Skip to content
  • Priyanka Sangam's avatar
    20703c78
    Bug 32367250 - ABORTED NDB BACKUP DUE TO FULL FILE SYSTEM CAUSES CLUSTER CRASH · 20703c78
    Priyanka Sangam authored
    
    
    Data nodes crash on encountering an 'out-of-disk-space' error during
    a compressed backup. The crash occurs when NDBFS attempts to append
    a block of data to the backup file with a compressed write and fails.
    The return code which indicates an error for compressed writes is
    incorrectly treated as a fatal error and a crash is triggered in
    response.
    
    This is fixed by adding a check to handle the error return code from
    compressed writes. The system error is extracted and set so that the
    backup fails with an appropriate error "No space left on the device:
    Temporary error: Temporary Resource error".
    
    An mtr test using error insertion is added to test this scenario. Error
    inserts are implemented to simulate out-of-disk errors in ndbzwrite
    for writes to backup datafiles.
    
    Reviewed-by: default avatarFrazer Clement <frazer.clement@oracle.com>
    Change-Id: Ia9d67188a95e7e14bef6a032e5b472ada954700b
    20703c78
    Bug 32367250 - ABORTED NDB BACKUP DUE TO FULL FILE SYSTEM CAUSES CLUSTER CRASH
    Priyanka Sangam authored
    
    
    Data nodes crash on encountering an 'out-of-disk-space' error during
    a compressed backup. The crash occurs when NDBFS attempts to append
    a block of data to the backup file with a compressed write and fails.
    The return code which indicates an error for compressed writes is
    incorrectly treated as a fatal error and a crash is triggered in
    response.
    
    This is fixed by adding a check to handle the error return code from
    compressed writes. The system error is extracted and set so that the
    backup fails with an appropriate error "No space left on the device:
    Temporary error: Temporary Resource error".
    
    An mtr test using error insertion is added to test this scenario. Error
    inserts are implemented to simulate out-of-disk errors in ndbzwrite
    for writes to backup datafiles.
    
    Reviewed-by: default avatarFrazer Clement <frazer.clement@oracle.com>
    Change-Id: Ia9d67188a95e7e14bef6a032e5b472ada954700b
Loading