$Id: README,v 1.7.4.1 2006-10-27 21:15:06 cbbrowne Exp $

Directory src/slon

This directory contains the main replication engine.

STATUS:
	
	Slon can subscribe to a data set, copy the initial data from
	the provider and sync. That is usually called "replication".

	Slon can forward log information.

	Slon can also subscribe to a set as a cascaded subscriber and
	change the data provider later on the fly.

TODO:

	All other admin events (especially drop)

	Use forward data for PITR.

Major components:

 - slon.c

   Main program file:

   - processes command line options
   - loads configuration for the local node from its database
   - starts various threads:
     - main loop thread
     - cleanup thread
     - SYNC event generator thread

 - cleanup_thread.c

  This thread runs every so often and does various forms of cleanup:
   - Cleans up event table to remove old entries
   - Removes obsolete entries from sl_log_1, sl_log_2, sl_seqlog
   - Vacuums tables that Slony-I uses
   - Once in a while, it calls logswitch_weekly(), which tries to
     rotate between sl_log_1 and sl_log_2 periodically

 - conf-file.l
   A lex grammar for slon configuration files

 - dbutils.c

   Tools for database access that provide abstractions to simplify
   usage of libpq functions.  For instance, slon_mkquery() provides a
   way to paste together queries using dynamic string buffer
   allocation.

 - local_listen.c

   Thread function that listens for events on the local node's
   database and takes action based on them.  This mostly involves
   requests for reconfiguration of the node.

 - remote_listen.c

   An "remote listener" thread is created for each node this node
   listens to.  The thread is in effect looking for configuration
   changes and SYNC events on sets to which the local node is
   subscribing to.  When events are received, they are forwarded to a
   queue on the remote worker for the node to be worked on.

 - remote_worker.c

   A "remote worker" thread is created for each node this node listens
   to.  When events are received by the remote listener, they are
   queued by the worker thread which then, in order, acts on them.

   Configuration changes are typically passed on to the runtime
   configuration system.  The two events that "do replication" are
   ENABLE_SUBSCRIPTION, which copies existing data from a provider to
   the subscriber, and SYNC, which leads to the worker thread reading
   updates from sl_log_1/sl_log_2 and sl_seqlog.

 - runtime_config.c

   Each slon process has a set of in-memory configuration information;
   these functions manage this information.

   The data includes:

   SlonNode - a linked list of nodes

    For each node, the following is stored:
      no_id         - node ID
      no_active     - its active state
      no_comment    - Comment field
      pa_conninfo   - path (libpq style DSN) to the node
      pa_connretry  - connection retry interval
      last_event    - last event received for this node
      listen_status - status of the listen thread
      listen_thread - thread ID of listen thread
      listen_head   - list of origins we listen for
      listen_tail   - tail of list of origins we listen for
      worker_status - status of worker thread
      worker_thread - thread ID of worker thread
      message_lock  - mutex for message queue
      message_cond  - condition variable for queue
      message_head  - list of work messages
      message_tail  - tail of list of work messages

   listen - a linked list of listen entries

   SlonSet - a linked list of sets

     For each node, the following is tracked:

      set_id        - ID of set
      set_origin    - node that is the origin for the set
      set_comment   - comment about the set
      sub_provider  - node from which this node receives data
      sub_forward   - does this node forward data?
      sub_active    - is the subscription active yet?

 - scheduler.c

   Event scheduler subsystem

 - sync_thread.c

   This thread periodically generates a SYNC event if local database
   activity has created new log information in sl_log_1/sl_log_2.  

   A SYNC is the basic "unit of replicated data."

 - misc.c

   Miscellaneous support functions.  Mostly about log file processing.

Work threads:
  - main
  - cleanup thread
  - sync thread
  - remote_listen thread - for each node being monitored
  - remote_worker thread - for each node being monitored