Apsara File Storage NAS:Mount an NFS file system on a Linux ECS instance

Step 1: Install an NFS client

Before you mount an NFS file system on a Linux ECS instance, you must install an NFS client. After you install the NFS client, you no longer need to install the client the next time you mount a file system on the ECS instance.

1. Connect to the ECS instance. For more information, see Connection methods.

2. Install an NFS client.

   Operating system	Installation command
   Alibaba Cloud Linux	sudo yum install nfs-utils
   CentOS
   Redhat
   Ubuntu	Run the following commands in sequence: sudo apt-get update sudo apt-get install nfs-common
   Debian

3. Increase the maximum number of concurrent NFS requests.

   Run the following command to set the maximum number of concurrent NFS requests to 128. For more information, see How do I change the maximum number of concurrent NFS requests from an NFS client?

   if (lsmod | grep sunrpc); then
   (modinfo sunrpc | grep tcp_max_slot_table_entries) && sysctl -w sunrpc.tcp_max_slot_table_entries=128
   (modinfo sunrpc | grep tcp_slot_table_entries) && sysctl -w sunrpc.tcp_slot_table_entries=128
   fi
   (modinfo sunrpc | grep tcp_max_slot_table_entries) && echo "options sunrpc tcp_max_slot_table_entries=128" >> /etc/modprobe.d/sunrpc.conf
   (modinfo sunrpc | grep tcp_slot_table_entries) && echo "options sunrpc tcp_slot_table_entries=128" >> /etc/modprobe.d/sunrpc.conf

Step 2: Mount the NFS file system

In Linux, NFS file systems can be mounted manually or automatically. When you restart the ECS instance on which an NFS file system is mounted, the mount information about the file system may be lost. To prevent this issue, we recommend that you enable automatic mounting for the ECS instance after you manually mount the NFS file system.

Appendix: How NFS caching works and the related issues

In traditional disks, all data is cached into the page cache, and modified pages are asynchronously flushed back to the server. The latency of traditional disks is low. However, in an NFS file system, NFS does not cache newly created files or newly written content into the page cache, but flushes them back to the NAS server as soon as possible. Therefore, when multiple ECS instances share an NFS file system, all NAS operations cause one more overhead than disk operations. This overhead is generally between 100 us and 1 ms. Flushing data back to the NAS server as soon as possible involves the following multi-node consistency models provided by NAS:

Timeout-based eventual consistency model

For more information about the timeout-based eventual consistency model provided by NFS, see NFS.

File-based close-to-open (CTO) consistency model

The timeout-based eventual consistency model cannot ensure that ECS-2 immediately reads the data written by ECS-1. Therefore, to improve performance, NFS provides the file-based CTO consistency model. When two or more compute nodes concurrently read and write the same file, the changes made by ECS-1 may not be immediately read by ECS-2. However, once ECS-1 writes data into the file and closes the file, reopening the file on any compute node ensures access to the newly written data.

For example, a producer ECS instance produces file X and then executes the close operation. Then, the producer ECS instance sends message X to Message Queue, stating that file X has been produced. A consumer ECS instance that has subscribed to Message Queue reads message X (file X has been produced). Then, the consumer ECS instance executes the open operation on the file and reads the file through fd returned by the open operation. This way, the consumer ECS instance can definitely read all the content of file X. Assume that the consumer ECS instance has executed the open operation on file X and obtained fd before the producer ECS instance completes the file production. In this case, the consumer ECS instance may not be able to read the latest file content by directly using the fd after receiving message X.

Typical issues

File creation delayed

• Issue

  ECS-1 creates the abc file, but it takes some time for ECS-2 to read the abc file. The latency ranges from 1 second to 1 minute.

• Cause

  The issue is caused by the lookup cache, which meets the expected time T. For example, ECS-2 accesses the abc file before ECS-1 has created it, causing the issue that the file does not exist on ECS-2. As a result, a record indicating that the abc file does not exist is cached. FileAttr has not expired within time T. Therefore, when ECS-2 accesses the file again, it still accesses the record indicating that the abc file does not exist, which was cached the first time that ECS-2 accessed the file.

• Solutions

  To ensure that ECS-2 can read the file immediately after ECS-1 creates it, you can use one of the following solutions:

  • Solution 1: Disable negative lookup cache for ECS-2 so that files that do not exist are not cached. This solution causes the minimum overhead.

    Add the lookupcache=positive (default value: lookupcache=all) field when you mount the file system. Run the following command to mount the file system:

    sudo mount -t nfs -o vers=3,nolock,proto=tcp,rsize=1048576,wsize=1048576,hard,timeo=600,retrans=2,noresvport,lookupcache=positive file-system-id.region.nas.aliyuncs.com:/ /mnt

  • Solution 2: Disable all caches on ECS-2. This solution results in poor performance. Select an appropriate solution based on your business requirements.

    Add the actimeo=0 field when you mount the file system. Run the following command to mount the file system:

    sudo mount -t nfs -o vers=3,nolock,proto=tcp,rsize=1048576,wsize=1048576,hard,timeo=600,retrans=2,noresvport,actimeo=0 file-system-id.region.nas.aliyuncs.com:/ /mnt

Writing data into a file delayed

• Issue

  ECS-1 has updated the abc file. However, when ECS-2 reads the file immediately, the file content is still not updated.

• Causes

  The following two causes are involved:

  • Cause 1: After ECS-1 writes the abc file, ECS-1 does not flush the content immediately. Instead, it caches the content into the page cache and relies on the application layer to call the fsync or close operation.

  • Cause 2: File caches exist on ECS-2. Therefore, ECS 2 may not immediately obtain the latest file content from the server. For example, ECS-2 has cached the data when ECS-1 updates the abc file. As a result, the cached content is still used when ECS-2 reads the file.

• Solutions

  To ensure that ECS-2 can read the file immediately after ECS-1 creates it, you can use one of the following solutions:

  • Solution 1: Apply the file-based close-to-open (CTO) consistency model so that the read and write operations on ECS-1 or ECS-2 conform to CTO consistency. This way, ECS-2 can definitely read the latest data. Specifically, ECS-1 executes the close or fsync operation after it updates a file. ECS-2 executes the open operation before it reads the file.

  • Solution 2: Disable all caches on ECS-1 and ECS-2. This solution results in poor performance. Select an appropriate solution based on your business requirements.

    • Disable caching on ECS-1. When you mount the file system, add the noac field to ensure that all written data is immediately flushed into the disk. Run the following command to mount the file system:

      sudo mount -t nfs -o vers=3,nolock,proto=tcp,rsize=1048576,wsize=1048576,hard,timeo=600,retrans=2,noresvport,noac file-system-id.region.nas.aliyuncs.com:/ /mnt

      Note

      • If you call the fsync operation after the write operation on ECS-1 is complete or you call the sync operation to write data, replace "noac" in the preceding command with "actimeo=0" to improve the performance slightly.

      • noac is equivalent to actimeo=0 plus sync. In this case, all write operations are forcibly executed by using sync.

    • Disable caching on ECS-2. When you mount the file system, add the actimeo=0 field to omit all caches. Run the following command to mount the file system:

      sudo mount -t nfs -o vers=3,nolock,proto=tcp,rsize=1048576,wsize=1048576,hard,timeo=600,retrans=2,noresvport,actimeo=0 file-system-id.region.nas.aliyuncs.com:/ /mnt

References

• Mount a file system across VPCs or Alibaba Cloud accounts

• Access file systems in on-premises data centers

• Use POSIX ACLs to control access

• Use NFSv4 ACLs to control access

• Back up files from a General-purpose NAS file system

• Recycle bin