Base API applications must implement a communication
infrastructure. The communication infrastructure consists of
three parts: a way to map environment IDs to particular sites,
the functions to send and receive messages, and the
application architecture that supports the particular
communication infrastructure used (for example, individual
threads per communicating site, a shared message handler for
all sites, a hybrid solution). The communication
infrastructure for ex_rep_base is implemented in the file
ex_rep/base/rep_net.c
, and each part
of that infrastructure is described as follows.
Ex_rep_base maintains a table of environment ID to TCP/IP
port mappings. A pointer to this table is stored in a
structure pointed to by the app_private field of the DB_ENV
object so it can be accessed by any function that has the
database environment handle. The table is represented by a
machtab_t structure which contains a reference to a linked
list of member_t's, both of which are defined in
ex_rep/base/rep_net.c
. Each member_t
contains the host and port identification, the environment ID,
and a file descriptor.
This design is particular to this application and communication infrastructure, but provides an indication of the sort of functionality that is needed to maintain the application-specific state for a TCP/IP-based infrastructure. The goal of the table and its interfaces is threefold: First, it must guarantee that given an environment ID, the send function can send a message to the appropriate place. Second, when given the special environment ID DB_EID_BROADCAST, the send function can send messages to all the machines in the group. Third, upon receipt of an incoming message, the receive function can correctly identify the sender and pass the appropriate environment ID to the DB_ENV->rep_process_message() method.
Mapping a particular environment ID to a specific port is accomplished by looping through the linked list until the desired environment ID is found. Broadcast communication is implemented by looping through the linked list and sending to each member found. Since each port communicates with only a single other environment, receipt of a message on a particular port precisely identifies the sender.
This is implemented in the quote_send, quote_send_broadcast
and quote_send_one functions, which can be found in
ex_rep/base/rep_net.c
.
The example provided is merely one way to satisfy these requirements, and there are alternative implementations as well. For instance, instead of associating separate socket connections with each remote environment, an application might instead label each message with a sender identifier; instead of looping through a table and sending a copy of a message to each member of the replication group, the application could send a single message using a broadcast protocol.
The quote_send function is passed as the callback to DB_ENV->rep_set_transport(); Berkeley DB automatically sends messages as needed for replication. The receive function is a mirror to the quote_send_one function. It is not a callback function (the application is responsible for collecting messages and calling DB_ENV->rep_process_message() on them as is convenient). In the sample application, all messages transmitted are Berkeley DB messages that get handled by DB_ENV->rep_process_message(), however, this is not always going to be the case. The application may want to pass its own messages across the same channels, distinguish between its own messages and those of Berkeley DB, and then pass only the Berkeley DB ones to DB_ENV->rep_process_message().
The final component of the communication infrastructure is
the process model used to communicate with all the sites in
the replication group. Each site creates a thread of control
that listens on its designated socket (as specified by the
-l command line argument)
and then creates a new channel for each site that contacts it.
In addition, each site explicitly connects to the sites
specified in the -r and
-R command line
arguments. This is a fairly standard TCP/IP process
architecture and is implemented by the connect_thread,
connect_all and connect_site functions in
ex_rep/base/rep_msg.c
and supporting
functions in
ex_rep/base/rep_net.c
.