Storage负载均衡

用户可以使用BALANCE语句平衡分片和Raft leader的分布，或者删除冗余的Storage服务器。

前提条件

为了平衡分片和Raft leader的分布，Nebula Graph中图空间的副本数都必须大于1。

均衡分片分布

BALANCE DATA语句会开始一个任务，将Nebula Graph集群中的分片平均分配到所有Storage服务器。通过创建和执行一组子任务来迁移数据和均衡分片分布。

示例

以横向扩容Nebula Graph为例，集群中增加新的Storage服务器后，新服务器上没有分片。

  1. 执行命令SHOW HOSTS检查分片的分布。

    1. nebual> SHOW HOSTS;
    2. +-------------+------+----------+--------------+-----------------------------------+------------------------+
    3. | Host        | Port | Status   | Leader count | Leader distribution               | Partition distribution |
    4. +-------------+------+----------+--------------+-----------------------------------+------------------------+
    5. | "storaged0" | 9779 | "ONLINE" | 4            | "basketballplayer:4"              | "basketballplayer:15"  |
    6. +-------------+------+----------+--------------+-----------------------------------+------------------------+
    7. | "storaged1" | 9779 | "ONLINE" | 8            | "basketballplayer:8"              | "basketballplayer:15"  |
    8. +-------------+------+----------+--------------+-----------------------------------+------------------------+
    9. | "storaged2" | 9779 | "ONLINE" | 3            | "basketballplayer:3"              | "basketballplayer:15"  |
    10. +-------------+------+----------+--------------+-----------------------------------+------------------------+
    11. | "storaged3" | 9779 | "ONLINE" | 0            | "No valid partition"              | "No valid partition"   |
    12. +-------------+------+----------+--------------+-----------------------------------+------------------------+
    13. | "storaged4" | 9779 | "ONLINE" | 0            | "No valid partition"              | "No valid partition"   |
    14. +-------------+------+----------+--------------+-----------------------------------+------------------------+
    15. | "Total"     |      |          | 15           | "basketballplayer:15"             | "basketballplayer:45"  |
    16. +-------------+------+----------+--------------+-----------------------------------+------------------------+

  2. 执行命令BALANCE DATA将所有分片均衡分布。

    1. nebula> BALANCE DATA;
    2. +------------+
    3. | ID         |
    4. +------------+
    5. | 1614237867 |
    6. +------------+

  3. 根据返回的任务ID，执行命令BALANCE DATA <balance_id>检查任务状态。

    1. nebula> BALANCE DATA 1614237867;
    2. +--------------------------------------------------------------+-------------------+
    3. | balanceId, spaceId:partId, src->dst                          | status            |
    4. +--------------------------------------------------------------+-------------------+
    5. | "[1614237867, 11:1, storaged1:9779->storaged3:9779]"         | "SUCCEEDED"       |
    6. +--------------------------------------------------------------+-------------------+
    7. | "[1614237867, 11:1, storaged2:9779->storaged4:9779]"         | "SUCCEEDED"       |
    8. +--------------------------------------------------------------+-------------------+
    9. | "[1614237867, 11:2, storaged1:9779->storaged3:9779]"         | "SUCCEEDED"       |
    10. +--------------------------------------------------------------+-------------------+
    11. ...
    12. +--------------------------------------------------------------+-------------------+
    13. | "Total:22, Succeeded:22, Failed:0, In Progress:0, Invalid:0" | 100               |
    14. +--------------------------------------------------------------+-------------------+

  4. 等待所有子任务完成，负载均衡进程结束，执行命令SHOW HOSTS确认分片已经均衡分布。

    1. nebula> SHOW HOSTS;
    2. +-------------+------+----------+--------------+-----------------------------------+------------------------+
    3. | Host        | Port | Status   | Leader count | Leader distribution               | Partition distribution |
    4. +-------------+------+----------+--------------+-----------------------------------+------------------------+
    5. | "storaged0" | 9779 | "ONLINE" | 4            | "basketballplayer:4"              | "basketballplayer:9"   |
    6. +-------------+------+----------+--------------+-----------------------------------+------------------------+
    7. | "storaged1" | 9779 | "ONLINE" | 8            | "basketballplayer:8"              | "basketballplayer:9"   |
    8. +-------------+------+----------+--------------+-----------------------------------+------------------------+
    9. | "storaged2" | 9779 | "ONLINE" | 3            | "basketballplayer:3"              | "basketballplayer:9"   |
    10. +-------------+------+----------+--------------+-----------------------------------+------------------------+
    11. | "storaged3" | 9779 | "ONLINE" | 0            | "No valid partition"              | "basketballplayer:9"   |
    12. +-------------+------+----------+--------------+-----------------------------------+------------------------+
    13. | "storaged4" | 9779 | "ONLINE" | 0            | "No valid partition"              | "basketballplayer:9"   |
    14. +-------------+------+----------+--------------+-----------------------------------+------------------------+
    15. | "Total"     |      |          | 15           | "basketballplayer:15"             | "basketballplayer:45"  |
    16. +-------------+------+----------+--------------+-----------------------------------+------------------------+

如果有子任务失败，请重新执行BALANCE DATA。如果重做负载均衡仍然不能解决问题，请到Nebula Graph社区寻求帮助。

停止负载均衡任务

停止负载均衡任务，请执行命令BALANCE DATA STOP。

  • 如果没有正在执行的负载均衡任务，会返回错误。

  • 如果有正在执行的负载均衡任务，会返回停止的任务ID（balance_id）。

BALANCE DATA STOP不会停止正在执行的子任务，而是取消所有后续子任务。用户可以执行命令BALANCE DATA <balance_id>检查停止的任务状态。

一旦所有子任务都完成或停止，用户可以再次执行命令BALANCE DATA。

  • 如果前一个负载均衡任务的任何一个子任务失败，Nebula Graph会重新启动之前的负载均衡任务。

  • 如果前一个负载均衡任务的任何一个子任务都没有失败，Nebula Graph会启动一个新的的负载均衡任务。

重置负载均衡任务

如果停止负载均衡任务后重新执行仍然失败，可以尝试用命令BALANCE DATA RESET PLAN重置负载均衡任务，该操作会清空旧的任务。之后再使用BALANCE DATA命令，会新建负载均衡任务，而不是执行旧的任务。

移除Storage服务器

移除指定的Storage服务器来缩小集群规模，可以使用命令BALANCE DATA REMOVE <host_list>。

示例

如果需要移除以下两台Storage服务器。

服务器名称IP地址端口
storage3192.168.0.89779
storage4192.168.0.99779

请执行如下命令：

  1. BALANCE DATA REMOVE 192.168.0.8:9779,192.168.0.9:9779;

Nebula Graph将启动一个负载均衡任务，迁移storage3和storage4中的分片，然后将服务器从集群中移除。

均衡leader分布

BALANCE DATA只能均衡分片分布，不能均衡Raft leader分布。用户可以使用命令BALANCE LEADER均衡leader分布。

示例

  1. nebula> BALANCE LEADER;

用户可以执行SHOW HOSTS检查结果。

  1. nebula> SHOW HOSTS;
  2. +-------------+------+----------+--------------+-----------------------------------+------------------------+
  3. | Host        | Port | Status   | Leader count | Leader distribution               | Partition distribution |
  4. +-------------+------+----------+--------------+-----------------------------------+------------------------+
  5. | "storaged0" | 9779 | "ONLINE" | 3            | "basketballplayer:3"              | "basketballplayer:9"   |
  6. +-------------+------+----------+--------------+-----------------------------------+------------------------+
  7. | "storaged1" | 9779 | "ONLINE" | 3            | "basketballplayer:3"              | "basketballplayer:9"   |
  8. +-------------+------+----------+--------------+-----------------------------------+------------------------+
  9. | "storaged2" | 9779 | "ONLINE" | 3            | "basketballplayer:3"              | "basketballplayer:9"   |
  10. +-------------+------+----------+--------------+-----------------------------------+------------------------+
  11. | "storaged3" | 9779 | "ONLINE" | 3            | "basketballplayer:3"              | "basketballplayer:9"   |
  12. +-------------+------+----------+--------------+-----------------------------------+------------------------+
  13. | "storaged4" | 9779 | "ONLINE" | 3            | "basketballplayer:3"              | "basketballplayer:9"   |
  14. +-------------+------+----------+--------------+-----------------------------------+------------------------+
  15. | "Total"     |      |          | 15           | "basketballplayer:15"             | "basketballplayer:45"  |
  16. +-------------+------+----------+--------------+-----------------------------------+------------------------+