This sample provides a script that can be used to warm a FSx for ONTAP volume. In other words, it tries to ensure that all the blocks for a volume are in the "performance tier" as opposed to the "capacity tier." It does that by simply reading every byte of every file in the volume. Doing that causes all blocks that are currently in the capacity tier to be pulled into the performance tier before being returned to the reader. At that point, assuming the tiering policy is not set to 'all' or 'snapshot', all the data should remain in the performance tier until ONTAP tiers it back based on the volume's tiering policy.
Note that, by default, Data ONTAP will not store data in the performance tier from the capacity tier if it detects that the data is being read sequentially. This is to keep things like backups and virus scans from filling up the performance tier. You can, and should, override this behavior by setting the cloud-retrieval-policy to "on-read" for the volume. Examples on how to do that are shown below.
In an additional effort to try to get ONTAP to keep data in the performance tier after reading it in, this script will read files in "reverse" order. Meaning it will read the last block of the file first, then the second to last block, and so on.
To speed up the process, the script will spawn multiple threads to process the volume. It will spawn a separate thread for each directory in the volume, and then a separate thread for each file in that directory. The number of directory threads is controlled by the -t option. The number of reader threads is controlled by the -x option. Note that the script will spawn -x reader threads per directory thread. So, for example, if you have 2 directory threads and 5 reader threads, you could have up to 10 reader threads running at one time.
Since the goal of this script is to force all the data that is currently
in the capacity tier to the performance tier you should ensure you have
enough free space in your performance tier to hold all the data in the volume.
You can use the volume show-footprint ONTAP command to see how much space
is currently in the capacity tier. You can then use storage aggregate show
to see how much space is available in the performance tier.
Note that it will not be uncommon for there to still be data in the capacity tier after running this script. There can be several reasons for that. For example:
- Space is from snapshots that aren't part of the live volume anymore.
- Space from blocks that are part of an object in the object store, but aren't part of the volume. This space will get consolidated eventually.
- Some is from metadata that is always kept in the capacity tier.
Even with the reasons mentioned above, we have found that running the script twice does, typically, get more data into the performance tier so if you are trying to get as much data as possible into the performance tier, it is recommended to run the script twice.
The first step is to ensure the volume's tiering policy is set to something other than "all" or "snapshot-only". You should also ensure that the cloud-retrieval-policy is set to "on-read". You can make both of these changes with the following commands:
set advanced -c off
volume modify -vserver <vserver> -volume <volume> -tiering-policy auto -cloud-retrieval-policy on-read
Where <vserver> is the name of the SVM and <volume> is the name of the volume.
The next step is to copy the script to a Linux based host that is able to NFS mount the volume to be warmed. If the volume is already mounted, then any user that has read access to all the files in the volume can run it. Otherwise, the script needs to be run as 'root' so it can mount the volume before reading the files.
If the 'root' user can't read the all files in the volume, then you should use the 'root' user just to mount the volume and then run the script from a user ID that can read the contents of all the files in the volume.
There are two main ways to run the script. The first is to just provide the script with a directory to start from using the -d option. The script will then read every file in that directory and all its subdirectories. The second way is to provide the script with the FSx for ONTAP file system data endpoint with the -f option and the volume name with the -v option. The script will then attempt to mount the volume if it isn't already mounted. If it does mount it, it will mount it read-only and unmount when it is done.
In order to mount the volume the script assumes that the junction path is the same as the volume name. If this isn't the case, then just mount the volume first before running the script and provide the path to the mount point with the -d option.
To run this script you just need to change the UNIX permissions on the file to be executable, then run it as a command:
chmod +x warm_performance_tier
./warm_performance_tier -d /path/to/mount/point
The above example will force the script to read every file in the /path/to/mount/point directory and any directory under it.
If you want the script to ensure the volume is mounted, you can specify the file system data endpoint and volume name:
./warm_performance_tier -f fsxfileserver.us-west-2.amazonaws.com -v myvolume
By default there won't be any output from the script. You can provide a -V option to get verbose output. This will display the thread ID, date (in epoch seconds), then the directory or file being processed.
If you run the script with a '-h' option, you will see the following help message:
Usage: warm_performance_tier [-f filesystem_endpoint] [-v volume_name] [-d directory] [-t max_directory_threads] [-x max_read_threads] [-n nfs_type] [-h] [-V]
Where:
-f filesystem_endpoint - Is the hostname or IP address of the FSx for ONTAP file system.
-v volume_name - Is the name of the volume.
-n nfs_type - Is the NFS version to use. Default is nfs4.
-d directory - Is the root directory to start the process from.
-t max_directory_threads - Is the maximum number of threads to use to process directories. The default is 2.
-x max_read_threads - Is the maximum number of threads to use to read files. The default is 5.
-V - Enable verbose output. Displays the thread ID, date (in epoch seconds), then the directory or file being processed.
-h - Prints this help information.
Notes:
* The filesystem_endpoint, volume_name and nfs_type are used to mount the volume
if it is not already mounted. It will be mounted read-only. It assumes that
the junction path is the same as the volume name.
* For each directory thread, there will be a maximum of max_read_threads threads
reading files.
After running the script, you should set the cloud-retrieval-policy back to "default" by running the following commands:
set advanced -c off
volume modify -vserver <vserver> -volume <volume> -cloud-retrieval-policy default
Where <vserver> is the name of the SVM and <volume> is the name of the volume.
This repository is maintained by the contributors listed on GitHub.
Licensed under the Apache License, Version 2.0 (the "License").
You may obtain a copy of the License at apache.org/licenses/LICENSE-2.0.
Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on an "AS IS" basis, without WARRANTIES or conditions of any kind, either express or implied.
See the License for the specific language governing permissions and limitations under the License.
© 2024 NetApp, Inc. All Rights Reserved.