rdblue commented on code in PR #18853:
URL: https://github.com/apache/iceberg/pull/18853#discussion_r4519801236
##########
core/src/main/java/org/apache/iceberg/V4Metadata.java:
##########
@@ -541,4 +548,730 @@ public Long pos() {
return null;
}
}
+
+ /** Base wrapper used to write a {@link TrackedFile} using the v4 tracked
file schema. */
+ private static void checkNullSequenceNumbers(TrackedFile file, Tracking
tracking) {
+ // null sequence numbers are inherited from the commit: ADDED entries
inherit both and
+ // MODIFIED entries inherit only the data sequence number
+ EntryStatus status = tracking.status();
+ Preconditions.checkState(
+ tracking.dataSequenceNumber() != null
+ || status == EntryStatus.ADDED
+ || status == EntryStatus.MODIFIED,
+ "Cannot write null data sequence number for status %s (must be ADDED
or MODIFIED): %s",
+ status,
+ file.location());
+ Preconditions.checkState(
+ tracking.fileSequenceNumber() != null || status == EntryStatus.ADDED,
+ "Cannot write null file sequence number for status %s (must be ADDED):
%s",
+ status,
+ file.location());
+ }
+
+ private abstract static class TrackedFileWrapper implements TrackedFile,
StructLike {
+ private static final int SIZE =
+ TrackedFile.schema(Types.StructType.of(),
Types.StructType.of()).columns().size();
+
+ private final Map<Integer, PartitionSpec> specsById;
+ private final Map<Integer, StructProjection> partitionProjections;
+ private final ContentStatsWrapper contentStats;
+
+ private TrackedFile wrapped = null;
+ private Integer specId = null;
+ private TrackingWrapper tracking = null;
+ private ManifestInfoWrapper manifestInfo = null;
+
+ private TrackedFileWrapper(
+ Map<Integer, PartitionSpec> specsById, Types.StructType
contentStatsType) {
+ this.specsById = specsById;
+ this.partitionProjections = partitionProjections(specsById);
+ this.contentStats = new ContentStatsWrapper(contentStatsType);
+ }
+
+ protected void wrapInternal(
+ TrackedFile file, TrackingWrapper newTracking, ManifestInfoWrapper
newManifestInfo) {
+ Integer fileSpecId = file.specId();
+ PartitionSpec spec = fileSpecId != null ? specsById.get(fileSpecId) :
null;
+ Preconditions.checkArgument(
+ fileSpecId == null || spec != null,
+ "Cannot find partition spec for spec ID: %s",
+ fileSpecId);
+
+ this.wrapped = file;
+ // v4 metadata stores a null spec ID and partition for unpartitioned
files
+ this.specId = spec != null && !spec.fields().isEmpty() ? fileSpecId :
null;
+ this.tracking = newTracking;
+ this.manifestInfo = newManifestInfo;
+ }
+
+ @Override
+ public int size() {
+ return SIZE;
+ }
+
+ @Override
+ public <T> void set(int pos, T value) {
+ throw new UnsupportedOperationException("Cannot modify a read-only
tracked file");
+ }
+
+ @Override
+ public <T> T get(int pos, Class<T> javaClass) {
+ return javaClass.cast(get(pos));
+ }
+
+ private Object get(int pos) {
+ return switch (pos) {
+ case 0 -> tracking;
+ case 1 -> wrapped.contentType().id();
+ case 2 -> wrapped.location();
+ case 3 -> wrapped.fileFormat().toString();
+ case 4 -> wrapped.recordCount();
+ case 5 -> wrapped.fileSizeInBytes();
+ case 6 -> specId;
+ case 7 -> unionPartition();
+ case 8 -> wrapped.contentStats() != null ?
contentStats.wrap(wrapped.contentStats()) : null;
+ case 9 -> wrapped.sortOrderId();
+ case 10 -> wrapped.deletionVector();
+ case 11 -> manifestInfo;
+ case 12 -> wrapped.keyMetadata();
+ case 13 -> wrapped.splitOffsets();
+ case 14 -> wrapped.columnFiles();
+ default -> throw new UnsupportedOperationException("Unknown field
ordinal: " + pos);
+ };
+ }
+
+ @Override
+ public Tracking tracking() {
+ return tracking;
+ }
+
+ @Override
+ public FileContent contentType() {
+ return wrapped.contentType();
+ }
+
+ @Override
+ public String location() {
+ return wrapped.location();
+ }
+
+ @Override
+ public FileFormat fileFormat() {
+ return wrapped.fileFormat();
+ }
+
+ @Override
+ public long recordCount() {
+ return wrapped.recordCount();
+ }
+
+ @Override
+ public long fileSizeInBytes() {
+ return wrapped.fileSizeInBytes();
+ }
+
+ @Override
+ public Integer specId() {
+ return specId;
+ }
+
+ @Override
+ public StructLike partition() {
+ return specId != null ? wrapped.partition() : null;
+ }
+
+ @Override
+ public ContentStats contentStats() {
+ return wrapped.contentStats();
+ }
+
+ @Override
+ public Integer sortOrderId() {
+ return wrapped.sortOrderId();
+ }
+
+ @Override
+ public DeletionVector deletionVector() {
+ return wrapped.deletionVector();
+ }
+
+ @Override
+ public ManifestInfo manifestInfo() {
+ return manifestInfo;
+ }
+
+ @Override
+ public ByteBuffer keyMetadata() {
+ return wrapped.keyMetadata();
+ }
+
+ @Override
+ public List<Long> splitOffsets() {
+ return wrapped.splitOffsets();
+ }
+
+ @Override
+ public List<ColumnFile> columnFiles() {
+ return wrapped.columnFiles();
+ }
+
+ @Override
+ public TrackedFile copy() {
+ return copy(null /* all stats */);
+ }
+
+ @Override
+ public TrackedFile copyWithStats(Set<Integer> requestedColumnIds) {
+ return copy(requestedColumnIds);
+ }
+
+ private TrackedFile copy(Set<Integer> statsIds) {
+ ContentStats stats = wrapped.contentStats();
+ ContentStats statsCopy;
+ if (stats == null || (statsIds != null && statsIds.isEmpty())) {
+ statsCopy = null;
+ } else {
+ statsCopy = statsIds != null ? stats.copy(statsIds) : stats.copy();
+ }
+
+ DeletionVector dv = wrapped.deletionVector();
+ return new TrackedFileStruct(
+ tracking.copy(),
+ wrapped.contentType(),
+ wrapped.location(),
+ wrapped.fileFormat(),
+ wrapped.recordCount(),
+ wrapped.fileSizeInBytes(),
+ specId,
+ StructLikeUtil.copy(partition()),
+ statsCopy,
+ wrapped.sortOrderId(),
+ dv != null ? dv.copy() : null,
+ manifestInfo != null ? manifestInfo.copy() : null,
+ wrapped.keyMetadata(),
+ wrapped.splitOffsets(),
+ copyColumnFiles(wrapped.columnFiles()));
+ }
+
+ private static List<ColumnFile> copyColumnFiles(List<ColumnFile>
columnFiles) {
+ if (columnFiles == null) {
+ return null;
+ }
+
+ List<ColumnFile> copies =
Lists.newArrayListWithCapacity(columnFiles.size());
+ for (ColumnFile columnFile : columnFiles) {
+ copies.add(columnFile.copy());
+ }
+
+ return copies;
+ }
+
+ private StructLike unionPartition() {
+ StructLike partition = partition();
+ if (partition == null) {
+ return null;
+ }
+
+ StructProjection projection = partitionProjections.get(specId);
+ return projection != null ? projection.wrap(partition) : partition;
+ }
+
+ private static Map<Integer, StructProjection> partitionProjections(
+ Map<Integer, PartitionSpec> specsById) {
+ Types.StructType unionType =
Partitioning.unionPartitionTypes(specsById.values());
+ Map<Integer, StructProjection> projections = Maps.newHashMap();
+ for (PartitionSpec spec : specsById.values()) {
+ Types.StructType partitionType = spec.partitionType();
+ StructProjection projection =
+ partitionType.equals(unionType)
+ ? null
+ : StructProjection.createAllowMissing(partitionType,
unionType);
+ projections.put(spec.specId(), projection);
+ }
+
+ return projections;
+ }
+ }
+
+ /** Wrapper used to write a data file to a v4 leaf manifest. */
+ static class LeafTrackedFileWrapper extends TrackedFileWrapper {
+ private final TrackingWrapper trackingWrapper = new TrackingWrapper();
+
+ LeafTrackedFileWrapper(
+ Map<Integer, PartitionSpec> specsById, Types.StructType
contentStatsType) {
+ super(specsById, contentStatsType);
+ }
+
+ LeafTrackedFileWrapper wrap(TrackedFile file, Tracking newTracking) {
+ Preconditions.checkArgument(file != null, "Invalid file: null");
+ Preconditions.checkArgument(newTracking != null, "Invalid tracking:
null");
+ Preconditions.checkArgument(
+ file.contentType() == FileContent.DATA,
+ "Invalid content type for leaf manifest: %s",
+ file.contentType());
+
+ checkNullSequenceNumbers(file, newTracking);
+
+ trackingWrapper.wrap(
+ newTracking,
+ newTracking.snapshotId(),
+ newTracking.dataSequenceNumber(),
+ newTracking.fileSequenceNumber(),
+ newTracking.firstRowId());
+ wrapInternal(file, trackingWrapper, null /* manifestInfo */);
+ return this;
+ }
+ }
+
+ /** Wrapper used to write a data file or leaf manifest to a v4 root
manifest. */
+ static class RootTrackedFileWrapper extends TrackedFileWrapper {
+ private final long commitSnapshotId;
+ private final long commitSequenceNumber;
+ private final TrackingWrapper trackingWrapper = new TrackingWrapper();
+ private final ManifestInfoWrapper manifestInfoWrapper = new
ManifestInfoWrapper();
+
+ RootTrackedFileWrapper(
+ long commitSnapshotId,
+ long commitSequenceNumber,
+ Map<Integer, PartitionSpec> specsById,
+ Types.StructType contentStatsType) {
+ super(specsById, contentStatsType);
+ this.commitSnapshotId = commitSnapshotId;
+ this.commitSequenceNumber = commitSequenceNumber;
+ }
+
+ /**
+ * Re-points this wrapper at a file and the tracking to write for it.
+ *
+ * @param file a data file or manifest
+ * @param newTracking the tracking to write for the file
+ * @param assignedFirstRowId a first row ID to assign to the file, or null
if it is already
+ * assigned
+ * @return this for method chaining
+ */
+ RootTrackedFileWrapper wrap(TrackedFile file, Tracking newTracking, Long
assignedFirstRowId) {
+ Preconditions.checkArgument(file != null, "Invalid file: null");
+ Preconditions.checkArgument(newTracking != null, "Invalid tracking:
null");
+
+ boolean isManifest = file.contentType() != FileContent.DATA;
+ Preconditions.checkArgument(
+ !isManifest || file.manifestInfo() != null,
+ "Missing manifest info for manifest: %s",
+ file.location());
+ Preconditions.checkArgument(
+ newTracking.snapshotId() != null, "Missing snapshot ID for file:
%s", file.location());
+ checkNullSequenceNumbers(file, newTracking);
+
+ long snapshotId = newTracking.snapshotId();
+ // a modified entry inherits the data sequence number of the commit that
modified it
+ Long dataSequenceSnapshotId =
+ newTracking.status() == EntryStatus.MODIFIED
+ ? newTracking.modifiedSnapshotId()
+ : newTracking.snapshotId();
+ trackingWrapper.wrap(
+ newTracking,
+ snapshotId,
+ resolveSequenceNumber(file, newTracking.dataSequenceNumber(),
dataSequenceSnapshotId),
+ resolveSequenceNumber(file, newTracking.fileSequenceNumber(),
snapshotId),
+ resolveFirstRowId(file, newTracking, assignedFirstRowId));
+
+ if (isManifest) {
+ ManifestInfo info = file.manifestInfo();
+ manifestInfoWrapper.wrap(
+ info, resolveSequenceNumber(file, info.minSequenceNumber(),
snapshotId));
+ wrapInternal(file, trackingWrapper, manifestInfoWrapper);
+ } else {
+ wrapInternal(file, trackingWrapper, null /* manifestInfo */);
+ }
+
+ return this;
+ }
+
+ private long resolveSequenceNumber(
+ TrackedFile file, Long existingSequenceNumber, Long entrySnapshotId) {
+ if (existingSequenceNumber == null
+ || existingSequenceNumber == ManifestWriter.UNASSIGNED_SEQ) {
+ // the sequence number is assigned here, so the file must be changed
by the current commit
+ Preconditions.checkState(
+ entrySnapshotId != null && entrySnapshotId == commitSnapshotId,
+ "Cannot assign sequence number to file from snapshot %s
(committing snapshot %s): %s",
+ entrySnapshotId,
+ commitSnapshotId,
+ file.location());
+ return commitSequenceNumber;
+ }
+
+ return existingSequenceNumber;
+ }
+
+ private static Long resolveFirstRowId(
Review Comment:
I would prefer to keep this the way it is in `ManifestFileWrapper`. At least
the structure shouldn't need to change, right?
I think the purpose of using `resolveFirstRowId` outside of the wrapper and
passing all of the values into the wrapper is to allow different behavior
between the leaf and root manifests. Is that correct?
That ends up making the `wrap` method do all of the work, which I found
confusing in some cases. `trackingWrapper.wrap` for leaf files passes all of
the tracking fields as their own overrides, for example. Both `wrap` calls pass
`newTracking.snapshotId()` through directly. And the `resolveSequenceNumber`
method is passed a snapshot ID to verify even though it doesn't have to do with
sequence number resolution. And this method is more about validating the first
row ID of `newTracking`, not resolving it.
It would be more readable not to mix together responsibilities for these
methods. And I also think it would be better to use the same logic or structure
as the v3 wrappers. Configure the wrapper for what is being written, then rely
on it.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
