在FlexGroup中跨目录移动文件的速度较慢

适用场景

问题描述

• 在FlexGroup上发现性能较差。
• 文件移动需要几分钟或几小时才能完成。
  • 示例： 文件重命名时间从下午5：46到次日上午12：13。

    • [2023-10-10 17:46:29,751] {bash_operator.py:126} INFO - 2023.10.10D22:46:29.751918000 [info][0] Moving :/mnt/home/dir1/dir2/dir3/dir4/dir5/dir6/dir7/dir8$/file1_lastModified -> :/mnt/home/dir1/dir2/dir3/dir4/dir5/dir6/dir7/dir/lastModified [2023-10-11 00:13:48,255] {bash_operator.py:126} INFO - 2023.10.11D05:13:48.255446000 [info][0] Executed successfully

  • 中显示了高延迟 qos statistics volume latency show

    • Cluster::> qos statistics volume latency show -vol vol1 -vserver svm1 Workload ID Latency Network Cluster Data Disk QoS Max QoS Min NVRAM Cloud FlexCache SM Sync VA AVSCAN --------------- ------ ---------- ---------- ---------- ---------- ---------- ---------- ---------- ---------- ---------- ---------- ---------- ---------- ---------- -total- - 8.58ms 2.99ms 247.00us 5.32ms 2.00us 0ms 0ms 28.00us 0ms 0ms 0ms 0ms 0ms fg1-wid50693 50693 35.39s 16.59s 628.87ms 18.18s 0ms 0ms 0ms 1.80ms 0ms 0ms 0ms 0ms 0ms

  • NFS Replay Cache的倒带计数较高(几分钟内达到数百万)、这会导致网络处理延迟较高、如上所示：

    • Cluster::> set -privilege diagnostic; statistics start -object rw_ctx -counter nfs_rewinds (wait 5 minutes) Cluster::*> statistics show Object: rw_ctx Instance: rw_ctx Start-time: 10/25/2023 15:18:01 End-time: 10/25/2023 15:22:44 Elapsed-time: 283s Scope: node2 Counter Value -------------------------------- -------------------------------- nfs_rewinds - Credentials 0 ... Replay Cache 12697850 <--- ... Cluster::*> statistics stop