nordisk/edda

crates/nornir-parquet-delta/src/error.rs

/// Everything that can go wrong mirroring a parquet file.
///
/// [`Error::Hole`] is the one that matters: it is the named failure that stands where a
/// sparse file would otherwise hand a reader a page of zeros and let it answer.
#[derive(Debug, thiserror::Error)]
#[non_exhaustive]
pub enum Error {
    /// **A read touched bytes the mirror does not hold.** Never a zero page, never a
    /// silent wrong answer. `want` is the range the parquet reader asked for.
    #[error(
        "mirror hole: bytes {}..{} are not held (the mirror holds {held_chunks} of {total_chunks} chunks)",
        want_offset,
        want_offset + want_len
    )]
    Hole {
        /// First byte the reader asked for.
        want_offset: u64,
        /// How many bytes it asked for.
        want_len: u64,
        /// Chunks the mirror holds right now.
        held_chunks: u64,
        /// Chunks the source has in total.
        total_chunks: u64,
    },

    /// The mirror on disk was taken from a DIFFERENT version of the source. The whole
    /// file is invalid — a parquet footer names byte offsets, and a re-written source
    /// moves every one of them.
    #[error("mirror is stale: it holds source version {held:?}, the source is now {expected:?}")]
    StaleSource {
        /// The token the mirror was created with.
        held: String,
        /// The token the caller says the source carries now.
        expected: String,
    },

    /// The bytes handed to [`crate::Mirror::accept`] are not the length the footer says
    /// that chunk is. Accepting them would put the file permanently out of phase.
    #[error("chunk (row group {row_group}, column {column}) is {want} bytes, got {got}")]
    WrongLength {
        /// The row group the piece belongs to.
        row_group: usize,
        /// The leaf column index.
        column: usize,
        /// What the footer says.
        want: u64,
        /// What was offered.
        got: u64,
    },

    /// A [`crate::Need`] named a row group or column the source does not have.
    #[error("no such {what} {index} (the source has {count})")]
    OutOfRange {
        /// `"row group"` or `"column"`.
        what: &'static str,
        /// What was asked for.
        index: usize,
        /// What exists.
        count: usize,
    },

    /// The bytes are not a parquet file (or not one this far).
    #[error("not a parquet file: {why}")]
    NotParquet {
        /// What gave it away.
        why: String,
    },

    /// The sidecar state file is not one of ours, or is from a future format.
    #[error("mirror state file is unreadable: {why}")]
    BadState {
        /// What gave it away.
        why: String,
    },

    /// Underlying I/O.
    #[error("mirror i/o: {0}")]
    Io(#[from] std::io::Error),

    /// The footer would not parse.
    #[error("footer: {0}")]
    Parquet(#[from] parquet::errors::ParquetError),
}