1================
  2RAID 4/5/6 cache
  3================
  4
  5Raid 4/5/6 could include an extra disk for data cache besides normal RAID
  6disks. The role of RAID disks isn't changed with the cache disk. The cache disk
  7caches data to the RAID disks. The cache can be in write-through (supported
  8since 4.4) or write-back mode (supported since 4.10). mdadm (supported since
  93.4) has a new option '--write-journal' to create array with cache. Please
 10refer to mdadm manual for details. By default (RAID array starts), the cache is
 11in write-through mode. A user can switch it to write-back mode by::
 12
 13	echo "write-back" > /sys/block/md0/md/journal_mode
 14
 15And switch it back to write-through mode by::
 16
 17	echo "write-through" > /sys/block/md0/md/journal_mode
 18
 19In both modes, all writes to the array will hit cache disk first. This means
 20the cache disk must be fast and sustainable.
 21
 22write-through mode
 23==================
 24
 25This mode mainly fixes the 'write hole' issue. For RAID 4/5/6 array, an unclean
 26shutdown can cause data in some stripes to not be in consistent state, eg, data
 27and parity don't match. The reason is that a stripe write involves several RAID
 28disks and it's possible the writes don't hit all RAID disks yet before the
 29unclean shutdown. We call an array degraded if it has inconsistent data. MD
 30tries to resync the array to bring it back to normal state. But before the
 31resync completes, any system crash will expose the chance of real data
 32corruption in the RAID array. This problem is called 'write hole'.
 33
 34The write-through cache will cache all data on cache disk first. After the data
 35is safe on the cache disk, the data will be flushed onto RAID disks. The
 36two-step write will guarantee MD can recover correct data after unclean
 37shutdown even the array is degraded. Thus the cache can close the 'write hole'.
 38
 39In write-through mode, MD reports IO completion to upper layer (usually
 40filesystems) after the data is safe on RAID disks, so cache disk failure
 41doesn't cause data loss. Of course cache disk failure means the array is
 42exposed to 'write hole' again.
 43
 44In write-through mode, the cache disk isn't required to be big. Several
 45hundreds megabytes are enough.
 46
 47write-back mode
 48===============
 49
 50write-back mode fixes the 'write hole' issue too, since all write data is
 51cached on cache disk. But the main goal of 'write-back' cache is to speed up
 52write. If a write crosses all RAID disks of a stripe, we call it full-stripe
 53write. For non-full-stripe writes, MD must read old data before the new parity
 54can be calculated. These synchronous reads hurt write throughput. Some writes
 55which are sequential but not dispatched in the same time will suffer from this
 56overhead too. Write-back cache will aggregate the data and flush the data to
 57RAID disks only after the data becomes a full stripe write. This will
 58completely avoid the overhead, so it's very helpful for some workloads. A
 59typical workload which does sequential write followed by fsync is an example.
 60
 61In write-back mode, MD reports IO completion to upper layer (usually
 62filesystems) right after the data hits cache disk. The data is flushed to raid
 63disks later after specific conditions met. So cache disk failure will cause
 64data loss.
 65
 66In write-back mode, MD also caches data in memory. The memory cache includes
 67the same data stored on cache disk, so a power loss doesn't cause data loss.
 68The memory cache size has performance impact for the array. It's recommended
 69the size is big. A user can configure the size by::
 70
 71	echo "2048" > /sys/block/md0/md/stripe_cache_size
 72
 73Too small cache disk will make the write aggregation less efficient in this
 74mode depending on the workloads. It's recommended to use a cache disk with at
 75least several gigabytes size in write-back mode.
 76
 77The implementation
 78==================
 79
 80The write-through and write-back cache use the same disk format. The cache disk
 81is organized as a simple write log. The log consists of 'meta data' and 'data'
 82pairs. The meta data describes the data. It also includes checksum and sequence
 83ID for recovery identification. Data can be IO data and parity data. Data is
 84checksummed too. The checksum is stored in the meta data ahead of the data. The
 85checksum is an optimization because MD can write meta and data freely without
 86worry about the order. MD superblock has a field pointed to the valid meta data
 87of log head.
 88
 89The log implementation is pretty straightforward. The difficult part is the
 90order in which MD writes data to cache disk and RAID disks. Specifically, in
 91write-through mode, MD calculates parity for IO data, writes both IO data and
 92parity to the log, writes the data and parity to RAID disks after the data and
 93parity is settled down in log and finally the IO is finished. Read just reads
 94from raid disks as usual.
 95
 96In write-back mode, MD writes IO data to the log and reports IO completion. The
 97data is also fully cached in memory at that time, which means read must query
 98memory cache. If some conditions are met, MD will flush the data to RAID disks.
 99MD will calculate parity for the data and write parity into the log. After this
100is finished, MD will write both data and parity into RAID disks, then MD can
101release the memory cache. The flush conditions could be stripe becomes a full
102stripe write, free cache disk space is low or free in-kernel memory cache space
103is low.
104
105After an unclean shutdown, MD does recovery. MD reads all meta data and data
106from the log. The sequence ID and checksum will help us detect corrupted meta
107data and data. If MD finds a stripe with data and valid parities (1 parity for
108raid4/5 and 2 for raid6), MD will write the data and parities to RAID disks. If
109parities are incompleted, they are discarded. If part of data is corrupted,
110they are discarded too. MD then loads valid data and writes them to RAID disks
111in normal way.