fsa_defines.h

linux集群服务器软件代码包

/* $Id: fsa_defines.h,v 1.34 2005/02/15 09:45:43 andrew Exp $ */
/* 
 * Copyright (C) 2004 Andrew Beekhof <[e-mail]>
 * 
 * This program is free software; you can redistribute it and/or
 * modify it under the terms of the GNU General Public
 * License as published by the Free Software Foundation; either
 * version 2.1 of the License, or (at your option) any later version.
 * 
 * This software is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
 * General Public License for more details.
 * 
 * You should have received a copy of the GNU General Public
 * License along with this library; if not, write to the Free Software
 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
 */
#ifndef FSA_DEFINES__H
#define FSA_DEFINES__H

/*======================================
 *	States the DC/CRMd can be in
 *======================================*/
enum crmd_fsa_state {
        S_IDLE = 0,	/* Nothing happening */

        S_ELECTION,	/* Take part in the election algorithm as 
			 * described below
			 */
	S_INTEGRATION,	/* integrate that status of new nodes (which is 
			 * all of them if we have just been elected DC)
			 * to form a complete and up-to-date picture of
			 * the CIB
			 */
	S_FINALIZE_JOIN,/* integrate that status of new nodes (which is 
			 * all of them if we have just been elected DC)
			 * to form a complete and up-to-date picture of
			 * the CIB
			 */
        S_NOT_DC,	/* we are in crmd/slave mode */
        S_POLICY_ENGINE,/* Determin the next stable state of the cluster
			 */
        S_RECOVERY,	/* Something bad happened, check everything is ok
			 * before continuing and attempt to recover if
			 * required
			 */
        S_RELEASE_DC,	/* we were the DC, but now we arent anymore,
			 * possibly by our own request, and we should 
			 * release all unnecessary sub-systems, finish
			 * any pending actions, do general cleanup and
			 * unset anything that makes us think we are
			 * special :)
			 */
	S_STARTING,	/* we are just starting out */
        S_PENDING,	/* we are not a full/active member yet */
        S_STOPPING,	/* We are in the final stages of shutting down */
        S_TERMINATE,	/* We are going to shutdown, this is the equiv of
			 * "Sending TERM signal to all processes" in Linux
			 * and in worst case scenarios could be considered
			 * a self STONITH
			 */
        S_TRANSITION_ENGINE,/* Attempt to make the calculated next stable
			     * state of the cluster a reality
			     */

        S_HALT,		/* Freeze - dont do anything
			 * Something ad happened that needs the admin to fix
			 * Wait for I_ELECTION
			 */

	/*  ----------- Last input found in table is above ---------- */
	S_ILLEGAL	/* This is an illegal FSA state */
			/* (must be last) */
};
#define MAXSTATE	S_ILLEGAL
/*
	A state diagram can be constructed from the dc_fsa.dot with the
	following command:

	dot -Tpng crmd_fsa.dot > crmd_fsa.png

Description:

      Once we start and do some basic sanity checks, we go into the
      S_NOT_DC state and await instructions from the DC or input from
      the CCM which indicates the election algorithm needs to run.

      If the election algorithm is triggered we enter the S_ELECTION state
      from where we can either go back to the S_NOT_DC state or progress
      to the S_INTEGRATION state (or S_RELEASE_DC if we used to be the DC
      but arent anymore).

      The election algorithm has been adapted from
      http://www.cs.indiana.edu/cgi-bin/techreports/TRNNN.cgi?trnum=TR521

      Loosly known as the Bully Algorithm, its major points are:
      - Election is initiated by any node (N) notices that the coordinator
	is no longer responding
      - Concurrent multiple elections are possible
      - Algorithm
	  + N sends ELECTION messages to all nodes that occur earlier in
	  the CCM's membership list.
	  + If no one responds, N wins and becomes coordinator
	  + N sends out COORDINATOR messages to all other nodes in the
	  partition
	  + If one of higher-ups answers, it takes over. N is done.
      
      Once the election is complete, if we are the DC, we enter the
      S_INTEGRATION state which is a DC-in-waiting style state.  We are
      the DC, but we shouldnt do anything yet because we may not have an
      up-to-date picture of the cluster.  There may of course be times
      when this fails, so we should go back to the S_RECOVERY stage and
      check everything is ok.  We may also end up here if a new node came
      online, since each node is authorative on itself and we would want
      to incorporate its information into the CIB.

      Once we have the latest CIB, we then enter the S_POLICY_ENGINE state
      where invoke the Policy Engine. It is possible that between
      invoking the Policy Engine and recieving an answer, that we recieve
      more input. In this case we would discard the orginal result and
      invoke it again.

      Once we are satisfied with the output from the Policy Engine we
      enter S_TRANSITION_ENGINE and feed the Policy Engine's output to the
      Transition Engine who attempts to make the Policy Engine's
      calculation a reality.  If the transition completes successfully,
      we enter S_IDLE, otherwise we go back to S_POLICY_ENGINE with the
      current unstable state and try again.
      
      Of course we may be asked to shutdown at any time, however we must
      progress to S_NOT_DC before doing so.  Once we have handed over DC
      duties to another node, we can then shut down like everyone else,
      that is by asking the DC for permission and waiting it to take all
      our resources away.

      The case where we are the DC and the only node in the cluster is a
      special case and handled as an escalation which takes us to
      S_SHUTDOWN.  Similarly if any other point in the shutdown
      fails or stalls, this is escalated and we end up in S_TERMINATE.

      At any point, the CRMd/DC can relay messages for its sub-systems,
      but outbound messages (from sub-systems) should probably be blocked
      until S_INTEGRATION (for the DC case) or the join protocol has
      completed (for the CRMd case)
      
*/

/*======================================
 *
 * 	Inputs/Events/Stimuli to be given to the finite state machine
 *
 *	Some of these a true events, and others a synthesised based on
 *	the "register" (see below) and the contents or source of messages.
 *
 *	At this point, my plan is to have a loop of some sort that keeps
 *	going until recieving I_NULL
 *
 *======================================*/
enum crmd_fsa_input {
/* 0 */
	I_NULL,		/* Nothing happened */
/* 1 */
	I_CCM_EVENT,
	
	I_CIB_OP,	/* An update to the CIB occurred */
	I_CIB_UPDATE,	/* An update to the CIB occurred */
	I_DC_TIMEOUT,	/* We have lost communication with the DC */
	I_ELECTION,	/* Someone started an election */
	I_PE_CALC,	/* The Policy Engine needs to be invoked */
	I_RELEASE_DC,	/* The election completed and we were not
				 * elected, but we were the DC beforehand
				 */
	I_ELECTION_DC,	/* The election completed and we were (re-)elected
			 * DC
			 */
	I_ERROR,	/* Something bad happened (more serious than
			 * I_FAIL) and may not have been due to the action
			 * being performed.  For example, we may have lost
			 * our connection to the CIB.
			 */
/* 10 */
	I_FAIL,		/* The action failed to complete successfully */
	I_INTEGRATED, 
	I_FINALIZED, 
	I_NODE_JOIN,	/* A node has entered the cluster */
	I_NOT_DC,	/* We are not and were not the DC before or after
			 * the current operation or state
			 */
	I_RECOVERED,	/* The recovery process completed successfully */
	I_RELEASE_FAIL,	/* We could not give up DC status for some reason
			 */
	I_RELEASE_SUCCESS,	/* We are no longer the DC */
	I_RESTART,	/* The current set of actions needs to be
			 * restarted
			 */
	I_TE_SUCCESS,	/* Some non-resource, non-ccm action is required
			 * of us, eg. ping
			 */
/* 20 */
	I_ROUTER,	/* Do our job as router and forward this to the
			 * right place
			 */
	I_SHUTDOWN,	/* We are asking to shutdown */
	I_TERMINATE,	/* We have been told to shutdown */
	I_STARTUP,
	I_PE_SUCCESS,	/* The action completed successfully */

	I_JOIN_OFFER,	/* The DC is offering membership */
	I_JOIN_REQUEST,	/* The client is requesting membership */
	I_JOIN_RESULT,	/* If not the DC: The result of a join request
			 * Else: A client is responding with its local state info
			 */
	
	I_WAIT_FOR_EVENT, /* we may be waiting for an async task to "happen"
			   * and until it does, we cant do anything else
			   */

	I_DC_HEARTBEAT,	/* The DC is telling us that it is alive and well */  


	I_LRM_EVENT,

/* 30 */
	I_PENDING,
	I_HALT,
	
	/*  ------------ Last input found in table is above ----------- */
	I_ILLEGAL	/* This is an illegal value for an FSA input */
			/* (must be last) */
};
#define MAXINPUT	I_ILLEGAL

#define I_MESSAGE	I_ROUTER

/*======================================
 *
 * actions
 *