This class helps with providing exactly-once semantics for read-write RPCs. More...

#include <ClientImpl.h>

Public Member Functions
	ExactlyOnceRPCHelper (ClientImpl *client)
	Constructor.
	~ExactlyOnceRPCHelper ()
	Destructor.
void	exit ()
	Prepare to shut down (join with thread).
Protocol::Client::ExactlyOnceRPCInfo	getRPCInfo (TimePoint timeout)
	Call this before sending an RPC.
void	doneWithRPC (const Protocol::Client::ExactlyOnceRPCInfo &)
	Call this after receiving an RPCs response.
Private Member Functions
Protocol::Client::ExactlyOnceRPCInfo	getRPCInfo (Core::HoldingMutex holdingMutex, TimePoint timeout)
	Internal version of getRPCInfo() to avoid deadlock with self.
void	doneWithRPC (const Protocol::Client::ExactlyOnceRPCInfo &, Core::HoldingMutex holdingMutex)
	Internal version of doneWithRPC() to avoid deadlock with self.
void	keepAliveThreadMain ()
	Main function for keep-alive thread.
	ExactlyOnceRPCHelper (const ExactlyOnceRPCHelper &)
ExactlyOnceRPCHelper &	operator= (const ExactlyOnceRPCHelper &)
Private Attributes
ClientImpl *	client
	Used to open a session with the cluster.
Core::Mutex	mutex
	Protects all the members of this class.
std::set< uint64_t >	outstandingRPCNumbers
	The numbers of the RPCs for which this client is still awaiting a response.
uint64_t	clientId
	The client's session ID as returned by the open session RPC, or 0 if one has not yet been assigned.
uint64_t	nextRPCNumber
	The number to assign to the next RPC.
Core::ConditionVariable	keepAliveCV
	keepAliveThread blocks on this.
bool	exiting
	Flag to keepAliveThread that it should shut down.
TimePoint	lastKeepAliveStart
	Time just before the last keep-alive or read-write request to the cluster was made.
std::chrono::milliseconds	keepAliveInterval
	How often session keep-alive requests are sent during periods of inactivity.
std::chrono::milliseconds	sessionCloseTimeout
	How long to wait for the CloseSession RPC before giving up.
std::unique_ptr < LeaderRPCBase::Call >	keepAliveCall
	If set, this is an ongoing keep-alive RPC.
std::thread	keepAliveThread
	Runs keepAliveThreadMain().

Detailed Description

This class helps with providing exactly-once semantics for read-write RPCs.

For example, it assigns sequence numbers to RPCs, which servers then use to prevent duplicate processing of duplicate requests.

This class is implemented in a monitor style.

Definition at line 211 of file ClientImpl.h.

Constructor & Destructor Documentation

LogCabin::Client::ClientImpl::ExactlyOnceRPCHelper::ExactlyOnceRPCHelper ( ClientImpl * client ) [explicit]

Constructor.

Parameters:

client Used to open a session with the cluster. Should not be NULL.

Definition at line 214 of file ClientImpl.cc.

LogCabin::Client::ClientImpl::ExactlyOnceRPCHelper::~ExactlyOnceRPCHelper ( )

Destructor.

Definition at line 236 of file ClientImpl.cc.

LogCabin::Client::ClientImpl::ExactlyOnceRPCHelper::ExactlyOnceRPCHelper ( const ExactlyOnceRPCHelper & ) [private]

Member Function Documentation

void LogCabin::Client::ClientImpl::ExactlyOnceRPCHelper::exit ( )

Prepare to shut down (join with thread).

Definition at line 241 of file ClientImpl.cc.

Protocol::Client::ExactlyOnceRPCInfo LogCabin::Client::ClientImpl::ExactlyOnceRPCHelper::getRPCInfo ( TimePoint timeout )

Call this before sending an RPC.

Parameters:

timeout If this timeout elapses before a session can be opened with the cluster, this method will return early and the returned information will have a client_id set to 0, which is not a valid ID.

Returns:: Info to be used with read-write RPCs, or if the timeout elapsed, a client_id set to 0.

Definition at line 286 of file ClientImpl.cc.

void LogCabin::Client::ClientImpl::ExactlyOnceRPCHelper::doneWithRPC ( const Protocol::Client::ExactlyOnceRPCInfo & rpcInfo )

Call this after receiving an RPCs response.

Definition at line 293 of file ClientImpl.cc.

Protocol::Client::ExactlyOnceRPCInfo LogCabin::Client::ClientImpl::ExactlyOnceRPCHelper::getRPCInfo	(	Core::HoldingMutex	holdingMutex,
		TimePoint	timeout
	)		`[private]`

Internal version of getRPCInfo() to avoid deadlock with self.

Definition at line 301 of file ClientImpl.cc.

void LogCabin::Client::ClientImpl::ExactlyOnceRPCHelper::doneWithRPC	(	const Protocol::Client::ExactlyOnceRPCInfo &	rpcInfo,
		Core::HoldingMutex	holdingMutex
	)		`[private]`

Internal version of doneWithRPC() to avoid deadlock with self.

Definition at line 351 of file ClientImpl.cc.

void LogCabin::Client::ClientImpl::ExactlyOnceRPCHelper::keepAliveThreadMain ( ) [private]

Main function for keep-alive thread.

Periodically makes requests to the cluster to keep the client's session active.

Definition at line 359 of file ClientImpl.cc.

ExactlyOnceRPCHelper& LogCabin::Client::ClientImpl::ExactlyOnceRPCHelper::operator= ( const ExactlyOnceRPCHelper & ) [private]

Member Data Documentation

ClientImpl* LogCabin::Client::ClientImpl::ExactlyOnceRPCHelper::client [private]

Used to open a session with the cluster.

const and non-NULL except for unit tests.

Definition at line 267 of file ClientImpl.h.

Core::Mutex LogCabin::Client::ClientImpl::ExactlyOnceRPCHelper::mutex [mutable, private]

Protects all the members of this class.

Definition at line 271 of file ClientImpl.h.

std::set<uint64_t> LogCabin::Client::ClientImpl::ExactlyOnceRPCHelper::outstandingRPCNumbers [private]

The numbers of the RPCs for which this client is still awaiting a response.

Definition at line 276 of file ClientImpl.h.

uint64_t LogCabin::Client::ClientImpl::ExactlyOnceRPCHelper::clientId [private]

The client's session ID as returned by the open session RPC, or 0 if one has not yet been assigned.

Definition at line 281 of file ClientImpl.h.

uint64_t LogCabin::Client::ClientImpl::ExactlyOnceRPCHelper::nextRPCNumber [private]

The number to assign to the next RPC.

Definition at line 285 of file ClientImpl.h.

Core::ConditionVariable LogCabin::Client::ClientImpl::ExactlyOnceRPCHelper::keepAliveCV [private]

keepAliveThread blocks on this.

Notified when lastKeepAliveStart, keepAliveIntervalMs, or exiting changes.

Definition at line 290 of file ClientImpl.h.

bool LogCabin::Client::ClientImpl::ExactlyOnceRPCHelper::exiting [private]

Flag to keepAliveThread that it should shut down.

Definition at line 294 of file ClientImpl.h.

TimePoint LogCabin::Client::ClientImpl::ExactlyOnceRPCHelper::lastKeepAliveStart [private]

Time just before the last keep-alive or read-write request to the cluster was made.

The next keep-alive request will be invoked keepAliveIntervalMs after this, if no intervening requests are made.

Definition at line 300 of file ClientImpl.h.

std::chrono::milliseconds LogCabin::Client::ClientImpl::ExactlyOnceRPCHelper::keepAliveInterval [private]

How often session keep-alive requests are sent during periods of inactivity.

Definition at line 305 of file ClientImpl.h.

std::chrono::milliseconds LogCabin::Client::ClientImpl::ExactlyOnceRPCHelper::sessionCloseTimeout [private]

How long to wait for the CloseSession RPC before giving up.

Definition at line 309 of file ClientImpl.h.

std::unique_ptr<LeaderRPCBase::Call> LogCabin::Client::ClientImpl::ExactlyOnceRPCHelper::keepAliveCall [private]

If set, this is an ongoing keep-alive RPC.

This call is canceled to interrupt keepAliveThread when exiting.

Definition at line 315 of file ClientImpl.h.

std::thread LogCabin::Client::ClientImpl::ExactlyOnceRPCHelper::keepAliveThread [private]

Runs keepAliveThreadMain().

Since this thread would be unexpected/wasteful for clients that only issue read-only requests (or no requests at all), it is spawned lazily, if/when the client opens its session with the cluster (upon its first read-write request).

Definition at line 324 of file ClientImpl.h.

The documentation for this class was generated from the following files:

Client/ClientImpl.h
Client/ClientImpl.cc

Public Member Functions

Private Member Functions

Private Attributes

Detailed Description

Constructor & Destructor Documentation

Member Function Documentation

Member Data Documentation