Telnetcommand to connect CLB listening ports
The difference between layer-4 and layer-7 instances is whether layer-4 or layer-7 information is used as the basis for determining how to forward traffic for load balancing on real servers.
For example, a layer-4 CLB instance determines which traffic needs load balancing based on the layer-3 IP address (VIP) and layer-4 port number. It performs Network Address Translation (NAT) on the traffic to be processed, and then forwards it to the real server. It also records which server has processed the TCP or UDP traffic, and forwards all subsequent traffic of this connection to the same server for processing.
A layer-7 CLB is layer-4 CLB instance combined with application layer features.
For example, CLB instances of the same web server can identify traffic that needs to be processed based on layer-7 URL, browser type, and language in addition to the VIP and port 80.
Layer-7 load balancing is also known as "content exchange", in which an internal server is selected based on meaningful application layer content in the message and the server selection method configured on the CLB instance.
To select the server based on the real application layer content, a layer-7 CLB instance must establish a connection (three-way handshake) with the client as a proxy of the final server to receive the message containing real application layer content from the client. It then selects the internal server according to the specific fields in the message and the server selection method configured on the CLB instance. In this case, the CLB instance is more like a proxy server, establishing a TCP connection with the frontend client and the real server respectively.
TCP is a connection-oriented protocol. Before receiving and sending data, TCP requires a reliable connection to be established with the other side. UDP is a message-oriented (connection-less) protocol. It directly sends data packets without performing a three-way handshake with the other side. UDP is suitable for scenarios that focus more on real-timeliness than reliability, such as video chat, real-time push of financial market information, DNS, and IoT.
In cookie insertion mode, the CLB instance inserts cookies and the real server does not need to make any modifications. When you make the first HTTP request, it (without a cookie) enters the CLB instance, which then selects a real server based on the load balancing algorithm policy and sends the request to the server. The real server then sends an HTTP response (without a cookie) that is sent back to the CLB instance, which then inserts the cookie and returns the HTTP response (with a cookie) to the client.
When you make the second HTTP request (with the cookie inserted by the CLB instance last time), it enters the CLB instance, which then reads the session persistence values in the cookie and sends the request (with the same cookie as above) to the specified real server. The server gives an HTTP response. Because the server does not write the cookie, the response does not contain the cookie. When the response traffic re-enters the CLB instance, CLB will write the updated session persistence cookie into the response.
You can specify the forwarding weight for each CVM instance in the real server pool, and the instance with a higher weight will be assigned with more access requests. You can configure the weights of the backend CVM instances based on their service capabilities and statuses.
If you have also enabled session persistence, the same access request may be forwarded to different real servers. We recommend temporarily disabling session persistence and then checking whether the problem persists.
Please troubleshoot by following the steps below:
iptables) on the CVM instance.
Suppose that you configured to send a health check packet every 5 seconds in the console. However, the real server receives one or even more health check requests in 1 second.
Highly frequent health check relates to the implementation mechanism of CLB health check. Suppose that 1 million requests from clients are distributed to 4 CLB instances before being sent to real servers, and each CLB real server conducts health checks separately. If the CLB instances are configured to send a health check request every 5 seconds, each CLB instance will send a health check request every 5 seconds. This is why the real server receives multiple health check requests. For example, if there are 8 CLB instances in a cluster and each sends a request every 5 seconds, then the real server may receive 8 health check requests in 5 seconds.
The advantages of this implementation scheme are high efficiency, accurate check, and avoidance of mistaken removal. For example, if one of the CLB servers in the cluster fails, the other 7 servers can still forward traffic normally.
Therefore, if your real server is checked too frequently, you can configure a longer check interval (for example, 15 seconds).
When you visit the website
http://example.com via a browser, a redirection to the root directory is required for the server. When you visit the website
http://example.com/ via a browser, the server will directly return the default page of the website's root directory. Similarly, if
http://cloud.tencent.com/movie is redirected to
http://cloud.tencent.com/movie/ through URL rewriting, entering
http://cloud.tencent.com/movie will result in an additional URL rewriting process, leading to slight performance degradation and time consumption. However, if
http://cloud.tencent.com/product is redirected to a page other than
http://cloud.tencent.com/product/ through URL rewriting, you need to consider whether to add "/" after the secondary URL.
In Tencent Cloud CLB, if the frontend and backend port numbers are different, "/" needs to be added after the secondary URL upon the visit to the secondary page, so as to avoid port number changes after HTTP redirection and ensure normal page access.
Assume that in layer-7 forwarding, port 80 on the CLB instance and port 8081 on the real server are listened to. If the client accesses
http://www.example.com/movie, the access request is forwarded to the real server via the CLB instance, and the server redirects the request to
http://www.example.com:8081/movie/(listening port is 8081). In this case, the client access fails (port error).
Therefore, we recommend rewriting the access request to a secondary URL with "/", such as
http://www.example.com/movie/. This can avoid HTTP redirection, eliminate unnecessary judgment, and reduce unwanted load. If HTTP redirection is required, make sure that the CLB listening port is the same as that of the real server.
You can perform load balancing for the following TCP ports: 21 (FTP), 25 (SMTP), 80 (HTTP), 443 (HTTPS), 1024–65535, etc.
When the CLB instance receives the policy request from port 843, it will return the general cross-domain policy configuration file. If no policy file is returned and the connection is directly closed, the flash server request may be incorrect.
Make sure the correct flash server request is sent:
It must end with
\0and contain 23 bytes.
\0indicates a character whose ASCII code is 0 and only occupies 1 byte.
The normal result returned by port 843 is as shown below:
X-Forwarded-Formethod to get real client IPs. Acquisition of client IPs is enabled on CLB instances by default but needs to be configured on real servers. For more information, please see Getting Real Client IPs.
No. To access port a on server A (10.66.*.101), the request can be forwarded to port b on server B (10.66.*.102) via a private network CLB instance. But it cannot be forwarded to port b on the same server A (10.66.*.101).
HTTP/2 is only supported in HTTPS but not in HTTP, and backward compatibility is allowed on both the client and server.
HTTP/2 is only supported in HTTPS, but Gzip can be used in any HTTP version supported by Tencent Cloud.
If security group rules have been configured for a real server, the CLB instance may not be able to communicate with the server. Therefore, in layer-4 and layer-7 forwarding, we recommend configuring the security group of a real server as allowing all access requests. If the security group is enabled and accesses from any protocols or IP segments are denied by default, IPs of all clients need to be configured to the security group rules of the server IP.
For some malicious IPs, you can add them to the top rules of the security group to prevent them from accessing the real server. You can then allow access requests from all IPs (0.0.0.0) to the local service port, so normal clients can access the server. Security group rules are arranged in priority order and matched from top to bottom.
If health check has been configured for layer-7 CLB forwarding in VPC, in the real server security rules, the CLB VIP must be allowed. Otherwise, the health check may fail.
If you need to configure a blocklist for some client IPs to deny their access requests, you can configure the security group associated with the cloud services. The security group rules need to be configured as follows:
Follow the steps below strictly in the given order, otherwise the blocklist configuration may fail.
clientA ip+port drop clientB ip+port drop 0.0.0.0/0+port accept
For more information on the security group, please see Security Group Configuration of the Real Server.
A CLB instance always communicates with real servers over the private network, even when the bound CVM instances have public IPs.
The requests of pinging the CLB VIP are responded to by the CLB cluster and will not be forwarded to real servers.
Telnetto test private network CLB instances.
Telnetcommand to connect CLB listening ports
Telnetto connect to their listening ports will fail. It will succeed if they are bound with real servers.
Telnetto connect to listening ports will succeed for layer-7 listeners (HTTP and HTTPS) not bound with real servers, as CLB instances will respond instead.
For private network CLB instances, a CVM cannot be both the client and server. When the CLB instances read the same client and server IPs, access will fail.
If your client needs to be used as a server, please bind at least 2 real servers. CLB has related policies to prevent automatic loopback. When client A accesses the CLB instance, the CLB instance will automatically schedule the request to a real server other than client A.
When a client accesses the same port of a real server via different intermediate nodes at the same time, these nodes cannot be specifically determined. The scenarios are detailed below:
The CLB instances passe the client IP to the real server, and
client_ip:client_port -> vip:vport -> rs_ip:rs_port will change to
client_ip:client_port --> rs_ip:rs_port.
To allow traffic to access a backend CVM instance passing through a CLB instance, the traffic from the public network needs to be allowed both on the backend CVM and CLB security groups. We recommend only allowing the access traffic from a CLB VIP over the public network on the backend CVM security group first, and then allowing public network access IPs on the CLB security group as needed.
Only the traffic going through CLB will be monitored. You can resolve the domain name to the VIP of the CLB instance and access the domain name to view the CLB monitoring information.
The port 843 is opened by default for users to reset the port for Flash access. If you want to close it, you can just do so by not binding any real servers after configuring the