Honda Yasuhisa created HBASE-30554:
--------------------------------------

             Summary: Concurrent HbckChore invocations can trigger redundant 
sequential scans
                 Key: HBASE-30554
                 URL: https://issues.apache.org/jira/browse/HBASE-30554
             Project: HBase
          Issue Type: Bug
            Reporter: Honda Yasuhisa


Currently, {{HbckChore.runChore()}} checks whether the chore is running before 
calling the {{chore()}} method.  However, the check and the assignment of 
{{running = true}} are not atomic.
{code:java}
  public boolean runChore() {
    // This function does the sanity checks of making sure the chore is not run 
when it is
    // disabled or when it's already running. It returns whether the chore was 
actually run or not.
    if (isDisabled() || isRunning()) {
      if (isDisabled()) {
        LOG.warn("hbck chore is disabled! Set " + HBCK_CHORE_INTERVAL + " > 0 
to enable it.");
      } else {
        LOG.warn("hbck chore already running. Can't run till it finishes.");
      }
      return false;
    }
    chore();
    return true;
  } {code}
ref: 
[https://github.com/apache/hbase/blob/23ffbcca7c589758d1ea83a1035a7d451a215ce3/hbase-server/src/main/java/org/apache/hadoop/hbase/master/hbck/HbckChore.java#L122]

 
{code:java}
  @Override
  protected synchronized void chore() {
    if (isDisabled() || isRunning()) {
      LOG.warn("hbckChore is either disabled or is already running. Can't run 
the chore");
      return;
    }
    running = true;
    :{code}
ref: 
[https://github.com/apache/hbase/blob/23ffbcca7c589758d1ea83a1035a7d451a215ce3/hbase-server/src/main/java/org/apache/hadoop/hbase/master/hbck/HbckChore.java#L92]

 

Multiple threads can pass the check before any of them set running to 
{{{}true{}}}. While synchronization prevents concurrent execution of HbckChore, 
it doesn't prevent redundant sequential scans. A competing callers block until 
the current scan finishes.

Concurrent requests should be rejected (return false) instead of blocking until 
current scan finishes and then starting another scan.



--
This message was sent by Atlassian Jira
(v10.3.26#10030026)

Reply via email to