Disable I/O errors

HCL® OneDB® divides disabling I/O errors into two general types: destructive and nondestructive. A disabling I/O error is destructive when the disk that contains a database becomes damaged in some way. This type of event threatens the integrity of data, and the database server marks the chunk and dbspace as down. The database server prohibits access to the damaged disk until you repair or replace the disk and perform a physical and logical restore.

A disabling I/O error is nondestructive when the error does not threaten the integrity of your data. Nondestructive errors occur when someone accidentally disconnects a cable, you somehow erase the symbolic link that you set up to point to a chunk, or a disk controller becomes damaged.

Before the database server considers an I/O error to be disabling, the error must meet two criteria. First, the error must occur when the database server attempts to perform an operation on a chunk that has at least one of the following characteristics:
  • The chunk has no mirror.
  • The primary or mirror companion of the chunk under question is offline.
Second, the error must occur when the database server attempts unsuccessfully to perform one of the following operations:
  • Seek, read, or write on a chunk
  • Open a chunk
  • Verify that chunk information about the first used page is valid

    The database server performs this verification as a sanity check immediately after it opens a chunk.

You can prevent the database server from marking a dbspace as down while you investigate disabling I/O errors. If you find that the problem is trivial, such as a loose cable, you can bring the database server offline and then online again without restoring the affected dbspace from backup. If you find that the problem is more serious, such as a damaged disk, you can use onmode -O to mark the affected dbspace as down and continue processing.