Known Issues

Network Disconnection Caused by a Windows Server DHCP Lease Longer Than 99 Days

Symptom:

If the DHCP lease is longer than 99 days, instance IP addresses cannot be automatically renewed. As a result, the instance network will be disconnected when the lease comes to an end.

Involved images:

Public and private images of Windows Server 2008, Windows Server 2012 R2, Windows Server 2016, and Windows Server 2019

Solutions:

1. Change the DHCP lease of the subnet where the instance is located to one day or unlimited.
2. Run the following command to make the change take effect:
   

   The following command will temporarily disconnect you from the network. Do it during off-peak hours.

   ipconfig /renew

Occasional System Errors Triggered By Adding or Deleting NICs

Symptom:

After an ECS is started, adding or deleting a NIC or other equivalent actions may:

• Trigger a kernel panic, and the OS automatically restarts.
• Trigger frequent software interrupts, and the network may fail to receive or send packets.

Patch link: https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux-stable.git/commit/?id=f00e35e259948b995aa1f3ee7fddb05f34a50157

Involved images:

CentOS 7 public and private images

Solutions:

Upgrade the kernel to a version matching the 3.10.0-1160.25.1.e17.x86_64 kernel of CentOS 7.

Kernels Are Occasionally Disconnected from the TCP Network

Symptom:

This issue is caused by the kernel vulnerability CVE-2019-11477 (TCP SACK). When the socket buffer is low, the network may be disconnected.

The involved images are listed in the following table.

Solutions:

Upgrade the kernel to the latest version. Run the following commands to upgrade the kernel of each image type:

• CentOS/Fedora: yum update kernel
• Ubuntu: apt-get update && apt-get install linux-image-generic
• openSUSE: zypper refresh && zypper install kernel-default
• Debian: apt-get update && apt search linux-image && apt-get install linux-image-xxx
  

  You can run the apt search linux-image command to query for the latest kernel version. The apt-get install linux-image-xxx command is used to upgrade a kernel to the latest version.

OS Parameter Settings Do Not Take Effect

Symptom:

After net.ipv4.tcp_max_tw_buckets is configured in the /etc/sysctl.conf file, the check result of sysctl -a indicates that the configuration does not take effect. The configurations in /etc/sysctl.d/huawei.conf and /etc/security/limits.d/huawei-nofile.conf have been built in public images and these configurations have higher priorities than those in /etc/sysctl.conf. As a result, the configurations in /etc/sysctl.conf do not take effect.

The involved parameters are listed in the following table.

Involved images:

• CentOS 7 public images from 2018-09-25 to 2019-09-29
• CentOS 6 public images from 2018-09-25 to 2019-10-10
• Ubuntu, openSUSE 15.0, Debian, and Fedora 29 public images from 2018-09-28 to 2019-10-15

Solutions:

1. Delete the built-in configuration files.

   rm -rf /etc/sysctl.d/huawei.conf

   rm -rf /etc/security/limits.d/huawei-nofile.conf

2. Modify the kernel parameter configuration files (limits.conf and sysctl.conf).

   cat >>/etc/security/limits.conf <<EOF

   root soft nofile 65535

   root hard nofile 65535

   * soft nofile 65535

   * hard nofile 65535

   EOF

   cat >>/etc/sysctl.conf <<EOF

   vm.swappiness=0

   net.core.somaxconn=1024

   net.ipv4.tcp_max_tw_buckets=5000

   net.ipv4.tcp_max_syn_backlog=1024

   EOF

1822 NIC-based Offloading Instance Is Incompatible with the Linux 3.16.x Kernel

Symptom:

ECSs that use the hardware offloading function provided by Huawei-developed 25GE intelligent high-speed NIC may be incompatible with Linux 3.16.47–3.16.x OSs, which may cause occasional network disconnections of ECSs. ECSs that have this issue include but are not limited to C3ne, M3ne, C6, M6, G5, P2v, G5r, P2vs, P2s, Pi2, FP1cn1, Ai1, e3.26xlarge.14, e3.52xlarge.14, e3.52xlarge.20, KC1, and KM1.

Involved images:

Debian 8.2.0 64bit and Debian 8.8.0 public images

Solutions:

Remove Debian 8 public images from flavors. Migrate services of ECSs using this offloading function to S3 and C3 ECSs as soon as possible.

Data Is Lost During Disk Reset Due to the Incompatibility Between Server Manager of Windows Server 2012 R2 and VMTools

Symptom:

A Windows Server 2012 R2 ECS is configured with two data disks. When Windows Server Manager resets the second data disk, the first data disk is reset. As a result, the data of the first data disk is lost.

Involved images:

Windows Server 2012 R2 public image before 2019-02-19

Solutions:

Upgrade VMTools of involved ECSs to 2.5.0.156 or later.

Service Interruptions Caused By Lingering CLOSE_WAIT Connections

Symptom:

Some services are interrupted because a socket in a TCP connection created by the one-click password reset plug-in process stays in the CLOSE_WAIT state.

Involved images:

• CentOS and EulerOS public images issued before June 5, 2019
• Ubuntu and Debian public images issued before June 3, 2019

Solutions:

Update the one-click password reset plug-ins for the ECSs.