diff options
author | qrort <qrort@yandex-team.com> | 2022-11-30 23:47:12 +0300 |
---|---|---|
committer | qrort <qrort@yandex-team.com> | 2022-11-30 23:47:12 +0300 |
commit | 22f8ae0e3f5d68b92aecccdf96c1d841a0334311 (patch) | |
tree | bffa27765faf54126ad44bcafa89fadecb7a73d7 /library/cpp/erasure/codec.h | |
parent | 332b99e2173f0425444abb759eebcb2fafaa9209 (diff) | |
download | ydb-22f8ae0e3f5d68b92aecccdf96c1d841a0334311.tar.gz |
validate canons without yatest_common
Diffstat (limited to 'library/cpp/erasure/codec.h')
-rw-r--r-- | library/cpp/erasure/codec.h | 80 |
1 files changed, 80 insertions, 0 deletions
diff --git a/library/cpp/erasure/codec.h b/library/cpp/erasure/codec.h new file mode 100644 index 0000000000..c03d25e9c8 --- /dev/null +++ b/library/cpp/erasure/codec.h @@ -0,0 +1,80 @@ +#pragma once + +#include "public.h" + +#include <optional> +#include <vector> + +namespace NErasure { + +//! Describes a generic way to generate parity blocks from data blocks and +//! to recover (repair) missing blocks. +/*! + * Given N data blocks (numbered from 0 to N - 1) one can call #Encode to generate + * another M parity blocks (numbered from N to N + M - 1). + * + * If some of the resulting N + M blocks ever become missing one can attempt to + * repair the missing blocks by calling #Decode. + * + * Here N and M are fixed (codec-specific) parameters. + * Call #GetDataPartCount and #GetParityPartCount to figure out the + * the values for N and M, respectively. + * + */ +template <class TBlobType> +struct ICodec { + //! Computes a sequence of parity blocks for given data blocks. + /*! + * The size of #blocks must be equal to #GetDataPartCount. + * The size of the returned array is equal to #GetParityPartCount. + */ + virtual std::vector<TBlobType> Encode(const std::vector<TBlobType>& blocks) const = 0; + + //! Decodes (repairs) missing blocks. + /*! + * #erasedIndices must contain the set of erased blocks indices. + * #blocks must contain known blocks (in the order specified by #GetRepairIndices). + * \returns The repaired blocks. + */ + virtual std::vector<TBlobType> Decode( + const std::vector<TBlobType>& blocks, + const TPartIndexList& erasedIndices) const = 0; + + //! Given a set of missing block indices, returns |true| if missing blocks can be repaired. + //! Due to performance reasons the elements of #erasedIndices must unique and sorted. + virtual bool CanRepair(const TPartIndexList& erasedIndices) const = 0; + + //! Rapid version that works with set instead of list. + virtual bool CanRepair(const TPartIndexSet& erasedIndices) const = 0; + + //! Given a set of missing block indices, checks if missing blocks can be repaired. + /*! + * \returns + * If repair is not possible, returns |std::nullopt|. + * Otherwise returns the indices of blocks (both data and parity) to be passed to #Decode + * (in this very order). Not all known blocks may be needed for repair. + */ + virtual std::optional<TPartIndexList> GetRepairIndices(const TPartIndexList& erasedIndices) const = 0; + + //! Returns the number of data blocks this codec can handle. + virtual int GetDataPartCount() const = 0; + + //! Returns the number of parity blocks this codec can handle. + virtual int GetParityPartCount() const = 0; + + //! Returns the maximum number of blocks that can always be repaired when missing. + virtual int GetGuaranteedRepairablePartCount() const = 0; + + //! Every block passed to this codec must have size divisible by the result of #GetWordSize. + virtual int GetWordSize() const = 0; + + // Extension methods + + //! Returns the sum of #GetDataPartCount and #GetParityPartCount. + int GetTotalPartCount() const { + return GetDataPartCount() + GetParityPartCount(); + } +}; + +} // namespace NErasure + |