exampledoloop.html [plain text]
<?xml version="1.0" encoding="UTF-8" standalone="no"?>
<!DOCTYPE html PUBLIC "-//W3C//DTD XHTML 1.0 Transitional//EN" "http://www.w3.org/TR/xhtml1/DTD/xhtml1-transitional.dtd">
<html xmlns="http://www.w3.org/1999/xhtml">
<head>
<meta http-equiv="Content-Type" content="text/html; charset=UTF-8" />
<title>Example Processing Loop</title>
<link rel="stylesheet" href="gettingStarted.css" type="text/css" />
<meta name="generator" content="DocBook XSL Stylesheets V1.62.4" />
<link rel="home" href="index.html" title="Getting Started with Replicated Berkeley DB Applications" />
<link rel="up" href="fwrkmasterreplica.html" title="Chapter 4. Replica versus Master Processes" />
<link rel="previous" href="processingloop.html" title="Processing Loop" />
<link rel="next" href="addfeatures.html" title="Chapter 5. Additional Features" />
</head>
<body>
<div class="navheader">
<table width="100%" summary="Navigation header">
<tr>
<th colspan="3" align="center">Example Processing Loop</th>
</tr>
<tr>
<td width="20%" align="left"><a accesskey="p" href="processingloop.html">Prev</a> </td>
<th width="60%" align="center">Chapter 4. Replica versus Master Processes</th>
<td width="20%" align="right"> <a accesskey="n" href="addfeatures.html">Next</a></td>
</tr>
</table>
<hr />
</div>
<div class="sect1" lang="en" xml:lang="en">
<div class="titlepage">
<div>
<div>
<h2 class="title" style="clear: both"><a id="exampledoloop"></a>Example Processing Loop</h2>
</div>
</div>
<div></div>
</div>
<p>
In this section we take the example
processing loop that we presented in the
previous section and we flesh it out to
provide a more complete example. We do this
by updating the
<tt class="function">doloop()</tt>
function that our original transaction
application used
<span>(see <a href="simpleprogramlisting.html#doloop_cxx">Method: RepMgr::doloop()</a>)</span>
to fully support our replicated application.
</p>
<p>
In the following example code, code that we
add to the original example is presented in
<b class="userinput"><tt>bold</tt></b>.
</p>
<p>
To begin, we include a new header file into
our application so that we can check for the
<tt class="literal">ENOENT</tt> return value later
in our processing loop. We also define our
<tt class="literal">APP_DATA</tt>
structure, and we define a
<tt class="literal">sleeptime</tt> value.
<span>
Finally, we update <tt class="classname">RepMgr</tt>
to have a new method for our event notification
callback, and to add a new data member for our
<tt class="literal">APP_DATA</tt> data member.
</span>
</p>
<pre class="programlisting">#include <db_cxx.h>
#include <iostream>
<b class="userinput"><tt>#include <errno.h></tt></b>
...
// Skipping all the RepHostInfoObj and RepConfigInfo code, which does not
// change.
...
using std::cout;
using std::cin;
using std::cerr;
using std::endl;
using std::flush;
#define CACHESIZE (10 * 1024 * 1024)
#define DATABASE "quote.db"
<b class="userinput"><tt>#define SLEEPTIME 3</tt></b>
const char *progname = "RepMgr";
<b class="userinput"><tt>// Struct used to store information in Db app_private field.
typedef struct {
int is_master;
} APP_DATA;</tt></b>
class RepMgr
{
public:
// Constructor.
RepMgr();
// Initialization method. Creates and opens our environment handle.
int init(RepConfigInfo* config);
// The doloop is where all the work is performed.
int doloop();
// terminate() provides our shutdown code.
int terminate();
<b class="userinput"><tt>// event notification callback
static void
event_callback(DbEnv * dbenv, u_int32_t which, void *info);</tt></b>
private:
// disable copy constructor.
RepMgr(const RepMgr &);
void operator = (const RepMgr &);
// internal data members.
<b class="userinput"><tt>APP_DATA app_data;</tt></b>
RepConfigInfo *app_config;
DbEnv dbenv;
// private methods.
// print_stocks() is used to display the contents of our database.
static int print_stocks(Db *dbp);
}; </pre>
<p>
That done, we can skip the
<span><tt class="methodname">main()</tt> method, because it does not change.</span>
Instead, we skip down to our
<span><tt class="classname">RepMgr</tt> constructor where we initialize our
<tt class="literal">APP_DATA is_master</tt> data member:</span>
</p>
<pre class="programlisting">
RepMgr::RepMgr() : app_config(0), dbenv(0)
{
<b class="userinput"><tt>app_data.is_master = 0; // assume I start out as client</tt></b>
}
</pre>
<p>
That done, we must also
update <tt class="methodname">RepMgr::init()</tt> to do a couple
of things. First, we need to register our event callback with
the environment handle. We also need to make our
<tt class="literal">APP_DATA</tt> data member available through our
environment handle's <tt class="methodname">app_private</tt>
field. This is a fairly trivial update, and it happens at the
top of the method (we skip the rest of the method's listing
since it does not change):
</p>
<pre class="programlisting">int RepMgr::init(RepConfigInfo *config)
{
int ret = 0;
app_config = config;
dbenv.set_errfile(stderr);
dbenv.set_errpfx(progname);
<b class="userinput"><tt>dbenv.set_app_private(&app_data);
dbenv.set_event_notify(event_callback);</tt></b>
... </pre>
<p>
That done, we need to implement our
<tt class="function">event_callback()</tt> callback. Note that what we use
here is no different from the callback that we described in
the previous section. However, for the sake of completeness we
provide the implementation here again.
</p>
<pre class="programlisting">
<b class="userinput">
<tt>/*
* A callback used to determine whether the local environment is a replica
* or a master. This is called by the replication framework
* when the local replication environment changes state.
*/
void RepMgr::event_callback(DbEnv *dbenv, u_int32_t which, void *info)
{
APP_DATA *app = dbenv->get_app_private();
info = NULL; /* Currently unused. */
switch (which) {
case DB_EVENT_REP_MASTER:
app->is_master = 1;
break;
case DB_EVENT_REP_CLIENT:
app->is_master = 0;
break;
case DB_EVENT_REP_STARTUPDONE: /* fallthrough */
case DB_EVENT_REP_NEWMASTER:
/* Ignore. */
break;
default:
dbenv->errx(dbenv, "ignoring event %d", which);
}
}</tt>
</b>
</pre>
<p>
That done, we need to update our
<tt class="function">doloop()</tt>
<span>method.</span>
</p>
<p>
We begin by updating our database handle open flags to
determine which flags to use, depending on whether the
application is running as a master.
</p>
<pre class="programlisting">#define BUFSIZE 1024
int RepMgr::doloop()
{
Db *dbp;
Dbt key, data;
char buf[BUFSIZE], *rbuf;
int ret;
dbp = NULL;
memset(&key, 0, sizeof(key));
memset(&data, 0, sizeof(data));
ret = 0;
for (;;) {
if (dbp == NULL) {
dbp = new Db(&dbenv, 0);
// Set page size small so page allocation is cheap.
if ((ret = dbp->set_pagesize(512)) != 0)
goto err;
try {
dbp->open(NULL, DATABASE, NULL, DB_BTREE,
<b class="userinput"><tt>app_data.is_master ? DB_CREATE | DB_AUTO_COMMIT :
DB_AUTO_COMMIT</tt></b>, 0); </pre>
<p>
When we open the database, we modify our error handling to
account for the case where the database does not yet exist. This can
happen if our code is running as a replica and the replication framework has not
yet had a chance to create the databases for us. Recall that replicas never
write to their own databases directly, and so they cannot
create databases on their own.
</p>
<p>
If we detect that the database does not yet exist, we simply
close the database handle, sleep for a short period of time
and then continue processing. This gives the replication framework a chance to
create the database so that our replica can continue
operations.
</p>
<pre class="programlisting"> } catch(DbException dbe) {
<b class="userinput"><tt>/* It is expected that this condition will be triggered
* when client sites start up.
* It can take a while for the master site to be found
* and synced, and no DB will be available until then.
*/
if (dbe.get_errno() == ENOENT) {
cout << "No stock db available yet - retrying." << endl;
try {
dbp->close(0);
} catch (DbException dbe2) {
cout << "Unexpected error closing after failed"
<< " open, message: " << dbe2.what() << endl;
dbp = NULL;
goto err;
}
dbp = NULL;
sleep(SLEEPTIME);
continue;
} else {</tt></b>
dbenv.err(ret, "DB->open");
throw dbe;
<b class="userinput"><tt>}</tt></b>
}
} </pre>
<p>
Next we modify our prompt, so that if the local process is running
as a replica, we can tell from the shell that the prompt is for a
read-only process.
</p>
<pre class="programlisting"> <b class="userinput"><tt>cout << "QUOTESERVER" ;
if (!app_data.is_master)
cout << "(read-only)";
cout << "> " << flush; </tt></b></pre>
<p>
When we collect data from the prompt, there is a case that says
if no data is entered then show the entire stocks database.
This display is performed by our
<tt class="function">print_stocks()</tt>
<span>method</span>
(which has not
required a modification since we first introduced it in
<a href="simpleprogramlisting.html#printstocks_c">
<span>Method: RepMgr::print_stocks()</span>
</a>).
</p>
<p>
When we call
<span><tt class="function">print_stocks()</tt>, </span>
we check for a dead replication handle. Dead
replication handles happen whenever a replication election
results in a previously committed transaction becoming
invalid. This is an error scenario caused by a new master having a
slightly older version of the data than the original
master and so all replicas must modify their database(s) to
reflect that of the new master. In this situation, some
number of previously committed transactions may have to be
unrolled. From the replica's perspective, the database
handles should all be closed and then opened again.
</p>
<pre class="programlisting">
if (fgets(buf, sizeof(buf), stdin) == NULL)
break;
if (strtok(&buf[0], " \t\n") == NULL) {
switch ((ret = print_stocks(dbp))) {
case 0:
continue;
<b class="userinput"><tt>case DB_REP_HANDLE_DEAD:
(void)dbp->close(DB_NOSYNC);
cout << "closing db handle due to rep handle dead" << endl;
dbp = NULL;
continue;</tt></b>
default:
dbp->err(ret, "Error traversing data");
goto err;
}
}
rbuf = strtok(NULL, " \t\n");
if (rbuf == NULL || rbuf[0] == '\0') {
if (strncmp(buf, "exit", 4) == 0 ||
strncmp(buf, "quit", 4) == 0)
break;
dbenv.errx("Format: TICKER VALUE");
continue;
} </pre>
<p>
That done, we need to add a little error checking to our
command prompt to make sure the user is not attempting to
modify the database at a replica. Remember, replicas must never
modify their local databases on their own. This guards against
that happening due to user input at the prompt.
</p>
<pre class="programlisting"> <b class="userinput"><tt>if (!app_data.is_master) {
dbenv->errx(dbenv, "Can't update at client");
continue;
}</tt></b>
key.set_data(buf);
key.set_size((u_int32_t)strlen(buf));
data.set_data(rbuf);
data.set_size((u_int32_t)strlen(rbuf));
if ((ret = dbp->put(NULL, &key, &data, 0)) != 0)
{
dbp->err(ret, "DB->put");
if (ret != DB_KEYEXIST)
goto err;
}
}
err: if (dbp != NULL)
(void)dbp->close(dbp, DB_NOSYNC);
return (ret);
} </pre>
<p>
With that completed, we are all done updating our application
for replication.
The only remaining
<span>method, <tt class="function">print_stocks()</tt>,</span>
is unmodified from when we
originally introduced it. For details on that function, see
<a href="simpleprogramlisting.html#printstocks_c">
<span>Method: RepMgr::print_stocks()</span>
</a>.
</p>
<div class="sect2" lang="en" xml:lang="en">
<div class="titlepage">
<div>
<div>
<h3 class="title"><a id="runningit"></a>Running It</h3>
</div>
</div>
<div></div>
</div>
<p>
To run our replicated application, we need to make
sure each participating environment has its own unique
home directory. We can do this by running
each site on a separate networked machine, but that
is not strictly necessary; multiple instances of this
code can run on the same machine provided the
environment home restriction is observed.
</p>
<p>
To run a process, make sure the environment home
exists and then start the process using the
<tt class="literal">-h</tt> option to specify that
directory. You must also use the <tt class="literal">-m</tt>
option to identify the local host and port that this
process will use to listen for replication messages, and
the <tt class="literal">-o</tt> option to identify the other
processes in the replication group. Finally, use the
<tt class="literal">-p</tt> option to specify a priority.
The process that you designate to have the highest priority will become
the master.
</p>
<pre class="programlisting">> mkdir env1
> ./RepMgr -h env1 -n 2 -m localhost:8080 -o localhost:8081 -p 10
No stock database yet available.
No stock database yet available. </pre>
<p>
Now, start another process. This time, change the environment
home to something else, use the <tt class="literal">-m</tt> to at
least change the port number the process is listening on, and
use the <tt class="literal">-o</tt> option to identify the host and
port of the other replication process:
</p>
<pre class="programlisting">> mkdir env2
> ./RepMgr -h env2 -n 2 -m localhost:8081 -o localhost:8080 -p 20</pre>
<p>
After a short pause, the second process should display the master
prompt:
</p>
<pre class="programlisting">
QUOTESERVER > </pre>
<p>
And the first process should
display the read-only prompt:
</p>
<pre class="programlisting">
QUOTESERVER (read-only)> </pre>
<p>
Now go to the master process and give it a couple of stocks and stock
prices:
</p>
<pre class="programlisting">QUOTESERVER> FAKECO 9.87
QUOTESERVER> NOINC .23
QUOTESERVER> </pre>
<p>
Then, go to the replica and hit <b class="userinput"><tt>return</tt></b> at the prompt to
see the new values:
</p>
<pre class="programlisting">QUOTESERVER (read-only)>
Symbol Price
====== =====
FAKECO 9.87
NOINC .23
QUOTESERVER (read-only)> </pre>
<p>
Doing the same at the master results in the same thing:
</p>
<pre class="programlisting">QUOTESERVER>
Symbol Price
====== =====
FAKECO 9.87
NOINC .23
QUOTESERVER> </pre>
<p>
You can change a stock by simply entering the stock value and
new price at the master's prompt:
</p>
<pre class="programlisting">QUOTESERVER> FAKECO 10.01
QUOTESERVER> </pre>
<p>
Then, go to either the master or the replica to see the updated
database:
</p>
<pre class="programlisting">QUOTESERVER>
Symbol Price
====== =====
FAKECO 10.01
NOINC .23
QUOTESERVER> </pre>
<p>
And on the replica:
</p>
<pre class="programlisting">QUOTESERVER (read-only)>
Symbol Price
====== =====
FAKECO 10.01
NOINC .23
QUOTESERVER (read-only)> </pre>
<p>
Finally, to quit the applications, simply type
<tt class="literal">quit</tt> at both prompts:
</p>
<pre class="programlisting">QUOTESERVER (read-only)> quit
> </pre>
<p>
And on the master as well:
</p>
<pre class="programlisting">QUOTESERVER> quit
> </pre>
</div>
</div>
<div class="navfooter">
<hr />
<table width="100%" summary="Navigation footer">
<tr>
<td width="40%" align="left"><a accesskey="p" href="processingloop.html">Prev</a> </td>
<td width="20%" align="center">
<a accesskey="u" href="fwrkmasterreplica.html">Up</a>
</td>
<td width="40%" align="right"> <a accesskey="n" href="addfeatures.html">Next</a></td>
</tr>
<tr>
<td width="40%" align="left" valign="top">Processing Loop </td>
<td width="20%" align="center">
<a accesskey="h" href="index.html">Home</a>
</td>
<td width="40%" align="right" valign="top"> Chapter 5. Additional Features</td>
</tr>
</table>
</div>
</body>
</html>