<?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 V1.73.2" /> <link rel="start" 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="prev" 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 class="toc"> <dl> <dt> <span class="sect2"> <a href="exampledoloop.html#runningit">Running It</a> </span> </dt> </dl> </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 <code class="function">doloop()</code> function that our original transaction application used <span>(see <a class="xref" href="simpleprogramlisting.html#doloop_cxx" title="Method: SimpleTxn::doloop()">Method: SimpleTxn::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 <strong class="userinput"><code>bold</code></strong>. </p> <p> To begin, we include a new header file into our application so that we can check for the <code class="literal">ENOENT</code> return value later in our processing loop. We also define our <code class="literal">APP_DATA</code> structure, and we define a <code class="literal">sleeptime</code> value. <span> Finally, we update <code class="classname">RepMgrGSG</code> to have a new method for our event notification callback, and to add a new data member for our <code class="literal">APP_DATA</code> data member. </span> </p> <pre class="programlisting">#include <db_cxx.h> #include <iostream> <strong class="userinput"><code>#include <errno.h></code></strong> ... // 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" <strong class="userinput"><code>#define SLEEPTIME 3</code></strong> const char *progname = "RepMgrGSG"; <strong class="userinput"><code>// Struct used to store information in Db app_private field. typedef struct { int is_master; } APP_DATA;</code></strong> class RepMgrGSG { public: // Constructor. RepMgrGSG(); // 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(); <strong class="userinput"><code>// event notification callback static void event_callback(DbEnv * dbenv, u_int32_t which, void *info);</code></strong> private: // disable copy constructor. RepMgrGSG(const RepMgrGSG &); void operator = (const RepMgrGSG &); // internal data members. <strong class="userinput"><code>APP_DATA app_data;</code></strong> 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><code class="methodname">main()</code> method, because it does not change.</span> Instead, we skip down to our <span><code class="classname">RepMgrGSG</code> constructor where we initialize our <code class="literal">APP_DATA is_master</code> data member:</span> </p> <pre class="programlisting"> RepMgrGSG::RepMgrGSG() : app_config(0), dbenv(0) { <strong class="userinput"><code>app_data.is_master = 0; // assume I start out as client</code></strong> } </pre> <p> That done, we must also update <code class="methodname">RepMgrGSG::init()</code> to do a couple of things. First, we need to register our event callback with the environment handle. We also need to make our <code class="literal">APP_DATA</code> data member available through our environment handle's <code class="methodname">app_private</code> 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 RepMgrGSG::init(RepConfigInfo *config) { int ret = 0; app_config = config; dbenv.set_errfile(stderr); dbenv.set_errpfx(progname); <strong class="userinput"><code>dbenv.set_app_private(&app_data); dbenv.set_event_notify(event_callback);</code></strong> ... </pre> <p> That done, we need to implement our <code class="function">event_callback()</code> 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"> <strong class="userinput"> <code>/* * A callback used to determine whether the local environment is a * replica or a master. This is called by the Replication Manager * when the local replication environment changes state. */ void RepMgrGSG::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); } }</code> </strong> </pre> <p> That done, we need to update our <code class="function">doloop()</code> <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 RepMgrGSG::doloop() { Db *dbp; Dbt key, data; char buf[BUFSIZE], *rbuf; int ret; dbp = 0; memset(&key, 0, sizeof(key)); memset(&data, 0, sizeof(data)); ret = 0; for (;;) { if (dbp == 0) { dbp = new Db(&dbenv, 0); try { dbp->open(NULL, DATABASE, NULL, DB_BTREE, <strong class="userinput"><code>app_data.is_master ? DB_CREATE | DB_AUTO_COMMIT : DB_AUTO_COMMIT</code></strong>, 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 Manager 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 Manager a chance to create the database so that our replica can continue operations. </p> <pre class="programlisting"> } catch(DbException dbe) { <strong class="userinput"><code>/* 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 {</code></strong> dbenv.err(ret, "DB->open"); throw dbe; <strong class="userinput"><code>}</code></strong> } } </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"> <strong class="userinput"><code>cout << "QUOTESERVER" ; if (!app_data.is_master) cout << "(read-only)"; cout << "> " << flush; </code></strong></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 <code class="function">print_stocks()</code> <span>method</span> (which has not required a modification since we first introduced it in <a class="xref" href="simpleprogramlisting.html#printstocks_c" title="Method: SimpleTxn::print_stocks()"> <span>Method: SimpleTxn::print_stocks()</span> </a>). </p> <p> When we call <span><code class="function">print_stocks()</code>, </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; <strong class="userinput"><code>case DB_REP_HANDLE_DEAD: (void)dbp->close(DB_NOSYNC); cout << "closing db handle due to rep handle dead" << endl; dbp = NULL; continue;</code></strong> 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"> <strong class="userinput"><code>if (!app_data.is_master) { dbenv->errx(dbenv, "Can't update at client"); continue; }</code></strong> 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 != 0) (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, <code class="function">print_stocks()</code>,</span> is unmodified from when we originally introduced it. For details on that function, see <a class="xref" href="simpleprogramlisting.html#printstocks_c" title="Method: SimpleTxn::print_stocks()"> <span>Method: SimpleTxn::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> <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 <code class="literal">-h</code> option to specify that directory. You must also use the <code class="literal">-l</code> option to identify the local host and port that this process will use to listen for replication messages, the <code class="literal">-n</code> option to specify the number of sites in the replication group, and the <code class="literal">-r</code> option to identify the other processes in the replication group. Finally, use the <code class="literal">-p</code> 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 > ./RepMgrGSG -h env1 -n 2 -l localhost:8080 -r 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 <code class="literal">-l</code> flag to at least change the port number the process is listening on, and use the <code class="literal">-r</code> option to identify the host and port of the other replication process: </p> <pre class="programlisting">> mkdir env2 > ./RepMgrGSG -h env2 -n 2 -l localhost:8081 -r 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 <strong class="userinput"><code>return</code></strong> 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. On the master: </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 <code class="literal">quit</code> at both prompts. On the replica: </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>