Package org.elasticsearch.xpack.esql.datasources.spi

Interface FileList

public interface FileList
Indexed view over a resolved set of files from an external data source. Implementations are package-private within the datasources.glob package; consumers should depend only on this interface.

Two sentinel instances encode distinct states:

  • UNRESOLVED — glob not yet expanded (isResolved() == false, isEmpty() == false)
  • EMPTY — glob expanded but matched zero files (isResolved() == true, isEmpty() == true)

  • Field Details

    • UNRESOLVED

      static final FileList UNRESOLVED
      Sentinel: single-file path, glob not yet applied (isResolved() == false).

    • EMPTY

      static final FileList EMPTY
      Sentinel: glob expanded but matched zero files (isResolved() == true, isEmpty() == true).

  • Method Details

    • fileCount

      int fileCount()

    • path

      StoragePath path(int i)

    • size

      long size(int i)

    • lastModifiedMillis

      long lastModifiedMillis(int i)

    • originalPattern

      @Nullable String originalPattern()

    • partitionMetadata

      @Nullable PartitionMetadata partitionMetadata()

    • isResolved

      boolean isResolved()

    • isEmpty

      boolean isEmpty()

    • estimatedBytes

      long estimatedBytes()

    • fileSetFingerprint

      @Nullable default FileSetFingerprint fileSetFingerprint()
      The 128-bit fingerprint identifying the resolved file SET: a commutative fold over every file's (path, mtime, size) plus the file count, computed once when the listing is built. The same set listed in any order yields the same fingerprint; any file added, removed, or modified (mtime or size) yields a different one. This makes the fingerprint a content-addressed cache key for dataset-level derived state (e.g. the warm COUNT(*) aggregate): keys derived from it are correct-or-miss by construction, with no separate invalidation protocol — and they survive listing refreshes (the 30s listing TTL) as long as the underlying files are unchanged, because the fingerprint derives from listing CONTENT, not listing object identity.

      null for the sentinels and for implementations that do not compute one.