LuciferYang commented on code in PR #10845:
URL: https://github.com/apache/paimon/pull/10845#discussion_r4400356172
##########
paimon-core/src/main/java/org/apache/paimon/utils/FileSystemBranchManager.java:
##########
@@ -224,6 +240,228 @@ public void fastForward(String branchName) {
}
}
+ /**
+ * Fast-forwards by appending when the current branch is a prefix of the
branch: its snapshots
+ * from the branch's earliest one (or its own earliest, if later) to its
latest are the
+ * branch's, and so are its schemas. Nothing is deleted and files are
written in ascending id
+ * order, so a failure leaves the current branch at one of the branch's
snapshots and a retry
+ * continues from there.
+ *
+ * @return false if the current branch is not a prefix of the branch
+ */
+ private boolean tryAppendFromBranch(String branchName, long
branchEarliestId, Lock lock)
+ throws IOException {
+ SnapshotManager branchSnapshotManager =
snapshotManager.copyWithBranch(branchName);
+ SchemaManager branchSchemaManager =
schemaManager.copyWithBranch(branchName);
+ // List rather than trust the LATEST hint: an interrupted replace by
an older version copies
+ // the branch's hint before the snapshots it names.
+ Map<Long, Long> sizes = snapshotSizes(snapshotManager);
+ Map<Long, Long> branchSizes = snapshotSizes(branchSnapshotManager);
+ if (sizes.isEmpty() || branchSizes.isEmpty()) {
+ return false;
+ }
+ long latestId = Collections.max(sizes.keySet());
+ long branchLatestId = Collections.max(branchSizes.keySet());
+ if (latestId < branchEarliestId
+ || latestId > branchLatestId
+ || !sameSizes(sizes, branchSizes, branchEarliestId, latestId)
+ || !sameContent(
+ snapshotManager.snapshotPath(latestId),
+ branchSnapshotManager.snapshotPath(latestId),
+ Snapshot::fromJson)
+ || !sameSchemas(branchSchemaManager)) {
+ return false;
+ }
+
+ long latestSchemaId = Collections.max(schemaManager.listAllIds());
+ for (long schemaId : sorted(branchSchemaManager.listAllIds())) {
+ if (schemaId > latestSchemaId) {
+ copyAtomically(
+ lock,
+ branchSchemaManager.toSchemaPath(schemaId),
+ schemaManager.toSchemaPath(schemaId));
+ }
+ }
+ copySnapshots(branchSnapshotManager, latestId + 1, branchLatestId,
lock);
+ copyBranchTags(branchName, lock);
+ snapshotManager.commitLatestHint(branchLatestId);
+ return true;
+ }
+
+ private Map<Long, Long> snapshotSizes(SnapshotManager manager) throws
IOException {
+ Map<Long, Long> sizes = new HashMap<>();
+ for (FileStatus status :
+ listVersionedFileStatus(fileIO, manager.snapshotDirectory(),
SNAPSHOT_PREFIX)
+ .collect(Collectors.toList())) {
+ String version =
status.getPath().getName().substring(SNAPSHOT_PREFIX.length());
+ try {
+ sizes.put(Long.parseLong(version), status.getLen());
+ } catch (NumberFormatException ignored) {
+ // not a snapshot file, for example a temporary file of a
Python writer
+ }
+ }
+ return sizes;
+ }
+
+ /**
+ * Whether the current branch has every snapshot from {@code from} (or its
earliest, if later)
+ * to {@code to}, each of the branch's size. A copy cut short by an older
version's non-atomic
+ * replace has another size.
+ */
+ private static boolean sameSizes(
+ Map<Long, Long> sizes, Map<Long, Long> branchSizes, long from,
long to) {
+ for (long id = Math.max(from, Collections.min(sizes.keySet())); id <=
to; id++) {
+ Long size = sizes.get(id);
+ if (size == null || !size.equals(branchSizes.get(id))) {
+ return false;
+ }
+ }
+ return true;
+ }
+
+ /** Whether the current branch's schemas are the branch's schemas up to
the current latest. */
+ private boolean sameSchemas(SchemaManager branchSchemaManager) throws
IOException {
+ Set<Long> ids = new HashSet<>(schemaManager.listAllIds());
+ if (ids.isEmpty()) {
+ return false;
+ }
+ long latestId = Collections.max(ids);
+ Set<Long> branchIds =
+ branchSchemaManager.listAllIds().stream()
+ .filter(id -> id <= latestId)
+ .collect(Collectors.toSet());
+ if (!ids.equals(branchIds)) {
+ return false;
+ }
+ for (long id : ids) {
+ if (!sameContent(
+ schemaManager.toSchemaPath(id),
+ branchSchemaManager.toSchemaPath(id),
+ TableSchema::fromJson)) {
+ return false;
+ }
+ }
+ return true;
+ }
+
+ /**
+ * Replaces the current branch's snapshots and tags from the branch's
earliest snapshot onwards,
+ * and its schemas, with the branch's. Schemas are replaced file by file
instead of deleted up
+ * front, so a failure never leaves the remaining snapshots without their
schemas.
+ */
+ private void replaceFromBranch(String branchName, long earliestSnapshotId,
Lock lock)
+ throws IOException {
+ SnapshotManager branchSnapshotManager =
snapshotManager.copyWithBranch(branchName);
+ SchemaManager branchSchemaManager =
schemaManager.copyWithBranch(branchName);
+ long earliestSchemaId =
branchSnapshotManager.snapshot(earliestSnapshotId).schemaId();
+
+ List<Path> deletePaths =
+ new ArrayList<>(snapshotManager.snapshotPaths(id -> id >=
earliestSnapshotId));
+ deletePaths.addAll(
+ tagManager.tagPaths(path -> Tag.fromPath(fileIO, path).id() >=
earliestSnapshotId));
+ lock.ensureValid();
+ snapshotManager.deleteLatestHint();
+ fileIO.deleteFilesQuietly(deletePaths);
+ // Main expired its own snapshots up to its EARLIEST hint; the
branch's snapshots below
+ // that hint become main's. Move the hint before copying them, so that
a retry after a
+ // failed copy, which appends the rest, finds it already moved.
+ Long earliestHint =
+ HintFileUtils.readHint(
+ fileIO, HintFileUtils.EARLIEST,
snapshotManager.snapshotDirectory());
+ if (earliestHint != null && earliestHint > earliestSnapshotId) {
+ snapshotManager.commitEarliestHint(earliestSnapshotId);
+ }
+
+ List<Long> branchSchemaIds = sorted(branchSchemaManager.listAllIds());
+ for (long id : branchSchemaIds) {
+ replaceAtomically(
+ lock, branchSchemaManager.toSchemaPath(id),
schemaManager.toSchemaPath(id));
+ }
+ for (long id : schemaManager.listAllIds()) {
+ if (id >= earliestSchemaId && !branchSchemaIds.contains(id)) {
+ lock.ensureValid();
+ fileIO.deleteQuietly(schemaManager.toSchemaPath(id));
+ }
+ }
+
+ // List rather than trust the branch's LATEST hint, as the append path
does.
+ long branchLatestId =
Collections.max(snapshotSizes(branchSnapshotManager).keySet());
+ copySnapshots(branchSnapshotManager, earliestSnapshotId,
branchLatestId, lock);
+ copyBranchTags(branchName, lock);
+ snapshotManager.commitLatestHint(branchLatestId);
+ }
+
+ private void copySnapshots(
+ SnapshotManager branchSnapshotManager, long fromId, long toId,
Lock lock)
+ throws IOException {
+ for (long id = fromId; id <= toId; id++) {
+ copyAtomically(
+ lock, branchSnapshotManager.snapshotPath(id),
snapshotManager.snapshotPath(id));
+ }
+ }
+
+ private void copyBranchTags(String branchName, Lock lock) throws
IOException {
+ // Continue fast-forward even without tags.
+ Path branchTagDirectory =
tagManager.copyWithBranch(branchName).tagDirectory();
+ if (!fileIO.exists(branchTagDirectory)) {
+ return;
+ }
+ for (FileStatus status : fileIO.listStatus(branchTagDirectory)) {
+ if (!status.isDir()) {
+ Path tag = status.getPath();
+ replaceAtomically(lock, tag, new
Path(tagManager.tagDirectory(), tag.getName()));
+ }
+ }
+ }
+
+ /** Copies a file without exposing partial content, accepting a target
already copied. */
+ private void copyAtomically(Lock lock, Path source, Path target) throws
IOException {
+ // A lease may expire while many files are copied; stop before writing
past it.
+ lock.ensureValid();
+ String content = readExactly(source);
+ if ((fileIO.exists(target) || !fileIO.tryToWriteAtomic(target,
content))
Review Comment:
Fixed in 02d7d676dc. `copyAtomically` now checks the lease after reading the
source, right before the write. Replacing main deletes one file at a time with
a check before each delete, and the LATEST and EARLIEST hint writes are checked
too. `testFastForwardRechecksLeaseAfterReadingSource` expires the lease while
the branch's snapshot 2 is opened and asserts main gets no snapshot 2;
`testReplaceStopsDeletingWhenLeaseExpires` expires it after the first delete
and asserts nothing else is deleted.
##########
paimon-core/src/main/java/org/apache/paimon/utils/FileSystemBranchManager.java:
##########
@@ -224,6 +240,228 @@ public void fastForward(String branchName) {
}
}
+ /**
+ * Fast-forwards by appending when the current branch is a prefix of the
branch: its snapshots
+ * from the branch's earliest one (or its own earliest, if later) to its
latest are the
+ * branch's, and so are its schemas. Nothing is deleted and files are
written in ascending id
+ * order, so a failure leaves the current branch at one of the branch's
snapshots and a retry
+ * continues from there.
+ *
+ * @return false if the current branch is not a prefix of the branch
+ */
+ private boolean tryAppendFromBranch(String branchName, long
branchEarliestId, Lock lock)
+ throws IOException {
+ SnapshotManager branchSnapshotManager =
snapshotManager.copyWithBranch(branchName);
+ SchemaManager branchSchemaManager =
schemaManager.copyWithBranch(branchName);
+ // List rather than trust the LATEST hint: an interrupted replace by
an older version copies
+ // the branch's hint before the snapshots it names.
+ Map<Long, Long> sizes = snapshotSizes(snapshotManager);
+ Map<Long, Long> branchSizes = snapshotSizes(branchSnapshotManager);
+ if (sizes.isEmpty() || branchSizes.isEmpty()) {
+ return false;
+ }
+ long latestId = Collections.max(sizes.keySet());
+ long branchLatestId = Collections.max(branchSizes.keySet());
+ if (latestId < branchEarliestId
+ || latestId > branchLatestId
+ || !sameSizes(sizes, branchSizes, branchEarliestId, latestId)
+ || !sameContent(
+ snapshotManager.snapshotPath(latestId),
+ branchSnapshotManager.snapshotPath(latestId),
+ Snapshot::fromJson)
+ || !sameSchemas(branchSchemaManager)) {
+ return false;
+ }
+
+ long latestSchemaId = Collections.max(schemaManager.listAllIds());
+ for (long schemaId : sorted(branchSchemaManager.listAllIds())) {
+ if (schemaId > latestSchemaId) {
+ copyAtomically(
+ lock,
+ branchSchemaManager.toSchemaPath(schemaId),
+ schemaManager.toSchemaPath(schemaId));
+ }
+ }
+ copySnapshots(branchSnapshotManager, latestId + 1, branchLatestId,
lock);
+ copyBranchTags(branchName, lock);
+ snapshotManager.commitLatestHint(branchLatestId);
+ return true;
+ }
+
+ private Map<Long, Long> snapshotSizes(SnapshotManager manager) throws
IOException {
+ Map<Long, Long> sizes = new HashMap<>();
+ for (FileStatus status :
+ listVersionedFileStatus(fileIO, manager.snapshotDirectory(),
SNAPSHOT_PREFIX)
+ .collect(Collectors.toList())) {
+ String version =
status.getPath().getName().substring(SNAPSHOT_PREFIX.length());
+ try {
+ sizes.put(Long.parseLong(version), status.getLen());
+ } catch (NumberFormatException ignored) {
+ // not a snapshot file, for example a temporary file of a
Python writer
+ }
+ }
+ return sizes;
+ }
+
+ /**
+ * Whether the current branch has every snapshot from {@code from} (or its
earliest, if later)
+ * to {@code to}, each of the branch's size. A copy cut short by an older
version's non-atomic
+ * replace has another size.
+ */
+ private static boolean sameSizes(
+ Map<Long, Long> sizes, Map<Long, Long> branchSizes, long from,
long to) {
+ for (long id = Math.max(from, Collections.min(sizes.keySet())); id <=
to; id++) {
+ Long size = sizes.get(id);
+ if (size == null || !size.equals(branchSizes.get(id))) {
Review Comment:
Fixed in 02d7d676dc. The prefix check now compares the content of every main
snapshot from the branch's earliest (or main's earliest, if later) to main's
latest with the branch's, bytes first and parsed snapshots if the bytes differ.
`testFastForwardReplacesDivergentSnapshotOfSameLength` builds the state you
describe: main's own snapshot 3, the same length as the branch's, next to the
branch's snapshot 4. `fast_forward` now replaces main, and snapshot 3 ends up
with the branch's bytes.
Replacing main also no longer deletes quietly. A file that is still there
after the delete fails the fast-forward
(`testReplaceFailsWhenSnapshotCannotBeDeleted`), so this version does not leave
that state behind itself.
--
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]
