PHP实时聊天备份服务器生产问题

Question

I have built out a real time chat application powered by PubNub. At present, I allow messages to be sent and received directly to and from a user through PubNub alone. However, I wish to "save" these messages by first pushing them to my server/db, built in PHP and MySQL, then upon successful entry, send to the user via PubNub.

The setup of my present PHP server accepts the request and directs it to the proper case statment given the defined action passed into the request as:

<?php

switch($action) 
{

    case "userLogin": 
        ...
    break;  


    case "signUpUser":
        ...
    break;  

        ...

    case "syncWithServer":
        ...
    break;  

?>

where each case statement allows for database insert and update and subsequent return upon the actions success given the data passed.

For a detailed example, we have the sendMessage case below. The data must be inserted and then return the latest sync_id, for that message just inserted, then sent back to the user so as to keep everything in sync:

case "sendMessage":

    // Output is: {"message_sync_id":"4"}
    if ($userId = authenticateUser($db, $username, $password, $gcmregid)) 
    {   
        //echo "passed1";
        if (isset($_REQUEST['to_user_sync_id']))
        {
            //echo "passed2";a
             $touserid = $_REQUEST['to_user_sync_id'];               
             $unescapedMessage = $_REQUEST['message_text'];
             $message = mysql_real_escape_string($unescapedMessage);             
             $campaign = $_REQUEST['shared_campaign_id'];
             $location = $_REQUEST['shared_campaign_location_id'];

             if($stmt1 = $db->prepare("INSERT INTO `messages` (`message_sync_id`,`from_user_sync_id`, `to_user_sync_id`, sent_dt, `message_text`,`shared_campaign_id`,`shared_campaign_location_id`) 
                                      SELECT 
                                      CASE 
                                      WHEN max(message_sync_id) is null
                                      THEN 0
                                      ELSE max(message_sync_id)
                                      END + 1
                                      , ?
                                      , ?
                                      , NOW()
                                      , ?
                                      , ?
                                      , ?
                                      from messages;"))
             {
                $stmt1->bind_param("sssss", $userId,$touserid,$message,$campaign,$location);
                $stmt1->execute();
                $stmt1->close();

                $sqlMaxSyncId = "select max(message_sync_id) as message_sync_id from messages;";

                if($resultMaxID = $db->query($sqlMaxSyncId))
                {

                    $row = $resultMaxID->fetch_assoc();

                    $out = json_encode(array('message_sync_id' => $row['message_sync_id']));

                    error_log("sendMessage", 3 , $row['message_sync_id']);

                }
                else
                {
                    $out = FAILED;
                }           

             }
             else
             {
                $out = FAILED;
             }

        }
        else
        {
            $out = FAILED;
        }   
    }
    else
    {
        $out = FAILED;
    }   
break;

Given chat applications require very low latency, what production problems may this formulation run into that I may not see during very low usage testing?

How can these be alleviated?

Would Heroku be an appropriate environment to deploy such a server?

Answer 1

The save to database (which I guess is "syncWithSErver") could become a bottleneck if many users are chatting at the same time as I guess, that case statement executes a database query, a database that would be overloaded, and thus, slowing down the message dispatch.

Instead of saving them directly, use any memory cache, such as memcache or redis to store them temporary, and use a background job (a php script attached to cron, for example) that will keep saving them.