Design and implementation of Game server monitoring (I.)

Monitoring is intended to facilitate maintenance of the game server, when the server is in a timely manner to alert, and can monitor the server related conditions. However, demand is constantly changing. This is all right, especially when writing this tool.

Demand iterations, many of the situations that have arisen have not been considered clearly. For the requirements to do the following list.

Iteration One:

At first, operation and maintenance said is relatively simple, as long as the server on the machine to write a monitoring, can monitor the server hangs, can be hung when the mailbox sent an e-mail. is not a monitoring tool, simple! Just write a tool on the same physical host with the server to start one on the line, once the server hangs up can be alerted to send messages, and should be shipped-dimensional requirements also made a small interface for viewing. Monitoring, game server process, number of online, number of clientsession, database connectivity and so on.

Iteration Two:

Operational requirements have changed because the game server will be deployed on multiple physical hosts, need to be a proxy to monitor server performance on all physical hosts, and be able to test the connection between the physical host on which the game server resides and the database (another server). However, the game server is running on, windows, the system where the agent is located is a Linus system. I have simply implemented the following:

Iteration Three:

The general framework is there, but the operation is still not very satisfactory. Their needs have changed again. This is mainly functional.

1, the timeliness of e-mail warning and the cause of the error

2, the game server can be a hang off implementation tracking, from the sending of email Alerts game Server exception, to the game server to solve the ability to distinguish the prompt reminder. And OPS is a multi-person shift, if you can track the solution better, and send an email to remind others.

3, multi-game server is located in the physical host, monitoring terminal program configuration is simple

4, Windows do not have a good UI interface, and Linus platform operation has problems need to be adjusted

5, the Game Server connection database way is no longer ODBC, adjust the verification means!

6, can achieve a simple game logic operation, check whether the game server is running properly, for the game server into the dead loop dead no good monitoring to, if you can test some of the game features better.

At this time, this is not a monitoring tool, and gradually to a monitoring system forward. But this is not the scariest, the scariest thing is there is an iteration four!

Iteration Four:

When I finally went to the distribution model of his physical mainframe, I thought it was too easy to start, or it was amazing what OPS was thinking! As the game is on-line in Korea, so the game server is on the physical machine in Korea, the agent is on a domestic machine, but the agent side of the view is on another machine. And because of the instability of international communications, sometimes South Korea's physical host may and domestic machine connection failure, but in fact, its own game server operating network is not a problem. In addition to ensuring that iteration three, the mentioned function, there are some of the following:

1, for the agent-side and foreign physical host network quality warning and can be e-mail reminders.

2, to ensure that the game server to do the shutdown service maintenance, can no longer remind the abnormal

3, add the deletion of the game server, can be very convenient configuration

4. Ability to detect local network quality

5, to provide the requirements of distributed operation (not critical), you can achieve a Linux-based operation of the operation on the

6, can be separate to a single physical host game server monitoring wake-up settings and operation

7, but also to ensure the independence of the game server, can not occupy a large amount of bandwidth

8, for the detection cycle, etc. can be set, and set the number of re-check rotation times set. Different detection data need to make a distinction, some immediate alarm, and some will need to detect a few times still failed to send an alert message.

I feel that if the monitoring system estimates are not appropriate, a typical distributed monitoring system is true.

Design and implementation of Game server monitoring (I.)