HBase
  1. HBase
  2. HBASE-8500

RollingRestart utility at regionserver level

    Details

    • Type: Bug Bug
    • Status: Open
    • Priority: Minor Minor
    • Resolution: Unresolved
    • Affects Version/s: 0.89-fb
    • Fix Version/s: None
    • Component/s: master
    • Labels:
      None

      Description

      With the option of blacklisting a region server from getting assigned new regions by the master, we need to have a utility which encapsulates this and uses it for rolling restart of a hbase cluster. The JIRA is to create a utility which restarts a single region server by:
      1. blacklisting the region server
      2. moving all the regions one by one
      3. restart the region server process
      4. moving all the regions back
      5. removing the region server from blacklist

      Along with this, the utility should also be able to estimate the total outage observed from client perspective.

        Activity

        Hide
        Rishit Shroff added a comment -

        Thanks Stack,

        The idea is almost the same, the difference is that of disabling load balancer vs blacklisting a region server. In 0.89 fb load balancer is responsible for assigning regions to preferred nodes when a region server dies. So if we disable the load balancer during a rolling restart, and some other region server dies during that time, then the regions won't failover and get reassigned. Because of this limitation in 0.89 fb, we went with the idea of blacklisting the region server under restart so that the load balancer is on and working fine. Only the region server under restart is not assigned new regions by the load balancer.

        I will take a look at the script and check if I can integrate any part of it into this idea.

        Show
        Rishit Shroff added a comment - Thanks Stack, The idea is almost the same, the difference is that of disabling load balancer vs blacklisting a region server. In 0.89 fb load balancer is responsible for assigning regions to preferred nodes when a region server dies. So if we disable the load balancer during a rolling restart, and some other region server dies during that time, then the regions won't failover and get reassigned. Because of this limitation in 0.89 fb, we went with the idea of blacklisting the region server under restart so that the load balancer is on and working fine. Only the region server under restart is not assigned new regions by the load balancer. I will take a look at the script and check if I can integrate any part of it into this idea.
        Hide
        stack added a comment -
        Show
        stack added a comment - Does this help Rishit Shroff : http://hbase.apache.org/book.html#decommission

          People

          • Assignee:
            Rishit Shroff
            Reporter:
            Rishit Shroff
          • Votes:
            0 Vote for this issue
            Watchers:
            3 Start watching this issue

            Dates

            • Created:
              Updated:

              Development