repmgr.h   [plain text]

 * See the file LICENSE for redistribution information.
 * Copyright (c) 2006,2008 Oracle.  All rights reserved.
 * $Id: repmgr.h,v 12.19 2008/01/08 20:58:18 bostic Exp $

#ifndef _DB_REPMGR_H_
#define	_DB_REPMGR_H_

#include "dbinc_auto/repmgr_auto.h"

#if defined(__cplusplus)
extern "C" {

 * Replication Framework message types.  These values are transmitted to
 * identify messages sent between sites, even sites running differing versions
 * of software.  Therefore, once assigned, the values are permanently "frozen".
 * New message types added in later versions always get new (higher) values.
 * For example, in repmgr wire protocol version 1 the highest assigned message
 * type value was 3, for REPMGR_REP_MESSAGE.  Wire protocol version 2 added the
 * HEARTBEAT message type (4).
 * We still list them in alphabetical order, for ease of reference.  But this
 * generally does not correspond to numerical order.
#define	REPMGR_ACK		1	/* Acknowledgement. */
#define	REPMGR_HANDSHAKE	2	/* Connection establishment sequence. */
#define	REPMGR_HEARTBEAT	4	/* Monitor connection health. */
#define	REPMGR_REP_MESSAGE	3	/* Normal replication message. */

/* Heartbeats were introduced in version 2. */

/* The range of protocol versions we're willing to support. */

#ifdef DB_WIN32
typedef SOCKET socket_t;
typedef HANDLE thread_id_t;
typedef HANDLE mgr_mutex_t;
typedef HANDLE cond_var_t;
typedef WSABUF db_iovec_t;
typedef int socket_t;
typedef pthread_t thread_id_t;
typedef pthread_mutex_t mgr_mutex_t;
typedef pthread_cond_t cond_var_t;
typedef struct iovec db_iovec_t;

 * The (arbitrary) maximum number of outgoing messages we're willing to hold, on
 * a queue per connection, waiting for TCP buffer space to become available in
 * the kernel.  Rather than exceeding this limit, we simply discard additional
 * messages (since this is always allowed by the replication protocol).
 *    As a special dispensation, if a message is destined for a specific remote
 * site (i.e., it's not a broadcast), then we first try blocking the sending
 * thread, waiting for space to become available (though we only wait a limited
 * time).  This is so as to be able to handle the immediate flood of (a
 * potentially large number of) outgoing messages that replication generates, in
 * a tight loop, when handling PAGE_REQ, LOG_REQ and ALL_REQ requests.
#define	OUT_QUEUE_LIMIT	10

 * The system value is available from sysconf(_SC_HOST_NAME_MAX).
 * Historically, the maximum host name was 256.

/* A buffer big enough for the string "site". */

/* Default timeout values, in seconds. */

struct __repmgr_connection;
    typedef struct __repmgr_connection REPMGR_CONNECTION;
struct __repmgr_queue; typedef struct __repmgr_queue REPMGR_QUEUE;
struct __queued_output; typedef struct __queued_output QUEUED_OUTPUT;
struct __repmgr_retry; typedef struct __repmgr_retry REPMGR_RETRY;
struct __repmgr_runnable; typedef struct __repmgr_runnable REPMGR_RUNNABLE;
struct __repmgr_site; typedef struct __repmgr_site REPMGR_SITE;
struct __ack_waiters_table;
    typedef struct __ack_waiters_table ACK_WAITERS_TABLE;

typedef TAILQ_HEAD(__repmgr_conn_list, __repmgr_connection) CONNECTION_LIST;
typedef STAILQ_HEAD(__repmgr_out_q_head, __queued_output) OUT_Q_HEADER;
typedef TAILQ_HEAD(__repmgr_retry_q, __repmgr_retry) RETRY_Q_HEADER;

/* Information about threads managed by Replication Framework. */
struct __repmgr_runnable {
	ENV *env;
	thread_id_t thread_id;
	void *(*run) __P((void *));
	int finished;

 * Information about pending connection establishment retry operations.
 * We keep these in order by time.  This works, under the assumption that the
 * DB_REP_CONNECTION_RETRY never changes once we get going (though that
 * assumption is of course wrong, so this needs to be fixed).
 * Usually, we put things onto the tail end of the list.  But when we add a new
 * site while threads are running, we trigger its first connection attempt by
 * scheduling a retry for "0" microseconds from now, putting its retry element
 * at the head of the list instead.
 * TODO: I think this can be fixed by defining "time" to be the time the element
 * was added (with some convention like "0" meaning immediate), rather than the
 * deadline time.
struct __repmgr_retry {
	TAILQ_ENTRY(__repmgr_retry) entries;
	u_int eid;
	db_timespec time;

 * We use scatter/gather I/O for both reading and writing.  The largest number
 * of buffers we ever try to use at once is 5, corresponding to the 5 segments
 * of a message described in the "wire protocol" (repmgr_net.c).
typedef struct {
	db_iovec_t vectors[5];

	 * Index of the first iovec to be used.  Initially of course this is
	 * zero.  But as we progress through partial I/O transfers, it ends up
	 * pointing to the first iovec to be used on the next operation.
	int offset;

	 * Total number of pieces defined for this message; equal to the number
	 * of times add_buffer and/or add_dbt were called to populate it.  We do
	 * *NOT* revise this as we go along.  So subsequent I/O operations must
	 * use count-offset to get the number of active vector pieces still
	 * remaining.
	int count;

	 * Total number of bytes accounted for in all the pieces of this
	 * message.  We do *NOT* revise this as we go along (though it's not
	 * clear we shouldn't).
	size_t total_bytes;

typedef struct {
	size_t length;		/* number of bytes in data */
	int ref_count;		/* # of sites' send queues pointing to us */
	u_int8_t data[1];	/* variable size data area */

struct __queued_output {
	STAILQ_ENTRY(__queued_output) entries;
	size_t offset;

 * The following is for input.  Once we know the sizes of the pieces of an
 * incoming message, we can create this struct (and also the data areas for the
 * pieces themselves, in the same memory allocation).  This is also the struct
 * in which the message lives while it's waiting to be processed by message
 * threads.
typedef struct __repmgr_message {
	STAILQ_ENTRY(__repmgr_message) entries;
	int originating_eid;
	DBT control, rec;

typedef enum {
} phase_t;

 * If another site initiates a connection to us, when we receive it the
 * connection state is immediately "connected".  But when we initiate the
 * connection to another site, it first has to go through a "connecting" state,
 * until the non-blocking connect() I/O operation completes successfully.
 *     With an outgoing connection, we always know the associated site (and so
 * we have a valid eid).  But with an incoming connection, we don't know the
 * site until we get a handshake message, so until that time the eid is
 * invalid.
struct __repmgr_connection {
	TAILQ_ENTRY(__repmgr_connection) entries;

	int eid;		/* index into sites array in machtab */
	socket_t fd;
#ifdef DB_WIN32
	WSAEVENT event_object;

	u_int32_t version;	/* Wire protocol version on this connection. */
				/* (0 means not yet determined.) */

#define	CONN_INCOMING	0x01	/* We received this via accept(). */
	u_int32_t flags;

 * When we initiate an outgoing connection, it starts off in CONNECTING state
 * (or possibly CONNECTED).  When the (non-blocking) connection operation later
 * completes, we move to CONNECTED state.  When we get the response to our
 * version negotiation, we move to READY.
 *     For incoming connections that we accept, we start in NEGOTIATE, then to
 * PARAMETERS, and then to READY.
 *     CONGESTED is a hierarchical substate of READY: it's just like READY, with
 * the additional wrinkle that we don't bother waiting for the outgoing queue to
 * drain in certain circumstances.
#define	CONN_CONGESTED	1	/* Long-lived full outgoing queue. */
#define	CONN_CONNECTED	2	/* Awaiting reply to our version negotiation. */
#define	CONN_CONNECTING	3	/* Awaiting completion of non-block connect. */
#define	CONN_DEFUNCT	4	/* Basically dead, awaiting clean-up. */
#define	CONN_NEGOTIATE	5	/* Awaiting version proposal. */
#define	CONN_PARAMETERS	6	/* Awaiting parameters handshake. */
#define	CONN_READY	7	/* Everything's fine. */
	int state;

	 * Output: usually we just simply write messages right in line, in the
	 * send() function's thread.  But if TCP doesn't have enough network
	 * buffer space for us when we first try it, we instead allocate some
	 * memory, and copy the message, and then send it as space becomes
	 * available in our main select() thread.  In some cases, if the queue
	 * gets too long we wait until it's drained, and then append to it.
	 * This condition variable's associated mutex is the normal per-repmgr
	 * db_rep->mutex, because that mutex is always held anyway whenever the
	 * output queue is consulted.
	OUT_Q_HEADER outbound_queue;
	int out_queue_length;
	cond_var_t drained;
	int blockers;		/* ref count of msg threads waiting on us */

	 * Input: while we're reading a message, we keep track of what phase
	 * we're in.  In both phases, we use a REPMGR_IOVECS to keep track of
	 * our progress within the phase.  Depending upon the message type, we
	 * end up with either a rep_message (which is a wrapper for the control
	 * and rec DBTs), or a single generic DBT.
	 *     Any time we're in DATA_PHASE, it means we have already received
	 * the message header (consisting of msg_type and 2 sizes), and
	 * therefore we have allocated buffer space to read the data.  (This is
	 * important for resource clean-up.)
	phase_t		reading_phase;

	u_int8_t	msg_type;
	u_int32_t	control_size_buf, rec_size_buf;

	union {
		REPMGR_MESSAGE *rep_message;
		struct {
			DBT cntrl, rec;
		} repmgr_msg;
	} input;

#define	IS_READY_STATE(s)	((s) == CONN_READY || (s) == CONN_CONGESTED)

typedef struct addrinfo	ADDRINFO;
 * Some windows platforms have getaddrinfo (Windows XP), some don't.  We don't
 * support conditional compilation in our Windows build, so we always use our
 * own getaddrinfo implementation.  Rename everything so that we don't collide
 * with the system libraries.
#define	AI_PASSIVE	0x01
#define	AI_CANONNAME	0x02
#define	AI_NUMERICHOST	0x04

typedef struct __addrinfo {
	int ai_family;		/* PF_xxx */
	int ai_socktype;	/* SOCK_xxx */
	int ai_protocol;	/* 0 or IPPROTO_xxx for IPv4 and IPv6 */
	size_t ai_addrlen;	/* length of ai_addr */
	char *ai_canonname;	/* canonical name for nodename */
	struct sockaddr *ai_addr;	/* binary address */
	struct __addrinfo *ai_next;	/* next structure in linked list */

typedef struct {
	char *host;		/* Separately allocated copy of string. */
	u_int16_t port;		/* Stored in plain old host-byte-order. */
	ADDRINFO *address_list;
	ADDRINFO *current;
} repmgr_netaddr_t;

 * Each site that we know about is either idle or connected.  If it's connected,
 * we have a reference to a connection object; if it's idle, we have a reference
 * to a retry object.
 *     We store site objects in a simple array in the machtab, indexed by EID.
 * (We allocate EID numbers for other sites simply according to their index
 * within this array; we use the special value INT_MAX to represent our own
 * EID.)
struct __repmgr_site {
	repmgr_netaddr_t net_addr;
	DB_LSN max_ack;		/* Best ack we've heard from this site. */
	u_int32_t priority;
	db_timespec last_rcvd_timestamp;

#define	SITE_IDLE 1		/* Waiting til time to retry connecting. */
	int state;

#define	SITE_HAS_PRIO	0x01	/* Set if priority field has valid value. */
	u_int32_t flags;

	union {
		REPMGR_RETRY *retry; /* when IDLE */
	} ref;

 * Repmgr message formats.  We pass these in the "control" portion of a message.
 * For an ack, we just let the "rec" part go unused.  But for a handshake, the
 * "rec" part contains the variable-length host name (including terminating NUL
 * character).

 * The hand-shake message is exchanged upon establishment of a connection.  The
 * message protocol version number here refers to the connection as a whole.  In
 * other words, it's an assertion that every message sent or received on this
 * connection shall be of the specified version.  Since repmgr uses TCP, a
 * reliable stream-oriented protocol, this assertion is meaningful.
typedef struct {
	u_int32_t version;
	u_int16_t port;
	u_int32_t priority;

 * We store site structs in a dynamically allocated, growable array, indexed by
 * EID.  We allocate EID numbers for remote sites simply according to their
 * index within this array.  We don't need (the same kind of) info for ourself
 * (the local site), so we use an EID value that won't conflict with any valid
 * array index.
#define	SITE_FROM_EID(eid)	(&db_rep->sites[eid])
#define	EID_FROM_SITE(s)	((int)((s) - (&db_rep->sites[0])))
#define	IS_VALID_EID(e)		((e) >= 0)

#define	IS_PEER_POLICY(p) ((p) == DB_REPMGR_ACKS_ALL_PEERS ||		\
    (p) == DB_REPMGR_ACKS_QUORUM ||		\

#define	LOCK_MUTEX(m) do {						\
	int __ret;							\
	if ((__ret = __repmgr_lock_mutex(&(m))) != 0)			\
		return (__ret);						\
} while (0)

#define	UNLOCK_MUTEX(m) do {						\
	int __ret;							\
	if ((__ret = __repmgr_unlock_mutex(&(m))) != 0)			\
		return (__ret);						\
} while (0)

/* POSIX/Win32 socket (and other) portability. */
#ifdef DB_WIN32

#define	net_errno		WSAGetLastError()
typedef int socklen_t;
typedef char * sockopt_t;

#define	iov_len len
#define	iov_base buf

typedef DWORD threadsync_timeout_t;

#define	REPMGR_SYNC_INITED(db_rep) (db_rep->waiters != NULL)

#define	INVALID_SOCKET		-1
#define	SOCKET_ERROR		-1

#define	net_errno		errno
typedef void * sockopt_t;

#define	closesocket(fd)		close(fd)

typedef struct timespec threadsync_timeout_t;

#define	REPMGR_SYNC_INITED(db_rep) (db_rep->read_pipe >= 0)

/* Macros to proceed, as with a cursor, through the address_list: */
#define	ADDR_LIST_CURRENT(na)	((na)->current)
#define	ADDR_LIST_FIRST(na)	((na)->current = (na)->address_list)
#define	ADDR_LIST_NEXT(na)	((na)->current = (na)->current->ai_next)

 * Various threads write onto TCP/IP sockets, and an I/O error could occur at
 * any time.  However, only the dedicated "select()" thread may close the socket
 * file descriptor, because under POSIX we have to drop our mutex and then call
 * select() as two distinct (non-atomic) operations.
 * To simplify matters, there is a single place in the select thread where we
 * close and clean up after any defunct connection.  Even if the I/O error
 * happens in the select thread we follow this convention.
 * When an error occurs, we disable the connection (mark it defunct so that no
 * one else will try to use it, and so that the select thread will find it and
 * clean it up), and then usually take some additional recovery action: schedule
 * a connection retry for later, and possibly call for an election if it was a
 * connection to the master.  (This happens in the function
 * __repmgr_bust_connection.)  But sometimes we don't want to do the recovery
 * part; just the disabling part.
#define DISABLE_CONNECTION(conn) do {					 \
	(conn)->state = CONN_DEFUNCT;					 \
	(conn)->eid = -1;						 \
} while (0)

#include "dbinc_auto/repmgr_ext.h"

#if defined(__cplusplus)
#endif /* !_DB_REPMGR_H_ */