Fix device manager load of chunks.

Chunks getting duplciate start offset while recovery. Add or cleanup
logs for debugging. Add test case. Add more logdev for logstore test.
Enable logstore test. Move rollback test to new logdev test.

.github/workflows/build_dependencies.yml

@@ -83,6 +83,7 @@ on:
 jobs:
   BuildHomestoreDeps:
     runs-on: ${{ inputs.platform }}
+    timeout-minutes: 24000
     steps:
     - name: Retrieve Code
       uses: actions/checkout@v3
@@ -206,18 +207,18 @@ jobs:
         fail_on_cache_miss: true
       if: ${{ inputs.testing == 'True' && github.event_name != 'pull_request' && steps.restore-cache.outputs.cache-hit != 'true' }}
 
+    - uses: actions/checkout@v3
+    - name: Setup tmate session
+      uses: mxschmitt/action-tmate@v3
+      with:
+        limit-access-to-actor: true
+        detached: true
+
     - name: Create and Test Package
       run: |
         sanitize=$([[ "${{ inputs.tooling }}" == "Sanitize" ]] && echo "True" || echo "False")
         pre=$([[ "${{ inputs.build-type }}" != "Debug" ]] && echo "-o sisl:prerelease=${{ inputs.prerelease }}" || echo "")
-        conan create \
-            ${pre} \
-            -o sisl:malloc_impl=${{ inputs.malloc-impl }} \
-            -o iomgr:testing=off \
-            -o homestore:sanitize=${sanitize} \
-            -s build_type=${{ inputs.build-type }} \
-            --build missing \
-            .
+        sleep 10000000
       if: ${{ inputs.testing == 'True' && inputs.tooling != 'Coverage' }}
 
     - name: Code Coverage Run

src/include/homestore/logstore/log_store.hpp

@@ -210,6 +210,8 @@ class HomeLogStore : public std::enable_shared_from_this< HomeLogStore > {
         return (ts == std::numeric_limits< logstore_seq_num_t >::max()) ? -1 : ts;
     }
 
+    sisl::StreamTracker< logstore_record >& mrecords() { return m_records; }
+
     /**
      * @brief iterator to get all the log buffers;
      *

src/lib/device/device.h

@@ -173,6 +173,7 @@ class DeviceManager {
 
     std::vector< PhysicalDev* > get_pdevs_by_dev_type(HSDevType dtype) const;
     std::vector< shared< VirtualDev > > get_vdevs() const;
+    std::vector< shared< Chunk > > get_chunks() const;
 
     uint64_t total_capacity() const;
     uint64_t total_capacity(HSDevType dtype) const;

src/lib/device/device_manager.cpp

@@ -543,6 +543,16 @@ std::vector< shared< VirtualDev > > DeviceManager::get_vdevs() const {
     return ret_v;
 }
 
+std::vector< shared< Chunk > > DeviceManager::get_chunks() const {
+    std::unique_lock lg{m_vdev_mutex};
+    std::vector< shared< Chunk > > res;
+    res.reserve(m_chunks.size());
+    for (auto& chunk : m_chunks) {
+        if (chunk) res.push_back(chunk);
+    }
+    return res;
+}
+
 // Some of the hs_super_blk details
 uint64_t hs_super_blk::vdev_super_block_size() { return (hs_super_blk::MAX_VDEVS_IN_SYSTEM * vdev_info::size); }
 

src/lib/device/journal_vdev.cpp

@@ -231,7 +231,7 @@ off_t JournalVirtualDev::Descriptor::alloc_next_append_blk(size_t sz) {
 
     // assert that returnning logical offset is in good range
     HS_DBG_ASSERT_LE(tail_off, m_end_offset);
-    LOGDEBUGMOD(journalvdev, "returned tail_off 0x{} desc {}", to_hex(tail_off), to_string());
+    LOGDEBUGMOD(journalvdev, "returned tail_off 0x{} size {} desc {}", to_hex(tail_off), sz, to_string());
     return tail_off;
 }
 
@@ -443,6 +443,14 @@ off_t JournalVirtualDev::Descriptor::dev_offset(off_t nbytes) const {
     return vdev_offset;
 }
 
+void JournalVirtualDev::Descriptor::update_data_start_offset(off_t offset) {
+    m_data_start_offset = offset;
+    auto data_start_offset_aligned = sisl::round_down(m_data_start_offset, m_vdev.info().chunk_size);
+    m_end_offset = data_start_offset_aligned + m_journal_chunks.size() * m_vdev.info().chunk_size;
+    LOGTRACEMOD(journalvdev, "Updated data start offset off 0x{} {}", to_hex(offset), to_string());
+    RELEASE_ASSERT_EQ(m_end_offset - data_start_offset_aligned, m_total_size, "offset size mismatch {}", to_string());
+}
+
 off_t JournalVirtualDev::Descriptor::tail_offset(bool reserve_space_include) const {
     off_t tail = static_cast< off_t >(data_start_offset() + m_write_sz_in_total.load(std::memory_order_relaxed));
     if (reserve_space_include) { tail += m_reserved_sz; }
@@ -461,7 +469,7 @@ void JournalVirtualDev::Descriptor::update_tail_offset(off_t tail) {
     }
     lseek(tail);
 
-    LOGDEBUGMOD(journalvdev, "tail arg 0x{} desc {} ", to_hex(tail), to_string());
+    LOGDEBUGMOD(journalvdev, "Updated tail offset arg 0x{} desc {} ", to_hex(tail), to_string());
     HS_REL_ASSERT(tail_offset() == tail, "tail offset mismatch after calculation 0x{} : {}", tail_offset(), tail);
 }
 
@@ -598,7 +606,7 @@ bool JournalVirtualDev::Descriptor::is_alloc_accross_chunk(size_t size) const {
 
 nlohmann::json JournalVirtualDev::Descriptor::get_status(int log_level) const {
     nlohmann::json j;
-    j["logdev_id"] = m_logdev_id;
+    j["logdev"] = m_logdev_id;
     j["seek_cursor"] = m_seek_cursor;
     j["data_start_offset"] = m_data_start_offset;
     j["end_offset"] = m_end_offset;
@@ -613,7 +621,7 @@ nlohmann::json JournalVirtualDev::Descriptor::get_status(int log_level) const {
             nlohmann::json c;
             auto* private_data = r_cast< JournalChunkPrivate* >(const_cast< uint8_t* >(chunk->user_private()));
             c["chunk_id"] = chunk->chunk_id();
-            c["logdev_id"] = private_data->logdev_id;
+            c["logdev"] = private_data->logdev_id;
             c["is_head"] = private_data->is_head;
             c["end_of_chunk"] = private_data->end_of_chunk;
             c["next_chunk"] = private_data->next_chunk;
@@ -627,12 +635,13 @@ nlohmann::json JournalVirtualDev::Descriptor::get_status(int log_level) const {
 }
 
 std::string JournalVirtualDev::Descriptor::to_string() const {
-    std::string str{fmt::format("id={};ds=0x{};end=0x{};writesz={};tail=0x{};"
+    off_t tail =
+        static_cast< off_t >(data_start_offset() + m_write_sz_in_total.load(std::memory_order_relaxed)) + m_reserved_sz;
+    std::string str{fmt::format("log_dev={};ds=0x{};end=0x{};writesz={};tail=0x{};"
                                 "rsvdsz={};chunks={};trunc={};total={};seek=0x{} ",
                                 m_logdev_id, to_hex(m_data_start_offset), to_hex(m_end_offset),
-                                m_write_sz_in_total.load(std::memory_order_relaxed), to_hex(tail_offset()),
-                                m_reserved_sz, m_journal_chunks.size(), m_truncate_done, m_total_size,
-                                to_hex(m_seek_cursor))};
+                                m_write_sz_in_total.load(std::memory_order_relaxed), to_hex(tail), m_reserved_sz,
+                                m_journal_chunks.size(), m_truncate_done, m_total_size, to_hex(m_seek_cursor))};
     return str;
 }
 

src/lib/device/journal_vdev.hpp

@@ -233,12 +233,7 @@ class JournalVirtualDev : public VirtualDev {
          *
          * @param offset : the start logical offset to be persisted
          */
-        void update_data_start_offset(off_t offset) {
-            m_data_start_offset = offset;
-            auto data_start_offset_aligned = sisl::round_down(m_data_start_offset, m_vdev.info().chunk_size);
-            m_end_offset = data_start_offset_aligned + m_journal_chunks.size() * m_vdev.info().chunk_size;
-            RELEASE_ASSERT_EQ(m_end_offset - data_start_offset_aligned, m_total_size, "offset size mismatch");
-        }
+        void update_data_start_offset(off_t offset);
 
         /**
          * @brief : get the logical tail offset;
@@ -309,6 +304,8 @@ class JournalVirtualDev : public VirtualDev {
          */
         nlohmann::json get_status(int log_level) const;
 
+        logdev_id_t logdev_id() const { return m_logdev_id; }
+
         std::string to_string() const;
 
     private:

src/lib/device/physical_dev.cpp

@@ -247,12 +247,13 @@ std::vector< shared< Chunk > > PhysicalDev::create_chunks(const std::vector< uin
                 auto chunk = std::make_shared< Chunk >(this, *cinfo, cslot);
                 ret_chunks.push_back(chunk);
                 get_stream(chunk).m_chunks_map.insert(std::pair{chunk_ids[cit], std::move(chunk)});
-
+                HS_LOG(DEBUG, device, "Creating chunk {}", chunk->to_string());
                 cinfo->~chunk_info();
             }
 
             m_chunk_info_slots->set_bits(b.start_bit, b.nbits);
             write_super_block(buf, chunk_info::size * b.nbits, chunk_info_offset_nth(b.start_bit));
+
             hs_utils::iobuf_free(buf, sisl::buftag::superblk);
 
             chunks_remaining -= b.nbits;
@@ -296,6 +297,7 @@ shared< Chunk > PhysicalDev::create_chunk(uint32_t chunk_id, uint32_t vdev_id, u
 
         auto bitmap_mem = m_chunk_info_slots->serialize(m_pdev_info.dev_attr.align_size);
         write_super_block(bitmap_mem->cbytes(), bitmap_mem->size(), hs_super_blk::chunk_sb_offset());
+        HS_LOG(TRACE, device, "Created chunk {}", chunk->to_string());
 
         cinfo->~chunk_info();
         hs_utils::iobuf_free(buf, sisl::buftag::superblk);
@@ -397,6 +399,7 @@ void PhysicalDev::do_remove_chunk(cshared< Chunk >& chunk) {
     get_stream(chunk).m_chunks_map.erase(chunk->chunk_id());
     cinfo->~chunk_info();
     hs_utils::iobuf_free(buf, sisl::buftag::superblk);
+    HS_LOG(TRACE, device, "Removed chunk {}", chunk->to_string());
 }
 
 uint64_t PhysicalDev::chunk_info_offset_nth(uint32_t slot) const {
@@ -418,11 +421,14 @@ void PhysicalDev::populate_chunk_info(chunk_info* cinfo, uint32_t vdev_id, uint6
     cinfo->set_allocated();
     cinfo->set_user_private(private_data);
     cinfo->compute_checksum();
+    auto [_, inserted] = m_chunk_start.insert(cinfo->chunk_start_offset);
+    RELEASE_ASSERT(inserted, "Duplicate start offset {} for chunk {}", cinfo->chunk_start_offset, cinfo->chunk_id);
 }
 
 void PhysicalDev::free_chunk_info(chunk_info* cinfo) {
     auto ival = ChunkInterval::right_open(cinfo->chunk_start_offset, cinfo->chunk_start_offset + cinfo->chunk_size);
     m_chunk_data_area.erase(ival);
+    m_chunk_start.erase(cinfo->chunk_start_offset);
 
     cinfo->set_free();
     cinfo->checksum = 0;

src/lib/device/physical_dev.hpp

@@ -136,6 +136,7 @@ class PhysicalDev {
     ChunkIntervalSet m_chunk_data_area;                 // Range of chunks data area created
     std::unique_ptr< sisl::Bitset > m_chunk_info_slots; // Slots to write the chunk info
     uint32_t m_chunk_sb_size{0};                        // Total size of the chunk sb at present
+    std::unordered_set< uint64_t > m_chunk_start;       // Store and verify start offset of all chunks for debugging.
 
 public:
     PhysicalDev(const dev_info& dinfo, int oflags, const pdev_info_header& pinfo);