Operating system maintenance FAQ

General issues

• Enabling the CONFIG_PARAVIRT_SPINLOCK kernel option may cause performance issues

• A printk deadlock in Alibaba Cloud Linux causes a system crash

• How to disable CPU vulnerability fixes in Alibaba Cloud Linux

• Known issues in NFS v4.x

• Resolving a segmentation fault caused by SysAK 2.2.0 when running the DNF command

• Preventing application performance jitter caused by cgroups

• Resolving Overlayfs operation failures caused by insufficient read permissions in Alibaba Cloud Linux

• OOM Killer: Causes and solutions

• Troubleshooting high slab_unreclaimable memory usage

• Mitigating Linux memory fragmentation

• Customizing DNS configuration with the /etc/resolv.conf file in an Alibaba Cloud Linux instance

• Resolving packet loss on ECS instances caused by the "kernel: nf_conntrack: table full, dropping packet" kernel log message

• Is it free to run Alibaba Cloud Linux on ECS instances?

• Which Alibaba Cloud ECS instance types does Alibaba Cloud Linux support?

• Does Alibaba Cloud Linux support 32-bit applications and libraries?

• Does Alibaba Cloud Linux support a graphical user interface (GUI)?

Alibaba Cloud Linux 3

• The sch_netem module is missing in Alibaba Cloud Linux 3

• Configuring XPS in Alibaba Cloud Linux 3: Methods and impacts

• Resolving kernel upgrade errors in the Alibaba Cloud Linux 3.8 image

• How to disable CPU vulnerability fixes in Alibaba Cloud Linux 3

• Resolving high system load with no workload in Alibaba Cloud Linux 3

• Resolving poor read performance of the NFS file system in Alibaba Cloud Linux 3

Alibaba Cloud Linux 2

• Tuning performance for transparent huge pages (THP) in Alibaba Cloud Linux 2

• Resolving the polkit memory leak in Alibaba Cloud Linux 2

• Troubleshooting systemd service issues in Alibaba Cloud Linux 2

• Enable or disable IPv6 in Alibaba Cloud Linux 2

• Installing and enabling a later version of curl in Alibaba Cloud Linux 2

• Resolving issues with querying and configuring routing information on an ECS instance that runs Alibaba Cloud Linux 2

• Resolving failures when creating a large number of processes on an ECS instance that runs Alibaba Cloud Linux 2

• Mitigating the network performance impact of the TCP BBR congestion control algorithm in Alibaba Cloud Linux 2

• Resolving poor Buffer I/O write performance of the ext4 file system on an ECS instance that runs Alibaba Cloud Linux 2

• Troubleshooting a Send-Q value of 0 in ss command output on an ECS instance that runs Alibaba Cloud Linux 2

• Tuning performance for transparent huge pages (THP) in Alibaba Cloud Linux 2

• Resolving Overlayfs mount failures caused by a shared mount directory on an ECS instance that runs Alibaba Cloud Linux 2

• Resolving the "integrity: Unable to open file" message in the kernel log of an ECS instance that runs Alibaba Cloud Linux 2

• Can I view the source code of Alibaba Cloud Linux 2 components?

• Is Alibaba Cloud Linux 2 backward-compatible with earlier versions of Aliyun Linux?

• Which third-party applications are supported on Alibaba Cloud Linux 2?

• How does Alibaba Cloud Linux 2 ensure data security?

• Does Alibaba Cloud Linux 2 support data encryption?

• How to configure permissions in Alibaba Cloud Linux 2

Linux guest OS: issues and solutions

Boot failures

• Check for block device information in the fstab file

• Verify that block devices are mounted correctly in the fstab file

• Verify the format of the fstab file

• Use the fsck command to check system files

Logon failures

• Check that the limits settings are correct

• Check if a password is set for the root account

• Verify the format of critical system files

• Verify that SSH access permissions are correctly configured

• Verify that required SSH files and directories exist

• Check if the huge pages setting is too large

• Check if the operating system is out of memory (OOM)

• Check if the system firewall is enabled

• Check if TCP SACK is enabled

• Check if the UDP cache has overflowed

• Check if SELinux is enabled

• Unable to log on to an instance with SSH or VNC

• Error when connecting to an instance

Instance access issues

• Verify that the kernel parameters for the NAT environment are correct

• Verify that processes have started and that common service ports are in a LISTENING state

Network connectivity issues

• Verify that the DHCP configuration is correct

• Check if network-related processes exist

• Check if multi-queue for the network interface controller is enabled

Performance issues

• Check if the TCP backlog has overflowed

• High CPU utilization

• Unable to write files to a disk

Windows guest OS: issues and solutions

Logon failures

• Verify that the Windows network interface controller is enabled

  You cannot log on to a Windows instance if its system network interface controller is unavailable. For more information, see The system network interface controller in a Windows instance is unavailable.

• Check if port 3389 is open

  The Remote Desktop Protocol (RDP) service lets you manage your Windows instance. If the Remote Desktop Protocol (RDP) service is disabled, you cannot establish a remote desktop connection. For more information, see Start the Remote Desktop Protocol (RDP) service for a Windows instance.

• Check if the virtio driver version is outdated

  An outdated virtio driver can prevent you from logging on to the instance. For more information, see Update the virtio driver of a Windows instance.

• Check that the firewall is correctly configured

  Incorrect firewall settings can prevent logon. For more information, see Windows system firewall policy configuration guide.

Performance issues

• High CPU utilization

  Sustained high CPU utilization can affect system stability and service performance. For more information, see Troubleshoot and resolve high CPU utilization on a Windows instance.

• Check the Windows operating system version

  Microsoft ended support for Windows Server 2008 and Windows Server 2008 R2 on January 14, 2020. Therefore, Alibaba Cloud no longer provides technical support for ECS instances that run these operating systems. If you have ECS instances that run these operating systems, update them to Windows Server 2012 or a later version as soon as possible. For information about the supported images, see public images or refer to the buy page.

• Check the disk capacity

  Insufficient space on the C drive can prevent the system from operating correctly. For more information, see How to troubleshoot low space on the C drive of a Windows instance.

Change the operating system (system disk)

You can change the operating system (replace the system disk) by changing the image of an ECS instance.

Change the OS with a shared custom image

Yes. Account A must first share the custom image with Account B. Then, Account B can use the image to change the operating system (replace the system disk).

Use an image with data disk snapshots

You can use a custom image that contains data disk snapshots to change the operating system. This operation replaces only the system disk of the instance and does not affect its data disks.

Change OS vs. re-initialize system disk

System disk resize failure during OS change

Resizing a system disk partition during an OS change may fail due to a timeout. If this occurs, you must manually extend the partition. For more information, see Extend partitions and file systems of a Linux disk. This method extends only the system disk partition and does not affect the operating system version.

Troubleshoot the non-I/O optimized instance error

Reset the system time zone

For a single instance, run the timedatectl set-timezone command to change the time zone.

You can use Cloud Assistant to change the time zones of multiple instances at once.

Container runtime in Windows Server with Container images

Due to changes in Microsoft's support policy for container runtimes (for more information, see Supported Container Runtime on Windows Server), Windows Server with Container images updated by Alibaba Cloud ECS since early 2024 no longer include Mirantis Container Runtime (MCR). Instead, they include the open-source containerd runtime. If you require MCR, you must purchase and install Mirantis Container Runtime from Mirantis.

Starting March 1, 2024, Windows Server with Container images provided by Alibaba Cloud ECS include the following container-related components:

• The Windows Server container feature. Hyper-V isolation is not supported. For more information, see Windows and containers.

• The containerd runtime, version 1.7.13. For more information, see containerd.

• nerdctl.exe, a command-line interface (CLI) for managing containers, version 1.7.13. For more information, see nerdctl.

• nat.exe, a Container Network Interface (CNI) plugin for Windows container networking, version 1.0.0. For more information, see windows-container-networking.

User data script write failure on Windows

Issue

When you use a user data script to write data to the C:\Users\Administrator\Desktop\userData_test.txt path, the operation fails with a "path not found" error.

Cause

In Windows, the C:\Users directory and its subdirectories are the default storage locations for user profiles and data. You can access these locations only after logging on to the system. The user data script runs during instance initialization, before a user logs on. Therefore, attempts to write data to the C:\Users directory fail.

Solution

To resolve this issue, modify the write path in your user data script to a directory that is not user-specific, such as the root of the C: drive. For example:

[bat]
echo "userData" > C:\userData_test.txt

For more information, see Customize instance initialization configurations.

Limitations of the Windows Server 2025 image

• vCPU: 1 to 640

• Memory: 2 GiB to 48 TiB

• ECS instance types

  Due to compatibility issues, the following ECS instance types do not support this image:

  • All ECS Bare Metal instance families

  • 6th generation AMD instance types (general-purpose instance family g6a, compute-optimized instance family c6a, and memory-optimized instance family r6a)

On an ECS instance running Windows Server 2008, frequent calls to the timeBeginPeriod Windows system API can make the system time inaccurate. To resolve this issue, perform the following steps:

1. Connect to the ECS instance.

   For instructions, see Log on to a Windows instance by using Workbench.

2. Download the tool.

3. Unzip CheckTimeBeginPeriod.zip.

4. Unzip bin.zip, go to the bin directory, and double-click the appropriate .exe file.

   • For a 64-bit operating system, double-click InjectDllx64.exe.

   • For a 32-bit operating system, double-click InjectDllx86.exe.

   The tool displays the process that is calling the timeBeginPeriod API.

5. Based on your business requirements, stop or update the application that is calling the timeBeginPeriod API.

If the issue persists, submit a ticket.

Getting technical support for Red Hat Enterprise Linux

Instead of contacting Red Hat directly, you can submit a ticket to Alibaba Cloud for technical support. An Alibaba Cloud support engineer will assist you with your issue. If Alibaba Cloud cannot resolve an issue with the Red Hat Enterprise Linux operating system, we will escalate the ticket to Red Hat.

Included Red Hat subscriptions

Red Hat images from Alibaba Cloud include a Red Hat Enterprise Linux (RHEL) subscription. The following software repositories are available:

• RHEL 7

  • Red Hat Enterprise Linux 7 Server - Extras from RHUI (RPMs)

  • Red Hat Enterprise Linux 7 Server - Optional from RHUI (RPMs)

  • Red Hat Enterprise Linux 7 Server from RHUI (RPMs)

• RHEL 8 & RHEL 9

  • BaseOS

  • AppStream

  The latest RHEL 8 and RHEL 9 images also include the CodeReady Linux Builder and Supplementary repositories by default. To use these software repositories on your RHEL 8 or RHEL 9 instance, contact Alibaba Cloud support.

  For details about the software repositories and package lists for RHEL 8 and RHEL 9, see the RHEL 8 Package Manifest and RHEL 9 Package Manifest.

Red Hat images from Alibaba Cloud provide packages for RHEL only. To install packages for other Red Hat products, such as Red Hat Satellite or Red Hat Ceph Storage, you must purchase a separate Red Hat subscription, register your host, and subscribe to the required products.

"Unknown" subscription status

This is expected behavior. When you use a Red Hat Enterprise Linux image from Alibaba Cloud, your instance receives updates from Alibaba Cloud's update sources. This differs from the traditional model, where you would use a Red Hat account to get updates directly from Red Hat. Therefore, running the subscription-manager command correctly reports the status as "Unknown":

+-------------------------------------------+
 System Status Details
+-------------------------------------------+
Overall Status: Unknown

System Purpose Status: Unknown

Support for SUSE Linux Enterprise Server (SLES)

Alibaba Cloud provides SUSE Linux Enterprise Server (SLES) public images that are regularly synchronized with SUSE update repositories. Alibaba Cloud Enterprise Support includes operating system support for instances created from SLES public images. If you have purchased Alibaba Cloud Enterprise Support, you can submit a ticket for technical support. Alibaba Cloud engineers will help you resolve issues with the SLES operating system.

"Command not found" error

"Permission denied" error

wget bash: /usr/bin/wget: Permission denied

Symptom

Installing an AD domain controller on a Windows ECS instance fails with the error message: "Installation of Active Directory Domain Services binaries failed".

Cause

The Event Viewer indicates that the Remote Registry service is disabled and cannot start.

Solution

Follow these steps to start the Remote Registry service.

1. Connect to the Windows instance.

   See Connect to a Windows instance by using a password or key.

2. Go to Start > Run, enter services.msc, and then click OK.

3. In the Services window, double-click the Remote Registry service to open the Remote Registry Properties window. Configure the following settings:

   • Set Startup Type to Automatic.

   • Under Service status, click Start. Ensure the Remote Registry service is running.

4. Click OK to save the settings.

Symptoms

When you install an AD domain controller on a Windows ECS instance, the message "This computer has dynamically assigned IP addresses" appears.

Cause

At least one physical network adapter on the Windows ECS instance does not have a static IP address.

Solution

1. Connect to the Windows instance.

   For instructions, see Log on to a Windows instance by using a password or key.

2. Install the AD domain controller. For instructions, see Build a Windows AD domain on an ECS instance.

3. When the Static IP Assignment dialog box appears, click Yes.

Symptoms

When you install an AD domain controller on a Windows ECS instance, the installation fails with the error code "0x0000232B RCODE_NAME_ERROR".

Cause

An incorrect IP address configuration in the DNS server may cause this issue.

Solution

Follow these steps to change the DNS server address on both the internal and external network adapters of the secondary instance to the private IP address of the primary instance.

1. Connect to the Windows instance.

   For more information, see Connect to a Windows instance by using a password or key.

2. Go to the Internet Protocol Version 4 (TCP/IPv4) Properties window, change the DNS server address, and then click OK.

   Note

   Set the DNS server address to the private IP address of your primary instance.

   

3. Verify that you can ping the DNS server's IP address.

Symptoms

When installing an AD domain controller on a Windows ECS instance, you receive the error message "The network path was not found".

Causes

Possible causes include the following:

• The TCP/IP NetBIOS Helper and Remote Registry services on the AD domain controller and the client are not running.

• The DNS configuration on the client or the AD domain controller is incorrect.

• A SID conflict exists between the client and the AD domain controller.

• A firewall or security software is blocking the connection.

Solutions

Follow these steps to troubleshoot the issue.

• Change the client SID

  To change the client SID, follow these steps:

  1. Connect to the Windows instance.

     For more information, see Connect to a Windows instance by using a password or key.

  2. Download the PowerShell script to change the client SID.

     • Download link: AutoSysprep.ps1

     • Source: Alibaba Cloud

  3. Open Command Prompt, enter PowerShell, and press Enter to open Windows PowerShell.

     Note

     If your instance runs a 64-bit operating system, do not use 32-bit PowerShell (Windows PowerShell (x86)). Otherwise, an error occurs.

  4. Navigate to the script's directory and run the following command to view the help information:

     .\AutoSysprep.ps1 -help

  5. Run the following command to re-initialize the server's SID:

     .\AutoSysprep.ps1 -ReserveHostname -ReserveNetwork -SkipRearm -PostAction "reboot"

     After the initialization is complete, the instance restarts. Note the following:

     • The IP address assignment method changes from DHCP to a static IP address. Ensure this static IP address matches the original IP address of the ECS instance. Alternatively, switch back to DHCP to automatically use the primary private IP address assigned on the console.

       Note

       Do not change the primary private IP address of the ECS instance on the console. Otherwise, access issues may occur.

     • After the SID is re-initialized, the firewall configuration reverts to the Microsoft default settings, which may block ping requests. To resolve this, you can turn off the firewall for the Guest or Public Network profile or add a rule to allow the required ports.

  6. Open Control Panel and turn off the firewall for the Guest or Public Network profile.

     You can then ping the server.

• Allow client connections through the firewall or security software

  For instructions, see Windows system firewall policy configuration guide.

Resolve CentOS DNS resolution timeouts

Cause

Due to a change in the DNS resolution mechanism of CentOS 6 and CentOS 7, DNS resolution timeouts may occur on ECS instances that run CentOS 6 or CentOS 7 and were created before February 22, 2017. The issue may also occur on instances created from custom images that were created before that date.

Solution

The following sections explain the script's function and logic.

• Script function

  The script checks if the options line in the /etc/resolv.conf file contains the single-request-reopen configuration. For more information, see the resolv.conf file description.

  CentOS 6 and CentOS 7 use the same network 5-tuple to send both IPv4 and IPv6 DNS requests. In this case, the single-request-reopen option should be enabled. This option forces the resolver to close the socket after the first request and open a new one for the second request. The change takes effect immediately without requiring an instance restart.

• Script logic

  1. The script checks whether the instance is running CentOS.

     • If the instance is not running CentOS, such as Ubuntu or Debian, the script exits.

     • If the instance is running CentOS, the script continues.

  2. The script checks the options configuration in the /etc/resolv.conf file.

     • If no options line exists:

       The script adds the default Alibaba Cloud options configuration: options timeout:2 attempts:3 rotate single-request-reopen.

     • If an options line exists:

       • If the single-request-reopen option is missing, the script appends it to the existing options line.

       • If the single-request-reopen option is already present, the script exits without changing the DNS configuration.

Check and fix missing IP addresses

For the cause and solution, see Fix missing IP addresses on CentOS 7 and Windows instances.

Resolve dump file generation failures

Symptoms

After a CentOS 7.9 ARM system crashes, you run the ls /var/crash command, but no vmcore file is generated.

Cause

The pre-installed version of the makedumpfile software is incompatible with this kernel, which prevents the system from generating a dump file. The kernel of the CentOS 7.9 ARM system includes the CONFIG_ARM64_USER_VA_BITS_52=y feature.

Solution

1. Run the following command to download the kexec-tools RPM package.

   wget http://mirrors.aliyun.com/centos-vault/7.9.2009/os/Source/SPackages/kexec-tools-2.0.15-51.el7.src.rpm

2. Run the following command to install the RPM package.

   rpm -ivh kexec-tools-2.0.15-51.el7.src.rpm

3. Run the following commands to download the patch file.

   cd /root/rpmbuild/SOURCES
   wget https://ecs-image-tools.oss-cn-hangzhou.aliyuncs.com/patch/rhelonly-kexec-tools-2.0.20-makedumpfile-arm64-Add-support-for-ARMv8.2-LVA-52-bi.patch

4. Modify the kexec-tools.spec file.

   1. Open the kexec-tools.spec file.

      cd /root/rpmbuild/SPECS/
      vi kexec-tools.spec

   2. Press i to enter insert mode, and then add the following two lines in the appropriate locations.

      Patch999: rhelonly-kexec-tools-2.0.20-makedumpfile-arm64-Add-support-for-ARMv8.2-LVA-52-bi.patch
      %patch999 -p1

      Insert the lines in the locations shown in the following figures.

      

      

   3. Press Esc to exit insert mode, and then enter :wq to save and exit.

5. Run the following command to check and install dependencies.

   yum-builddep kexec-tools.spec 

6. Run the following commands to build the RPM package.

   yum -y install rpm-build
   rpmbuild -ba kexec-tools.spec

7. Run the following commands to install the modified RPM package.

   cd /root/rpmbuild/RPMS/aarch64
   rpm -ivh kexec-tools-2.0.15-51.el7.aarch64.rpm

After another crash, run the ls -lh /var/crash command. If a vmcore file is now generated, the issue is resolved.

• Recover from rescue mode

  Upgrading systemd to version systemd-219-71.el7 on a CentOS 7 or Red Hat 7 series system can cause the instance to enter rescue mode after a restart. This disrupts network services and general application services. For the solution, see A CentOS 7.X or Red Hat 7.X system enters rescue mode after a systemd upgrade and restart.

• Resolve slow startup of Red Hat images

  On an ECS Bare Metal Instance, an instance running Red Hat 8.1 or 8.2 starts 1 to 2 minutes slower than an instance running Red Hat 7. To resolve this issue, open the /boot/grub2/grubenv file on your Red Hat 8.1 or 8.2 system, change the kernel boot parameter from console=ttyS0 console=ttyS0,115200n8 to console=tty0 console=ttyS0,115200n8, and then restart the server to apply the change.

High system load from Server Guard on Ubuntu

The system load on an ECS instance running certain versions of Ubuntu, such as Ubuntu 18.04, increases after the Server Guard process starts.

For the cause and solution, see High system load after the Server Guard process starts on an ECS instance that runs Ubuntu 18.04.

Why do Ubuntu system services restart automatically?

By default, Ubuntu uses the unattended-upgrade service to automatically update packages. This service periodically checks for and installs updates, which can cause related services to restart. If you do not need automatic updates, you can disable this service.

systemctl stop unattended-upgrade
systemctl disable unattended-upgrade

Patch and compile the FreeBSD kernel

Alibaba Cloud's public FreeBSD images use patched kernels to support booting on generation V or later instance families. You can query the instance families by using the Generation parameter of the DescribeInstanceTypeFamilies operation.

Patching is not required for FreeBSD 13 or later. This topic demonstrates how to patch and compile the FreeBSD kernel, using FreeBSD 12.3 as an example.

1. Download and decompress the FreeBSD kernel source code.

   wget http://ftp-archive.freebsd.org/pub/FreeBSD-Archive/old-releases/amd64/12.3-RELEASE/src.txz -O /src.txz
   cd /
   tar -zxvf /src.txz

2. Download the patch package.

   In this example, the 0001-virtio.patch is applied to the virtio driver.

   cd /usr/src/sys/dev/virtio/
   wget https://ecs-image-tools.oss-cn-hangzhou.aliyuncs.com/0001-virtio.patch
   patch -p4 < 0001-virtio.patch

3. Copy the kernel file, and then build and install the kernel.

   make -j<N> specifies the number of parallel jobs for compilation. You should determine this value based on the configuration of your build environment. For example, in a 1 vCPU environment, we recommend setting -j2. This means the ratio of vCPU cores to the variable N is 1:2.

   cd /usr/src/
   cp ./sys/amd64/conf/GENERIC .
   make -j2 buildworld KERNCONF=GENERIC
   make -j2 buildkernel KERNCONF=GENERIC
   make -j2 installkernel KERNCONF=GENERIC

4. After the build is complete, delete the source code.

   rm -rf /usr/src/*
   rm -rf /usr/src/.*

FreeBSD system disk not found in KVM

Problem

• Why can't I connect to a Fedora 33 instance with an ssh-rsa key pair?

  You may be unable to connect over SSH to an ECS instance running the 64-bit Fedora 33 operating system if you use an SSH key pair with the ssh-rsa signature algorithm. To resolve this issue, use one of the following methods:

  • Replace the ssh-rsa SSH key pair with one that uses a different signature algorithm, such as ECDSA.

  • Run the update-crypto-policies --set LEGACY command on the instance to switch the encryption policy from POLICY to LEGACY. This allows you to use an SSH key pair that uses the ssh-rsa signature algorithm.

• Why do Fedora CoreOS instances show only half the specified CPUs?

  On some instances created from a Fedora CoreOS image, such as those from the g5 general-purpose instance family, the On-line CPU(s) list in the lscpu command output shows only half the number of CPUs for the selected instance type. For example, if you select an instance type with 2 CPU cores, the On-line CPU(s) list shows only one CPU. The following figure shows an example.

  Note

  The On-line CPU(s) list field shows the IDs of the online CPUs.

  This behavior occurs because the kernel of the Fedora CoreOS image includes the mitigations=auto,nosmt boot parameter by default. This parameter automatically disables simultaneous multi-threading (SMT) on systems with known vulnerabilities, which reduces the number of available CPUs by half. You can view the mitigations=auto,nosmt parameter by running the cat /proc/cmdline command.

  For more information about SMT, see Automatically disable SMT when needed to address vulnerabilities and Policy for disabling SMT.

• Crash dump failure on ECS bare metal instances

  See What to do if some ECS instances fail to generate a crash dump file?.

• Softlockup exception during Linux kernel writeback

  A softlockup exception can occur during file cache writeback in some earlier versions of the Linux kernel. For a solution, see Solution for a softlockup exception that occurs during kernel writeback on a Linux operating system.

• Softlockup exception when deleting a cgroup

  See Solution for a softlockup exception that occurs when you delete a cgroup in an ECS instance.

• ECS instance downtime

  • What to do if an ECS instance has downtime and reports an "Out of memory and no killable processes" error?

  • What to do if an ECS instance has downtime and generates an "Objects remaining in kmalloc" alert log?

  • What to do if an ECS instance has downtime and generates a "RIP:get_target_pstate_use_performance" log?

  • What to do if an ECS instance has downtime and generates a "VFS: Unable to mount root fs on unknown-block" log?

  • Solution for a CentOS instance that hangs or unexpectedly restarts due to an excessive number of instances in a vSwitch

• Do public images include FTP services?

  No. You must install and configure FTP services yourself. For more information, see Build an FTP site (Windows) and Build an FTP site (Linux).

• Default virtual memory and swap configuration

  An operating system's memory manager uses a swap partition or a virtual memory file to temporarily move rarely accessed memory data to disk when physical memory is insufficient. This process increases the amount of available memory.

  However, this mechanism can degrade performance if memory usage is already high and I/O performance is poor. ECS cloud disks use a distributed file system as the storage backend and maintain multiple strongly consistent replicas for each data block. This approach ensures data durability but also increases I/O operations threefold, reducing storage and I/O performance.

  To avoid degrading cloud disk I/O performance under memory pressure, virtual memory is not enabled by default on Windows instances, and swap partitions are not configured by default on Linux instances.

• How do I enable Kdump in a public image?

  The Kdump service is disabled by default in public images. To generate a core file for downtime analysis, enable the Kdump service. This procedure uses a CentOS 7.2 public image as an example. The steps may vary based on your operating system.

  1. Set the directory where the core file is generated.

     1. Run vim /etc/kdump.conf to open the kdump configuration file. For more information about the vim command, see Vim editor.

     2. Set path to the directory where core files are generated. In this example, to generate core files in the /var/crash directory, the path is set as follows.

        path /var/crash

     3. Save and close the /etc/kdump.conf file.

  2. Enable the Kdump service.

     Use the method supported by your operating system.

     • Method 1: Run the following commands to enable the Kdump service.

       systemctl enable kdump.service

       systemctl start kdump.service

     • Method 2: Run the following commands to enable the Kdump service.

       chkconfig kdump on

       service kdump start

     • Method 3: If Cloud Assistant is installed on your server, see How do I resolve instance downtime issues after a migration? to enable the Kdump service.

• Set a static IP for a Linux instance

  You must establish a remote connection to the ECS instance to configure the settings. For more information, see Set a static IP address in a Linux instance.

• Customize DNS on a Linux instance

  For more information, see Customize the DNS configuration on a Linux instance.

• NTP sync failure after IPv6 configuration

  Symptom

  When you run the ntpq -p command on the server to synchronize the time, a timeout error occurs, as shown in the following figure.

  Solution

  Note

  This method applies to CentOS 7 and earlier, Ubuntu 20.04 and earlier, Anolis OS (ANCK and RHCK), Alibaba Cloud Linux, and Debian.

  1. Establish a remote connection to the Linux instance.

     For more information, see Log on to a Linux instance by using Workbench.

  2. Run the following command to modify the /etc/ntp.conf configuration file:

     vi /etc/ntp.conf

  3. Press the I key to enter the insert mode.

  4. Add the restrict -6 ::1 line to the file, as shown in the following figure.

  5. Press the Esc key, enter :wq, and press the Enter key to save the changes and exit.

  6. Run the following command to restart the NTP service:

     systemctl restart ntp 

• Hot-plug failure on custom image instances

  Symptom

  Hot-plugging a cloud disk refers to mounting or detaching a cloud disk while an instance is in the Running state. Hot-plugging a network interface refers to attaching or detaching an elastic network interface while an instance is in the Running state.

  Alibaba Cloud supports hot-plug for cloud disks and NICs, but this operation requires support from the operating system kernel. If the kernel does not support hot-plug, the following issues can occur:

  • After you attach a cloud disk or bind an ENI, the corresponding device is not visible in the operating system.

  • Detaching a cloud disk or unbinding an ENI fails.

  Solution

  Kernel support requirements for hot-plug differ between standard cloud instances and bare metal instances. Ensure your kernel supports both Peripheral Component Interconnect (PCI) and Advanced Configuration and Power Management Interface (ACPI) hot-plug features. These features are enabled by default on most systems, except for older versions such as CentOS 5. To verify that your kernel supports PCI and ACPI hot-plug, follow these steps:

  1. Establish a remote connection to the Linux instance.

     For more information, see Log on to a Linux instance by using Workbench.

  2. Run the following command to check the current kernel version of the instance:

     uname -r

     The following output indicates that the current kernel version is 3.10.0-1127.19.1.el7.x86_64.

     

  3. Run the following command to view the files in the /boot directory:

     ll /boot

     The following output indicates that config-3.10.0-1127.19.1.el7.x86_64 is the system's kernel configuration file.

     

  4. Run the following command to view the kernel configuration file of the system:

     cat /boot/config-3.10.0-1127.19.1.el7.x86_64

     • If the following configuration items are all set to y, the features are built into the kernel, and the operating system supports hot-plugging.

       CONFIG_HOTPLUG_PCI_PCIE=y
       CONFIG_HOTPLUG_PCI=y
       CONFIG_HOTPLUG_PCI_ACPI=y

     • If a configuration item is set to is not set, the feature was not compiled into the kernel. You must recompile the kernel to enable the feature.

     • If a configuration item is set to m, the feature is compiled as a module. For example, if CONFIG_HOTPLUG_PCI_ACPI is compiled as a module, you must load the corresponding module.

       CONFIG_HOTPLUG_PCI_PCIE=y
       CONFIG_HOTPLUG_PCI=y
       CONFIG_HOTPLUG_PCI_ACPI=m

       For example, in the 2.6 kernel of CentOS 5.x, the module for CONFIG_HOTPLUG_PCI_ACPI is acpiphp.ko. To load the module, run the modprobe acpiphp command. If the module fails to load, you can upgrade to a newer kernel version or stop the instance and perform a cold-plug operation.

       Important

       We recommend that you do not arbitrarily upgrade the kernel or operating system of your ECS instance. If you need to upgrade the kernel, see Prevent a Linux instance from failing to start after a kernel upgrade.

• Instance shutdown after a kernel panic

  Symptom

  When a kernel panic occurs, the system loads a second kernel (also known as a capture kernel) to perform a memory dump and generate Kdump logs. Due to a compatibility issue with bare metal instance types, disk discovery can fail during the boot of the capture kernel. This failure prevents the collection of Kdump logs, causes the capture kernel to fail, and shuts down the instance. You must then restart the instance from the console.

  For more information about bare metal instance types, see Instance families.

  Cause

  On bare metal instances, using the Kdump service to generate a dump file may fail.

  • This issue occurs on 6th-generation ebm* series bare metal instances running one of the following images:

    • CentOS 8.3 and earlier versions

    • Ubuntu 16 and 18

    • Debian 10

    • Alibaba Cloud Linux 2 with a kernel version that is earlier than 4.19.91-24.al7 (This issue is fixed in 4.19.91-24.al7 and later versions.)

  • This issue occurs on 7th-generation ebm* series bare metal instances running the Debian 10 image.

  Solution

  • CentOS and other images

    We recommend that you change the operating system to a more recent version. For more information, see Change the operating system (replace the system disk).

  • Alibaba Cloud Linux 2 images

    We recommend that you upgrade the kernel to version 4.19.91-24.al7 or later. Follow these steps:

    1. Establish a remote connection to the ECS instance.

       For more information, see Log on to a Linux instance by using Workbench.

    2. Run the following command to check the kernel version:

       uname -r

    3. Run the following command to upgrade the kernel:

       sudo yum update kernel

    4. Run the following command to restart the ECS instance for the new kernel to take effect:

       sudo reboot