handle missing pdev

eBay · May 8, 2024 · ca6db16 · ca6db16
1 parent 12390e7
commit ca6db16
Show file tree

Hide file tree

Showing 4 changed files with 65 additions and 4 deletions.
diff --git a/src/include/homestore/homestore_decl.hpp b/src/include/homestore/homestore_decl.hpp
@@ -83,6 +83,7 @@ static constexpr uint32_t INVALID_PDEV_ID{std::numeric_limits< uint32_t >::max()
 static constexpr uint32_t INVALID_VDEV_ID{std::numeric_limits< uint32_t >::max()};
 static constexpr uint32_t INVALID_CHUNK_ID{std::numeric_limits< uint32_t >::max()};
 static constexpr uint32_t INVALID_DEV_ID{std::numeric_limits< uint32_t >::max()};
+static constexpr uint64_t INVALID_DEV_OFFSET{std::numeric_limits< uint64_t >::max()};
 constexpr uint16_t MAX_UUID_LEN{128};
 static constexpr hs_uuid_t INVALID_SYSTEM_UUID{0};
 
@@ -177,7 +178,10 @@ struct hs_input_params {
     nlohmann::json to_json() const;
     std::string to_string() const { return to_json().dump(4); }
     uint64_t io_mem_size() const { return (hugepage_size != 0) ? hugepage_size : app_mem_size; }
-    bool has_fast_dev() const { return std::any_of(devices.begin(), devices.end(), [](const dev_info& d) { return d.dev_type == HSDevType::Fast; }); }
+    bool has_fast_dev() const {
+        return std::any_of(devices.begin(), devices.end(),
+                           [](const dev_info& d) { return d.dev_type == HSDevType::Fast; });
+    }
 };
 
 struct hs_engine_config {

diff --git a/src/lib/device/device.h b/src/lib/device/device.h
@@ -159,12 +159,16 @@ class DeviceManager {
 
     const Chunk* get_chunk(uint32_t chunk_id) const {
         std::unique_lock lg{m_vdev_mutex};
-        return (chunk_id == INVALID_CHUNK_ID) ? nullptr : m_chunks[chunk_id].get();
+        // if a pdev is misssing when restart, chunk_id from cleint might be larger than m_chunks.size()
+        if (chunk_id == INVALID_CHUNK_ID || chunk_id >= m_chunks.size()) return nullptr;
+        return m_chunks[chunk_id].get();
     }
 
     Chunk* get_chunk_mutable(uint32_t chunk_id) {
         std::unique_lock lg{m_vdev_mutex};
-        return (chunk_id == INVALID_CHUNK_ID) ? nullptr : m_chunks[chunk_id].get();
+        // if a pdev is misssing when restart, chunk_id from client might be larger than m_chunks.size()
+        if (chunk_id == INVALID_CHUNK_ID || chunk_id >= m_chunks.size()) return nullptr;
+        return m_chunks[chunk_id].get();
     }
 
     uint32_t atomic_page_size(HSDevType dtype) const;

diff --git a/src/lib/device/virtual_dev.cpp b/src/lib/device/virtual_dev.cpp
@@ -33,6 +33,7 @@
 #include <sisl/logging/logging.h>
 #include <sisl/utility/atomic_counter.hpp>
 #include <iomgr/iomgr_flip.hpp>
+#include <homestore/homestore_decl.hpp>
 
 #include "device/chunk.h"
 #include "device/physical_dev.hpp"
@@ -309,13 +310,19 @@ uint64_t VirtualDev::get_len(const iovec* iov, int iovcnt) {
     return len;
 }
 
+// for all writes functions, we don't expect to get invalid dev_offset, since we will never allocate blkid from missing
+// chunk(missing pdev);
 ////////////////////////// async write section //////////////////////////////////
 folly::Future< std::error_code > VirtualDev::async_write(const char* buf, uint32_t size, BlkId const& bid,
                                                          bool part_of_batch) {
     HS_DBG_ASSERT_EQ(bid.is_multi(), false, "async_write needs individual pieces of blkid - not MultiBlkid");
 
     Chunk* chunk;
     uint64_t const dev_offset = to_dev_offset(bid, &chunk);
+    if (sisl_unlikely(dev_offset == INVALID_DEV_OFFSET)) {
+        // TODO: define a new error code for missing pdev case;
+        return folly::makeFuture< std::error_code >(std::make_error_code(std::errc::resource_unavailable_try_again));
+    }
     auto* pdev = chunk->physical_dev_mutable();
 
     HS_LOG(TRACE, device, "Writing in device: {}, offset = {}", pdev->pdev_id(), dev_offset);
@@ -328,6 +335,9 @@ folly::Future< std::error_code > VirtualDev::async_write(const char* buf, uint32
 
 folly::Future< std::error_code > VirtualDev::async_write(const char* buf, uint32_t size, cshared< Chunk >& chunk,
                                                          uint64_t offset_in_chunk) {
+    if (sisl_unlikely(!is_chunk_loaded(chunk))) {
+        return folly::makeFuture< std::error_code >(std::make_error_code(std::errc::resource_unavailable_try_again));
+    }
     auto const dev_offset = chunk->start_offset() + offset_in_chunk;
     auto* pdev = chunk->physical_dev_mutable();
 
@@ -345,6 +355,9 @@ folly::Future< std::error_code > VirtualDev::async_writev(const iovec* iov, cons
 
     Chunk* chunk;
     uint64_t const dev_offset = to_dev_offset(bid, &chunk);
+    if (sisl_unlikely(dev_offset == INVALID_DEV_OFFSET)) {
+        return folly::makeFuture< std::error_code >(std::make_error_code(std::errc::resource_unavailable_try_again));
+    }
     auto const size = get_len(iov, iovcnt);
     auto* pdev = chunk->physical_dev_mutable();
 
@@ -358,6 +371,9 @@ folly::Future< std::error_code > VirtualDev::async_writev(const iovec* iov, cons
 
 folly::Future< std::error_code > VirtualDev::async_writev(const iovec* iov, const int iovcnt, cshared< Chunk >& chunk,
                                                           uint64_t offset_in_chunk) {
+    if (sisl_unlikely(!is_chunk_loaded(chunk))) {
+        return folly::makeFuture< std::error_code >(std::make_error_code(std::errc::resource_unavailable_try_again));
+    }
     auto const dev_offset = chunk->start_offset() + offset_in_chunk;
     auto const size = get_len(iov, iovcnt);
     auto* pdev = chunk->physical_dev_mutable();
@@ -376,11 +392,17 @@ std::error_code VirtualDev::sync_write(const char* buf, uint32_t size, BlkId con
 
     Chunk* chunk;
     uint64_t const dev_offset = to_dev_offset(bid, &chunk);
+    if (sisl_unlikely(dev_offset == INVALID_DEV_OFFSET)) {
+        return std::make_error_code(std::errc::resource_unavailable_try_again);
+    }
     return chunk->physical_dev_mutable()->sync_write(buf, size, dev_offset);
 }
 
 std::error_code VirtualDev::sync_write(const char* buf, uint32_t size, cshared< Chunk >& chunk,
                                        uint64_t offset_in_chunk) {
+    if (sisl_unlikely(!is_chunk_loaded(chunk))) {
+        return std::make_error_code(std::errc::resource_unavailable_try_again);
+    }
     return chunk->physical_dev_mutable()->sync_write(buf, size, chunk->start_offset() + offset_in_chunk);
 }
 
@@ -389,6 +411,9 @@ std::error_code VirtualDev::sync_writev(const iovec* iov, int iovcnt, BlkId cons
 
     Chunk* chunk;
     uint64_t const dev_offset = to_dev_offset(bid, &chunk);
+    if (sisl_unlikely(dev_offset == INVALID_DEV_OFFSET)) {
+        return std::make_error_code(std::errc::resource_unavailable_try_again);
+    }
     auto const size = get_len(iov, iovcnt);
     auto* pdev = chunk->physical_dev_mutable();
 
@@ -402,6 +427,9 @@ std::error_code VirtualDev::sync_writev(const iovec* iov, int iovcnt, BlkId cons
 
 std::error_code VirtualDev::sync_writev(const iovec* iov, int iovcnt, cshared< Chunk >& chunk,
                                         uint64_t offset_in_chunk) {
+    if (sisl_unlikely(!is_chunk_loaded(chunk))) {
+        return std::make_error_code(std::errc::resource_unavailable_try_again);
+    }
     uint64_t const dev_offset = chunk->start_offset() + offset_in_chunk;
     auto const size = get_len(iov, iovcnt);
     auto* pdev = chunk->physical_dev_mutable();
@@ -414,13 +442,18 @@ std::error_code VirtualDev::sync_writev(const iovec* iov, int iovcnt, cshared< C
     return pdev->sync_writev(iov, iovcnt, size, dev_offset);
 }
 
+// for read, chunk might be missing in case of pdev is gone(for example , breakfix), so we need to check if chunk is
+// loaded before proceeding with read;
 ////////////////////////////////// async read section ///////////////////////////////////////////////
 folly::Future< std::error_code > VirtualDev::async_read(char* buf, uint64_t size, BlkId const& bid,
                                                         bool part_of_batch) {
     HS_DBG_ASSERT_EQ(bid.is_multi(), false, "async_read needs individual pieces of blkid - not MultiBlkid");
 
     Chunk* pchunk;
     uint64_t const dev_offset = to_dev_offset(bid, &pchunk);
+    if (sisl_unlikely(dev_offset == INVALID_DEV_OFFSET)) {
+        return folly::makeFuture< std::error_code >(std::make_error_code(std::errc::resource_unavailable_try_again));
+    }
     return pchunk->physical_dev_mutable()->async_read(buf, size, dev_offset, part_of_batch);
 }
 
@@ -430,6 +463,9 @@ folly::Future< std::error_code > VirtualDev::async_readv(iovec* iovs, int iovcnt
 
     Chunk* pchunk;
     uint64_t const dev_offset = to_dev_offset(bid, &pchunk);
+    if (sisl_unlikely(dev_offset == INVALID_DEV_OFFSET)) {
+        return folly::makeFuture< std::error_code >(std::make_error_code(std::errc::resource_unavailable_try_again));
+    }
     return pchunk->physical_dev_mutable()->async_readv(iovs, iovcnt, size, dev_offset, part_of_batch);
 }
 
@@ -439,10 +475,16 @@ std::error_code VirtualDev::sync_read(char* buf, uint32_t size, BlkId const& bid
 
     Chunk* chunk;
     uint64_t const dev_offset = to_dev_offset(bid, &chunk);
+    if (sisl_unlikely(dev_offset == INVALID_DEV_OFFSET)) {
+        return std::make_error_code(std::errc::resource_unavailable_try_again);
+    }
     return chunk->physical_dev_mutable()->sync_read(buf, size, dev_offset);
 }
 
 std::error_code VirtualDev::sync_read(char* buf, uint32_t size, cshared< Chunk >& chunk, uint64_t offset_in_chunk) {
+    if (sisl_unlikely(!is_chunk_loaded(chunk))) {
+        return std::make_error_code(std::errc::resource_unavailable_try_again);
+    }
     return chunk->physical_dev_mutable()->sync_read(buf, size, chunk->start_offset() + offset_in_chunk);
 }
 
@@ -451,6 +493,9 @@ std::error_code VirtualDev::sync_readv(iovec* iov, int iovcnt, BlkId const& bid)
 
     Chunk* chunk;
     uint64_t const dev_offset = to_dev_offset(bid, &chunk);
+    if (sisl_unlikely(dev_offset == INVALID_DEV_OFFSET)) {
+        return std::make_error_code(std::errc::resource_unavailable_try_again);
+    }
     auto const size = get_len(iov, iovcnt);
     auto* pdev = chunk->physical_dev_mutable();
 
@@ -463,6 +508,9 @@ std::error_code VirtualDev::sync_readv(iovec* iov, int iovcnt, BlkId const& bid)
 }
 
 std::error_code VirtualDev::sync_readv(iovec* iov, int iovcnt, cshared< Chunk >& chunk, uint64_t offset_in_chunk) {
+    if (sisl_unlikely(!is_chunk_loaded(chunk))) {
+        return std::make_error_code(std::errc::resource_unavailable_try_again);
+    }
     uint64_t const dev_offset = chunk->start_offset() + offset_in_chunk;
     auto const size = get_len(iov, iovcnt);
     auto* pdev = chunk->physical_dev_mutable();
@@ -600,8 +648,12 @@ int VirtualDev::cp_progress_percent() { return 100; }
 ///////////////////////// VirtualDev Private Methods /////////////////////////////
 uint64_t VirtualDev::to_dev_offset(BlkId const& b, Chunk** chunk) const {
     *chunk = m_dmgr.get_chunk_mutable(b.chunk_num());
-    RELEASE_ASSERT(*chunk, "Chunk got null {}", b.chunk_num());
+    if (!(*chunk)) return INVALID_DEV_OFFSET;
     return uint64_cast(b.blk_num()) * block_size() + uint64_cast((*chunk)->start_offset());
 }
 
+bool VirtualDev::is_chunk_loaded(cshared< Chunk >& chunk) const {
+    return m_dmgr.get_chunk(chunk->chunk_id()) != nullptr;
+}
+
 } // namespace homestore
diff --git a/src/lib/device/virtual_dev.hpp b/src/lib/device/virtual_dev.hpp
@@ -299,6 +299,7 @@ class VirtualDev {
 
 private:
     uint64_t to_dev_offset(BlkId const& b, Chunk** chunk) const;
+    bool is_chunk_loaded(cshared< Chunk >& chunk) const;
     BlkAllocStatus alloc_blks_from_chunk(blk_count_t nblks, blk_alloc_hints const& hints, MultiBlkId& out_blkid,
                                          Chunk* chunk);
 };