Note:
This topic has been translated from a Chinese forum by GPT and might contain errors.
Original topic: TiKV服务部署 每个实例指定data_dir 多盘部署 故障域是怎么划分?
[TiDB Usage Environment] Testing
[TiDB Version] 7.5
[Reproduction Path] Use TiUP to set data_dir to multiple filesystem paths, such as: /data1, /data2, /data3, /data4
[Encountered Issue: If disk data1 is damaged but data2-4 are normal, will this TiKV instance be marked as faulty?]
[Resource Configuration] Enter TiDB Dashboard - Cluster Info - Hosts and take a screenshot of this page
[Attachment: Screenshot/Logs/Monitoring]
It seems that only TiFlash supports configuring multiple directories in data_dir, while TiKV does not.
Marking it, haven’t used it this way before. Is this for hot and cold data separation?
A single TiKV instance does not support multiple data_dirs.
Each directory for one TiKV or one TiKV using multiple disks in a RAID configuration.
Can it be played like this? It feels a bit like RAID.
Can this thing be made into a RAID? Looking forward to it.
I also encountered this problem. Using a cloud host, a single disk has an IO limit. I also want to deploy multiple disks, but there is no solution yet.
TiKV does not support fault domains.
TiKV indeed does not support multiple data_dirs.
TiKV does not support configuring multiple directories for a single node; only TiFlash supports this.
Only TiFlash supports it; TiKV does not support fault domains.
TiKV does not support this, right?
Tikv does not support configuring multiple directories for a single node.
TiKV does not support it unless you configure your disk with RAID.
You can use LVM striped volumes to achieve multi-disk acceleration for your issue.
Thank you all for your answers. During the experiment, I found the answer: the data_dir
of TiKV does not support configuring multiple disk paths.
My current situation is deploying TiKV on two physical machines (with 3 data disks each) and planning how to utilize the disks.
The current solution is:
Deploy the corresponding TiKV instances on each physical machine according to the number of disks and bind a single disk to each instance.
Then use the labeling feature to divide fault domains.
As @zhh_912 mentioned, there is no concept of fault domains, but the labeling feature can be used to achieve the functionality of fault domains.
Schedule Replicas by Topology Labels | PingCAP Documentation Center
This is a bit frustrating. After using Doris for a while, I’ve gotten used to attaching as many disks as possible on bare metal to improve IO. 