Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Enable RAID support for FDP IO path #319

Open
wants to merge 1 commit into
base: main
Choose a base branch
from
+61 −56
Open

Enable RAID support for FDP IO path #319

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
117 changes: 61 additions & 56 deletions cachelib/navy/common/Device.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -220,12 +220,13 @@ class CompletionHandler : public folly::EventHandler {
// Per-thread context for AsyncIO like libaio or io_uring
class AsyncIoContext : public IoContext {
public:
AsyncIoContext(std::unique_ptr<folly::AsyncBase>&& asyncBase,
size_t id,
folly::EventBase* evb,
size_t capacity,
bool useIoUring,
std::vector<std::shared_ptr<FdpNvme>> fdpNvmeVec);
AsyncIoContext(
std::unique_ptr<folly::AsyncBase>&& asyncBase,
size_t id,
folly::EventBase* evb,
size_t capacity,
bool useIoUring,
const std::unordered_map<int, std::shared_ptr<FdpNvme>>& fdpNvmeDevs);

~AsyncIoContext() override = default;

Expand Down Expand Up @@ -279,18 +280,16 @@ class AsyncIoContext : public IoContext {
size_t numSubmitted_ = 0;
size_t numCompleted_ = 0;

// Device info vector for FDP support
const std::vector<std::shared_ptr<FdpNvme>> fdpNvmeVec_{};
// As of now, only one FDP enabled Device is supported
static constexpr uint16_t kDefaultFdpIdx = 0u;
// Map of file descriptors to FdpNvme device objects
const std::unordered_map<int, std::shared_ptr<FdpNvme>>& fdpNvmeDevs_;
};

// An FileDevice manages direct I/O to either a single or multiple (RAID0)
// block device(s) or regular file(s).
class FileDevice : public Device {
public:
FileDevice(std::vector<folly::File>&& fvec,
std::vector<std::shared_ptr<FdpNvme>>&& fdpNvmeVec,
std::unordered_map<int, std::shared_ptr<FdpNvme>>&& fdpNvmeDevs,
uint64_t size,
uint32_t blockSize,
uint32_t stripeSize,
Expand All @@ -317,8 +316,8 @@ class FileDevice : public Device {
// File vector for devices or regular files
const std::vector<folly::File> fvec_{};

// Device info vector for FDP support
const std::vector<std::shared_ptr<FdpNvme>> fdpNvmeVec_{};
// Map of file descriptors to FdpNvme device objects
const std::unordered_map<int, std::shared_ptr<FdpNvme>> fdpNvmeDevs_;

// RAID stripe size when multiple devices are used
const uint32_t stripeSize_;
Expand Down Expand Up @@ -750,20 +749,21 @@ bool SyncIoContext::submitIo(IOOp& op) {
/*
* AsyncIoContext
*/
AsyncIoContext::AsyncIoContext(std::unique_ptr<folly::AsyncBase>&& asyncBase,
size_t id,
folly::EventBase* evb,
size_t capacity,
bool useIoUring,
std::vector<std::shared_ptr<FdpNvme>> fdpNvmeVec)
AsyncIoContext::AsyncIoContext(
std::unique_ptr<folly::AsyncBase>&& asyncBase,
size_t id,
folly::EventBase* evb,
size_t capacity,
bool useIoUring,
const std::unordered_map<int, std::shared_ptr<FdpNvme>>& fdpNvmeDevs)
: asyncBase_(std::move(asyncBase)),
id_(id),
qDepth_(capacity),
useIoUring_(useIoUring),
fdpNvmeVec_(fdpNvmeVec) {
fdpNvmeDevs_(fdpNvmeDevs) {
#ifdef CACHELIB_IOURING_DISABLE
// io_uring is not available on the system
XDCHECK(!useIoUring_ && !(fdpNvmeVec_.size() > 0));
XDCHECK(!useIoUring_ && !(fdpNvmeDevs_.size() > 0));
useIoUring_ = false;
#endif
if (evb) {
Expand All @@ -781,7 +781,7 @@ AsyncIoContext::AsyncIoContext(std::unique_ptr<folly::AsyncBase>&& asyncBase,
"[{}] Created new async io context with qdepth {}{} io_engine {} {}",
getName(), qDepth_, qDepth_ == 1 ? " (sync wait)" : "",
useIoUring_ ? "io_uring" : "libaio",
(fdpNvmeVec_.size() > 0) ? "FDP enabled" : "");
(fdpNvmeDevs_.size() > 0) ? "FDP enabled" : "");
}

void AsyncIoContext::pollCompletion() {
Expand Down Expand Up @@ -820,7 +820,7 @@ void AsyncIoContext::handleCompletion(
}

auto len = aop->result();
if (fdpNvmeVec_.size() > 0) {
if (fdpNvmeDevs_.size() > 0) {
// 0 means success here, so get the completed size from iop
len = !len ? iop->size_ : 0;
}
Expand Down Expand Up @@ -869,7 +869,7 @@ bool AsyncIoContext::submitIo(IOOp& op) {
}

std::unique_ptr<folly::AsyncBaseOp> AsyncIoContext::prepAsyncIo(IOOp& op) {
if (fdpNvmeVec_.size() > 0) {
if (fdpNvmeDevs_.size() > 0) {
return prepNvmeIo(op);
}

Expand Down Expand Up @@ -905,10 +905,10 @@ std::unique_ptr<folly::AsyncBaseOp> AsyncIoContext::prepNvmeIo(IOOp& op) {
iouringCmdOp->initBase();
struct io_uring_sqe& sqe = iouringCmdOp->getSqe();
if (req.opType_ == OpType::READ) {
fdpNvmeVec_[kDefaultFdpIdx]->prepReadUringCmdSqe(sqe, op.data_, op.size_,
op.offset_);
fdpNvmeDevs_.at(op.fd_)->prepReadUringCmdSqe(sqe, op.data_, op.size_,
op.offset_);
} else {
fdpNvmeVec_[kDefaultFdpIdx]->prepWriteUringCmdSqe(
fdpNvmeDevs_.at(op.fd_)->prepWriteUringCmdSqe(
sqe, op.data_, op.size_, op.offset_, op.placeHandle_.value_or(-1));
}
io_uring_sqe_set_data(&sqe, iouringCmdOp.get());
Expand All @@ -921,23 +921,24 @@ std::unique_ptr<folly::AsyncBaseOp> AsyncIoContext::prepNvmeIo(IOOp& op) {
/*
* FileDevice
*/
FileDevice::FileDevice(std::vector<folly::File>&& fvec,
std::vector<std::shared_ptr<FdpNvme>>&& fdpNvmeVec,
uint64_t fileSize,
uint32_t blockSize,
uint32_t stripeSize,
uint32_t maxIOSize,
uint32_t maxDeviceWriteSize,
IoEngine ioEngine,
uint32_t qDepthPerContext,
std::shared_ptr<DeviceEncryptor> encryptor)
FileDevice::FileDevice(
std::vector<folly::File>&& fvec,
std::unordered_map<int, std::shared_ptr<FdpNvme>>&& fdpNvmeDevs,
uint64_t fileSize,
uint32_t blockSize,
uint32_t stripeSize,
uint32_t maxIOSize,
uint32_t maxDeviceWriteSize,
IoEngine ioEngine,
uint32_t qDepthPerContext,
std::shared_ptr<DeviceEncryptor> encryptor)
: Device(fileSize * fvec.size(),
std::move(encryptor),
blockSize,
maxIOSize,
maxDeviceWriteSize),
fvec_(std::move(fvec)),
fdpNvmeVec_(std::move(fdpNvmeVec)),
fdpNvmeDevs_(std::move(fdpNvmeDevs)),
stripeSize_(stripeSize),
ioEngine_(ioEngine),
qDepthPerContext_(qDepthPerContext) {
Expand Down Expand Up @@ -974,7 +975,7 @@ FileDevice::FileDevice(std::vector<folly::File>&& fvec,
"num_fdp_devices {}",
fvec_.size(), getSize(), blockSize, stripeSize, maxDeviceWriteSize,
maxIOSize, getIoEngineName(ioEngine_), qDepthPerContext_,
fdpNvmeVec_.size());
fdpNvmeDevs_.size());
}

bool FileDevice::readImpl(uint64_t offset, uint32_t size, void* value) {
Expand Down Expand Up @@ -1030,7 +1031,7 @@ IoContext* FileDevice::getIoContext() {
std::unique_ptr<folly::AsyncBase> asyncBase;
if (useIoUring) {
#ifndef CACHELIB_IOURING_DISABLE
if (fdpNvmeVec_.size() > 0) {
if (fdpNvmeDevs_.size() > 0) {
// Big sqe/cqe is mandatory for NVMe passthrough
// https://elixir.bootlin.com/linux/v6.7/source/drivers/nvme/host/ioctl.c#L742
folly::IoUringOp::Options options;
Expand All @@ -1051,7 +1052,7 @@ IoContext* FileDevice::getIoContext() {
auto idx = incrementalIdx_++;
tlContext_.reset(new AsyncIoContext(std::move(asyncBase), idx, evb,
qDepthPerContext_, useIoUring,
fdpNvmeVec_));
fdpNvmeDevs_));

{
// Keep pointers in a vector to ease the gdb debugging
Expand All @@ -1067,10 +1068,20 @@ IoContext* FileDevice::getIoContext() {
}

int FileDevice::allocatePlacementHandle() {
static constexpr uint16_t kDefaultFdpIdx = 0u;
#ifndef CACHELIB_IOURING_DISABLE
if (fdpNvmeVec_.size() > 0) {
return fdpNvmeVec_[kDefaultFdpIdx]->allocateFdpHandle();
if (fdpNvmeDevs_.size() > 0) {
auto fdpHandle = -1;
// Ensuring that same FDP placement handle is allocated for all FdpNvme
// devices for RAID, and returns the allocated handle if successful,
// or -1 if there is a conflict
for (auto& nvmeFdp : fdpNvmeDevs_) {
auto tempHandle = nvmeFdp.second->allocateFdpHandle();
if (fdpHandle != -1 && (tempHandle != fdpHandle)) {
return -1;
}
fdpHandle = tempHandle;
}
return fdpHandle;
}
#endif
return -1;
Expand Down Expand Up @@ -1186,31 +1197,25 @@ std::unique_ptr<Device> createDirectIoFileDevice(
XDCHECK(folly::isPowTwo(blockSize));

uint32_t maxIOSize = maxDeviceWriteSize;
std::vector<std::shared_ptr<FdpNvme>> fdpNvmeVec{};
std::unordered_map<int, std::shared_ptr<FdpNvme>> fdpNvmeDevs;
#ifndef CACHELIB_IOURING_DISABLE
if (isFDPEnabled) {
try {
if (filePaths.size() > 1) {
throw std::invalid_argument(folly::sformat(
"{} input files; but FDP mode does not support RAID files yet",
filePaths.size()));
}

for (const auto& path : filePaths) {
auto fdpNvme = std::make_shared<FdpNvme>(path);
for (size_t i = 0; i < filePaths.size(); i++) {
auto fdpNvme = std::make_shared<FdpNvme>(filePaths[i]);

auto maxDevIOSize = fdpNvme->getMaxIOSize();
if (maxDevIOSize != 0u &&
(maxIOSize == 0u || maxDevIOSize < maxIOSize)) {
maxIOSize = maxDevIOSize;
}

fdpNvmeVec.push_back(std::move(fdpNvme));
fdpNvmeDevs.insert({fVec[i].fd(), std::move(fdpNvme)});
}
} catch (const std::exception& e) {
XLOGF(ERR, "NVMe FDP mode could not be enabled {}, Errno: {}", e.what(),
errno);
fdpNvmeVec.clear();
fdpNvmeDevs.clear();
maxIOSize = 0u;
}
}
Expand All @@ -1221,7 +1226,7 @@ std::unique_ptr<Device> createDirectIoFileDevice(
}

return std::make_unique<FileDevice>(std::move(fVec),
std::move(fdpNvmeVec),
std::move(fdpNvmeDevs),
fileSize,
blockSize,
stripeSize,
Expand Down