that
@@ -146,9 +149,7 @@ void GCManager::start() {
     LOGINFOMOD(gcmgr, "gc scheduler timer has started, interval is set to {} seconds", gc_scan_interval_sec);
 }
 
-bool GCManager::is_started() { return m_gc_timer_hdl != iomgr::null_timer_handle; }
-
-void GCManager::stop() {
+void GCManager::stop_gc_scan_timer() {
     if (m_gc_timer_hdl == iomgr::null_timer_handle) {
         LOGWARNMOD(gcmgr, "gc scheduler timer is not running, no need to stop it");
         return;
@@ -162,6 +163,10 @@ void GCManager::stop() {
         m_gc_timer_hdl = iomgr::null_timer_handle;
     });
     m_gc_timer_fiber = nullptr;
+}
+
+void GCManager::stop() {
+    stop_gc_scan_timer();
 
     for (const auto& [pdev_id, gc_actor] : m_pdev_gc_actors) {
         gc_actor->stop();
@@ -170,9 +175,20 @@ void GCManager::stop() {
 }
 
 folly::SemiFuture< bool > GCManager::submit_gc_task(task_priority priority, chunk_id_t chunk_id) {
-    if (!is_started()) return folly::makeFuture< bool >(false);
+    auto ex_vchunk = m_chunk_selector->get_extend_vchunk(chunk_id);
+    if (ex_vchunk == nullptr) {
+        LOGERRORMOD(gcmgr, "chunk {} not found when submit gc task!", chunk_id);
+        return folly::makeFuture< bool >(false);
+    }
+
+    // if the chunk has no garbage to be reclaimed, we don`t need to gc it , return true directly
+    const auto defrag_blk_num = ex_vchunk->get_defrag_nblks();
+    if (!defrag_blk_num && task_priority::normal == priority) {
+        LOGERRORMOD(gcmgr, "chunk {} has no garbage to be reclaimed, skip gc for this chunk!", chunk_id);
+        return folly::makeFuture< bool >(true);
+    }
 
-    auto pdev_id = m_chunk_selector->get_extend_vchunk(chunk_id)->get_pdev_id();
+    auto pdev_id = ex_vchunk->get_pdev_id();
     auto it = m_pdev_gc_actors.find(pdev_id);
     if (it == m_pdev_gc_actors.end()) {
         LOGINFOMOD(gcmgr, "pdev gc actor not found for pdev_id={}, chunk={}", pdev_id, chunk_id);
@@ -218,7 +234,7 @@ bool GCManager::is_eligible_for_gc(chunk_id_t chunk_id) {
 
     const auto total_blk_num = chunk->get_total_blks();
     const auto gc_garbage_rate_threshold = HS_BACKEND_DYNAMIC_CONFIG(gc_garbage_rate_threshold);
-    bool should_gc = 100 * defrag_blk_num >= total_blk_num * gc_garbage_rate_threshold;
+    bool should_gc = 100 * defrag_blk_num > total_blk_num * gc_garbage_rate_threshold;
 
     LOGDEBUGMOD(gcmgr,
                 "gc scan chunk_id={}, use_blks={}, available_blks={}, total_blks={}, defrag_blks={}, should_gc={}",
@@ -334,6 +350,11 @@ void GCManager::pdev_gc_actor::add_reserved_chunk(
 }
 
 folly::SemiFuture< bool > GCManager::pdev_gc_actor::add_gc_task(uint8_t priority, chunk_id_t move_from_chunk) {
+    if (m_is_stopped.load()) {
+        LOGWARNMOD(gcmgr, "pdev gc actor for pdev_id={} is not started yet or already stopped, cannot add gc task!",
+                   m_pdev_id);
+        return folly::makeSemiFuture< bool >(false);
+    }
     auto EXvchunk = m_chunk_selector->get_extend_vchunk(move_from_chunk);
     // it does not belong to any pg, so we don't need to gc it.
     if (!EXvchunk->m_pg_id.has_value()) {

src/lib/homestore_backend/gc_manager.hpp

@@ -310,7 +310,11 @@ class GCManager {
 
     void start();
     void stop();
-    bool is_started();
+
+    // the following two functions should not be called concurrently. if we need to call them concurrently, we need to
+    // add lock to protect
+    void start_gc_scan_timer();
+    void stop_gc_scan_timer();
 
     void scan_chunks_for_gc();
     void drain_pg_pending_gc_task(const pg_id_t pg_id);