SSTables 3.0 Statistics File Format¶

// Each bucket represents values from (previous bucket offset, current offset].
// Offset for last bucket is +inf.
using estimated_histogram = array<be32<int32_t>, bucket>;

struct bucket {
    // Offset of the previous bucket
    // In the first bucket this is offset of the first bucket itself because there's no previous bucket.
    // The offset of the first bucket is repeated in second bucket as well.
    be64<int64_t> prev_bucket_offset;
    // This bucket value
    be64<int64_t> value;
}

struct streaming_histogram {
    // Maximum number of buckets this historgam can have
    be32<int32_t> bucket_number_limit;
    array<be32<int32_t>, bucket> buckets;
}

struct bucket {
    // Offset of this bucket
    be64<double> offset;
    // Bucket value
    be64<int64_t> value;
}

struct commit_log_position {
    be64<int64_t> segment_id;
    be32<int32_t> position_in_segment;
}

struct statistics {
    // In bytes, uncompressed sizes of partitions
    estimated_histogram partition_sizes;
    // Number of cells per partition
    estimated_histogram column_counts;
    commit_log_position commit_log_upper_bound;
    // Typically in microseconds since the unix epoch, although this is not enforced
    be64<int64_t> min_timestamp;
    // Typically in microseconds since the unix epoch, although this is not enforced
    be64<int64_t> max_timestamp;
    // In seconds since the unix epoch
    be32<int32_t> min_local_deletion_time;
    // In seconds since the unix epoch
    be32<int32_t> max_local_deletion_time;
    be32<int32_t> min_ttl;
    be32<int32_t> max_ttl;
    // compressed_size / uncompressed_size
    be64<double> compression_rate;
    // Histogram of cell tombstones.
    // Keys are local deletion times of tombstones
    streaming_histogram tombstones;
    be32<int32_t> level;
    // The difference, measured in milliseconds, between repair time and midnight, January 1, 1970 UTC
    be64<int64_t> repaired_at;
    // Minimum and Maximum clustering key prefixes present in the SSTable (valid since the "md" SSTable format).
    // Note that:
    // - Clustering rows always have the full clustering key.
    // - Range tombstones may have a partial clustering key prefix.
    // - Partition tombstones implicitly apply to the full, unbound clustering range.
    // Therefore, an empty (min|max)_clustering_key denotes a respective unbound range,
    // derived either from an open-ended range tombstone, or from a partition tombstone.
    clustering_bound min_clustering_key;
    clustering_bound max_clustering_key;
    be8<bool> has_legacy_counters;
    be64<int64_t> number_of_columns;
    be64<int64_t> number_of_rows;

    // Version MA of SSTable 3.x format ends here.
    // It contains only one commit log position interval - [NONE = new CommitLogPosition(-1, 0), upper bound of commit log]

    commit_log_position commit_log_lower_bound;

    // Version MB of SSTable 3.x format ends here.
    // It contains only one commit log position interval - [lower bound of commit log, upper bound of commit log].

    array<be32<int32_t>, commit_log_interval> commit_log_intervals;

    // Versions MC and MD of SSTable 3.x format end here.

    // UUID of the host that wrote the SSTable.
    // Qualifies all commitlog positions in the SSTable Statistics file.

    UUID host_id;
}

using clustering_bound = array<be32<int32_t>, clustering_column>;
using clustering_column = array<be16<uint16_t>, be8>;

struct commit_log_interval {
    commit_log_position start;
    commit_log_position end;
}