a11::stores::LocalChunkStore::State Struct Reference

Inheritance diagram for a11::stores::LocalChunkStore::State:

Classes
struct	ReadCompletion
struct	ReadRequest

Public Types
enum class	ReadKind { kSequence , kArrivalOrder }

Public Member Functions
	State (std::string id)
a11::Future< data::NodeFragment >	Read (ReadKind kind, std::uint64_t value, absl::Time deadline)
std::optional< absl::StatusOr< data::NodeFragment > >	LookupReadLocked (const ReadRequest &request) const ABSL_EXCLUSIVE_LOCKS_REQUIRED(mu)
void	CollectReadsLocked (std::vector< ReadCompletion > *completions) ABSL_EXCLUSIVE_LOCKS_REQUIRED(mu)
void	RemoveRead (const ReadRequest *absl_nonnull request)
std::optional< absl::Status >	CollectNext (std::vector< std::optional< data::NodeFragment > > &fragments, size_t limit, std::shared_ptr< thread::PermanentEvent > *waiter) ABSL_LOCKS_EXCLUDED(mu)
	Append whatever Next can return right now, without ever blocking.
absl::flat_hash_map< std::uint32_t, std::uint64_t > seq_to_arrival_order	ABSL_GUARDED_BY (mu)
absl::flat_hash_map< std::uint64_t, std::uint32_t > arrival_order_to_seq	ABSL_GUARDED_BY (mu)
absl::flat_hash_map< std::uint32_t, data::Chunk > chunks	ABSL_GUARDED_BY (mu)
std::optional< std::uint32_t > final_seq	ABSL_GUARDED_BY (mu)
std::uint64_t total_chunks_put	ABSL_GUARDED_BY (mu)=0
std::uint64_t total_chunks_read	ABSL_GUARDED_BY (mu)=0
std::optional< absl::Status > status	ABSL_GUARDED_BY (mu)
std::vector< std::shared_ptr< ReadRequest > > pending_reads	ABSL_GUARDED_BY (mu)
std::shared_ptr< thread::PermanentEvent > changed	ABSL_GUARDED_BY (mu)

Static Public Member Functions
static void	CompleteReads (std::vector< ReadCompletion > completions)

Public Attributes
thread::Mutex	mu
const std::string	node_id

Member Enumeration Documentation

ReadKind

enum class a11::stores::LocalChunkStore::State::ReadKind

strong

Enumerator
kSequence
kArrivalOrder

Constructor & Destructor Documentation

State()

a11::stores::LocalChunkStore::State::State ( std::string  id )

inlineexplicit

Member Function Documentation

ABSL_GUARDED_BY() [1/9]

absl::flat_hash_map< std::uint32_t, std::uint64_t > seq_to_arrival_order a11::stores::LocalChunkStore::State::ABSL_GUARDED_BY

(

mu

)

ABSL_GUARDED_BY() [2/9]

absl::flat_hash_map< std::uint64_t, std::uint32_t > arrival_order_to_seq a11::stores::LocalChunkStore::State::ABSL_GUARDED_BY

(

mu

)

ABSL_GUARDED_BY() [3/9]

absl::flat_hash_map< std::uint32_t, data::Chunk > chunks a11::stores::LocalChunkStore::State::ABSL_GUARDED_BY

(

mu

)

ABSL_GUARDED_BY() [4/9]

std::optional< std::uint32_t > final_seq a11::stores::LocalChunkStore::State::ABSL_GUARDED_BY

(

mu

)

ABSL_GUARDED_BY() [5/9]

std::optional< absl::Status > status a11::stores::LocalChunkStore::State::ABSL_GUARDED_BY

(

mu

)

ABSL_GUARDED_BY() [6/9]

std::vector< std::shared_ptr< ReadRequest > > pending_reads a11::stores::LocalChunkStore::State::ABSL_GUARDED_BY

(

mu

)

ABSL_GUARDED_BY() [7/9]

std::shared_ptr< thread::PermanentEvent > changed a11::stores::LocalChunkStore::State::ABSL_GUARDED_BY

(

mu

)

ABSL_GUARDED_BY() [8/9]

std::uint64_t total_chunks_put a11::stores::LocalChunkStore::State::ABSL_GUARDED_BY ( mu  )

pure virtual

ABSL_GUARDED_BY() [9/9]

std::uint64_t total_chunks_read a11::stores::LocalChunkStore::State::ABSL_GUARDED_BY ( mu  )

pure virtual

CollectNext()

std::optional< absl::Status > a11::stores::LocalChunkStore::State::CollectNext ( std::vector< std::optional< data::NodeFragment > > &  fragments, size_t  limit, std::shared_ptr< thread::PermanentEvent > *  waiter  )

inline

Append whatever Next can return right now, without ever blocking.

Split out of Next() so the common case – the fragments are already here – can run on the caller's thread. Wrapping the whole loop in a Submit() would spawn a fiber even when nothing needs waiting for, and a fiber spawn costs more than an order of magnitude what the read itself does.

Parameters

fragments	Accumulator, appended to. Carries across waits, so a partial batch collected before a wait is still there after it.
limit	Maximum fragments to accumulate, counting what is already there.
waiter	Set to the generation event only when the caller must wait. It is snapshotted inside the lock, which is what closes the lost-wakeup window – see WaitForChange().

Returns

An engaged status when the call has its answer: OK means fragments is that answer, and a non-OK status is the terminal one to fail with. std::nullopt means the caller must wait on *waiter and try again.

CollectReadsLocked()

void a11::stores::LocalChunkStore::State::CollectReadsLocked ( std::vector< ReadCompletion > *  completions )

inline

CompleteReads()

static void a11::stores::LocalChunkStore::State::CompleteReads ( std::vector< ReadCompletion >  completions )

inlinestatic

LookupReadLocked()

std::optional< absl::StatusOr< data::NodeFragment > > a11::stores::LocalChunkStore::State::LookupReadLocked ( const ReadRequest &  request ) const

inline

Read()

a11::Future< data::NodeFragment > a11::stores::LocalChunkStore::State::Read ( ReadKind  kind, std::uint64_t  value, absl::Time  deadline  )

inline

RemoveRead()

void a11::stores::LocalChunkStore::State::RemoveRead ( const ReadRequest *absl_nonnull  request )

inline

Member Data Documentation

mu

thread::Mutex a11::stores::LocalChunkStore::State::mu

node_id

const std::string a11::stores::LocalChunkStore::State::node_id

---

The documentation for this struct was generated from the following file:

• cpp/a11/stores/local_chunk_store.cc