Skip to content

error_codes.md

Latest commit

 

History

History
35 lines (27 loc) · 2.72 KB

error_codes.md

File metadata and controls

35 lines (27 loc) · 2.72 KB
Raw

RAID async_iov Error Code Reference

Error codes returned directly by Raid0Disk::async_iov and Raid1Disk::async_iov. "Retry safe" means the caller can reissue the identical I/O without risk of data duplication or corruption.

RAID0

Op Condition Code Retry safe Notes
Any stripe_tasks.reserve() throws bad_alloc -EAGAIN Yes No SQEs submitted; no I/O started
Any __distribute returns error (currently unreachable) -EIO No Invariant: lambda always succeeds today; guard for future
READ io_uring_submit fails after fan-out -EAGAIN Yes SQEs in ring but not yet flushed; no data read
WRITE / DISCARD io_uring_submit fails after fan-out -EIO No Stripes may partially commit once ring is eventually flushed

In all submit-failure cases the started stripe tasks are drained before returning to avoid dangling cqe_state::_waiter handles.

RAID1

Op Condition Code Retry safe Notes
Any unknown Op is not READ / WRITE / DISCARD / WRITE_ZEROES / FLUSH -EINVAL Yes Programming error; FLUSH returns 0
READ Primary fails; no failover device available -EAGAIN Yes Both mirrors unavailable or dirty-region suppressed failover
READ Primary fails; failover attempted propagated - Failover result returned as-is
WRITE Active disk fails; __become_degraded fails; backup succeeded backup result Yes Backup holds valid data; array is in-memory degraded
WRITE Active disk fails; __become_degraded fails; backup failed or skipped -EAGAIN Yes No mirror holds a complete write; safe to retry
WRITE Active disk fails; __become_degraded succeeds; backup succeeded backup result Yes Array transitions to degraded; backup is authoritative
WRITE Active disk fails; __become_degraded succeeds; backup failed -EAGAIN Yes Active write failed, backup write failed; no data committed
WRITE Active succeeds; backup mode SKIP (degraded + unavail/dirty) active result - Degraded write; bitmap dirtied; resync will reconcile
WRITE Active succeeds; backup fails; __become_degraded fails -EIO No CAS lost or SB write failed; state indeterminate
WRITE Active succeeds; backup fails; __become_degraded succeeds active result - Array transitions to degraded; active write is authoritative

Notes on RAID1 -EAGAIN vs -EIO

  • -EAGAIN is used when no mirror holds a committed write - the caller can retry without risk of duplication.
  • -EIO is used when the superblock CAS/write fails during __become_degraded after a backup failure - the on-disk array state is indeterminate and a retry could produce inconsistent data.