Fokko commented on code in PR #3320:
URL: https://github.com/apache/iceberg-python/pull/3320#discussion_r3600324981
##########
pyiceberg/table/update/snapshot.py:
##########
@@ -353,11 +397,116 @@ def new_manifest_output(self) -> OutputFile:
location_provider = self._transaction._table.location_provider()
file_name =
_new_manifest_file_name(num=next(self._manifest_num_counter),
commit_uuid=self.commit_uuid)
file_path = location_provider.new_metadata_location(file_name)
+ self._written_manifests.append(file_path)
return self._io.new_output(file_path)
def fetch_manifest_entry(self, manifest: ManifestFile, discard_deleted:
bool = True) -> list[ManifestEntry]:
return manifest.fetch_manifest_entry(io=self._io,
discard_deleted=discard_deleted)
+ def commit(self) -> None:
+ self._transaction._register_snapshot_producer(self)
+ super().commit()
+
+ def _cleanup_uncommitted(self) -> None:
+ """Delete manifest files and manifest lists from failed retry
attempts."""
+ for path in self._uncommitted_manifests:
+ try:
+ self._io.delete(path)
+ except Exception:
+ logger.warning("Failed to delete uncommitted manifest: %s",
path, exc_info=True)
+ self._uncommitted_manifests.clear()
+ # Delete all manifest lists except the last one (which the committed
snapshot references)
+ if len(self._written_manifest_lists) > 1:
+ for path in self._written_manifest_lists[:-1]:
+ try:
+ self._io.delete(path)
+ except Exception:
+ logger.warning("Failed to delete uncommitted manifest
list: %s", path, exc_info=True)
+ self._written_manifest_lists = self._written_manifest_lists[-1:]
+
+ def _clean_all_uncommitted(self) -> None:
+ """Clean up all manifests and manifest lists on abort."""
+ for path in itertools.chain(self._uncommitted_manifests,
self._written_manifests):
+ try:
+ self._io.delete(path)
+ except Exception:
+ logger.warning("Failed to delete uncommitted manifest: %s",
path, exc_info=True)
+ for path in self._written_manifest_lists:
+ try:
+ self._io.delete(path)
+ except Exception:
+ logger.warning("Failed to delete uncommitted manifest list:
%s", path, exc_info=True)
+ self._uncommitted_manifests.clear()
+ self._written_manifests.clear()
+ self._written_manifest_lists.clear()
+
+ def _refresh_for_retry(self) -> None:
+ """Reset state for a retry attempt with refreshed metadata."""
+ self._uncommitted_manifests.extend(self._written_manifests)
+ self._written_manifests.clear()
+ self._parent_snapshot_id = (
+ snapshot.snapshot_id if (snapshot :=
self._transaction.table_metadata.snapshot_by_name(self._target_branch)) else
None
+ )
Review Comment:
nit, should we make a helper for getting the latest head of a branch? This
is not the easiest to read, and I believe we use this more than once
##########
pyiceberg/table/update/snapshot.py:
##########
@@ -91,19 +97,50 @@ def _new_manifest_list_file_name(snapshot_id: int, attempt:
int, commit_uuid: uu
return f"snap-{snapshot_id}-{attempt}-{commit_uuid}.avro"
+@dataclass(frozen=True)
+class CommitWindow:
+ """Tracks the commit range to validate against during retry.
+
+ base: The snapshot when the operation began (exclusive lower bound, fixed
across retries).
+ head: The branch HEAD snapshot after metadata refresh (inclusive upper
bound).
+ """
+
+ base: Snapshot | None
+ head: Snapshot | None
+
+ @classmethod
+ def resolve(cls, metadata: TableMetadata, base_id: int | None, branch: str
| None) -> CommitWindow:
+ """Resolve a CommitWindow from metadata, starting snapshot ID, and
target branch."""
+ head = metadata.snapshot_by_name(branch)
+ base = metadata.snapshot_by_id(base_id) if base_id is not None else
None
+ if base_id is not None and base is None:
+ raise ValidationException(f"Cannot find starting snapshot
{base_id}")
Review Comment:
This is dead code because `snapshot_by_id` will throw a `StopIteration`
before it gets here.
https://github.com/apache/iceberg-python/blob/48e710d20ceeeaa637d5aeae7746b787410859f8/pyiceberg/table/metadata.py#L239-L241
##########
pyiceberg/table/update/snapshot.py:
##########
@@ -353,11 +397,116 @@ def new_manifest_output(self) -> OutputFile:
location_provider = self._transaction._table.location_provider()
file_name =
_new_manifest_file_name(num=next(self._manifest_num_counter),
commit_uuid=self.commit_uuid)
file_path = location_provider.new_metadata_location(file_name)
+ self._written_manifests.append(file_path)
return self._io.new_output(file_path)
def fetch_manifest_entry(self, manifest: ManifestFile, discard_deleted:
bool = True) -> list[ManifestEntry]:
return manifest.fetch_manifest_entry(io=self._io,
discard_deleted=discard_deleted)
+ def commit(self) -> None:
+ self._transaction._register_snapshot_producer(self)
+ super().commit()
+
+ def _cleanup_uncommitted(self) -> None:
+ """Delete manifest files and manifest lists from failed retry
attempts."""
+ for path in self._uncommitted_manifests:
+ try:
+ self._io.delete(path)
+ except Exception:
+ logger.warning("Failed to delete uncommitted manifest: %s",
path, exc_info=True)
+ self._uncommitted_manifests.clear()
+ # Delete all manifest lists except the last one (which the committed
snapshot references)
+ if len(self._written_manifest_lists) > 1:
+ for path in self._written_manifest_lists[:-1]:
+ try:
+ self._io.delete(path)
+ except Exception:
+ logger.warning("Failed to delete uncommitted manifest
list: %s", path, exc_info=True)
+ self._written_manifest_lists = self._written_manifest_lists[-1:]
+
+ def _clean_all_uncommitted(self) -> None:
+ """Clean up all manifests and manifest lists on abort."""
+ for path in itertools.chain(self._uncommitted_manifests,
self._written_manifests):
+ try:
+ self._io.delete(path)
+ except Exception:
+ logger.warning("Failed to delete uncommitted manifest: %s",
path, exc_info=True)
+ for path in self._written_manifest_lists:
+ try:
+ self._io.delete(path)
+ except Exception:
+ logger.warning("Failed to delete uncommitted manifest list:
%s", path, exc_info=True)
+ self._uncommitted_manifests.clear()
+ self._written_manifests.clear()
+ self._written_manifest_lists.clear()
+
+ def _refresh_for_retry(self) -> None:
+ """Reset state for a retry attempt with refreshed metadata."""
+ self._uncommitted_manifests.extend(self._written_manifests)
+ self._written_manifests.clear()
+ self._parent_snapshot_id = (
+ snapshot.snapshot_id if (snapshot :=
self._transaction.table_metadata.snapshot_by_name(self._target_branch)) else
None
+ )
+ self._snapshot_id = self._transaction.table_metadata.new_snapshot_id()
+ self._manifest_num_counter = itertools.count(0)
+ self.commit_uuid = uuid.uuid4()
+
+ def _validate_concurrency(self) -> None:
+ """Validate that concurrent changes do not conflict with this
operation.
+
+ Uses the CommitWindow to determine which catalog commits to validate
against.
+ The window spans from base (when the operation began) to head (latest
branch
+ HEAD), covering all external concurrent commits.
+
+ Subclasses that do not require validation (e.g. fast append) should
override
+ with a no-op.
+ """
+ from pyiceberg.table import TableProperties
+ from pyiceberg.table.snapshots import IsolationLevel
+ from pyiceberg.table.update.validate import (
+ _validate_added_data_files,
+ _validate_deleted_data_files,
+ _validate_no_new_delete_files,
+ _validate_no_new_deletes_for_data_files,
+ )
+
+ if self._commit_window is None or self._commit_window.is_empty():
+ return
+
+ catalog_head = self._commit_window.head
+ starting_snapshot = self._commit_window.base
+
+ if catalog_head is None or starting_snapshot is None:
+ return
+
+ table = self._transaction._table
+ isolation_level_str = table.metadata.properties.get(
+ self._isolation_level_property,
TableProperties.WRITE_ISOLATION_LEVEL_DEFAULT
+ )
+ isolation_level = IsolationLevel(isolation_level_str)
Review Comment:
How about moving this to a property on `TableMetadata`? So we can do:
```suggestion
isolation_level = table.metadata.isolation_level(Operation)
```
##########
pyiceberg/table/update/snapshot.py:
##########
@@ -353,11 +397,116 @@ def new_manifest_output(self) -> OutputFile:
location_provider = self._transaction._table.location_provider()
file_name =
_new_manifest_file_name(num=next(self._manifest_num_counter),
commit_uuid=self.commit_uuid)
file_path = location_provider.new_metadata_location(file_name)
+ self._written_manifests.append(file_path)
return self._io.new_output(file_path)
def fetch_manifest_entry(self, manifest: ManifestFile, discard_deleted:
bool = True) -> list[ManifestEntry]:
return manifest.fetch_manifest_entry(io=self._io,
discard_deleted=discard_deleted)
+ def commit(self) -> None:
+ self._transaction._register_snapshot_producer(self)
+ super().commit()
+
+ def _cleanup_uncommitted(self) -> None:
+ """Delete manifest files and manifest lists from failed retry
attempts."""
+ for path in self._uncommitted_manifests:
+ try:
+ self._io.delete(path)
+ except Exception:
+ logger.warning("Failed to delete uncommitted manifest: %s",
path, exc_info=True)
+ self._uncommitted_manifests.clear()
+ # Delete all manifest lists except the last one (which the committed
snapshot references)
+ if len(self._written_manifest_lists) > 1:
+ for path in self._written_manifest_lists[:-1]:
+ try:
+ self._io.delete(path)
+ except Exception:
+ logger.warning("Failed to delete uncommitted manifest
list: %s", path, exc_info=True)
+ self._written_manifest_lists = self._written_manifest_lists[-1:]
+
+ def _clean_all_uncommitted(self) -> None:
+ """Clean up all manifests and manifest lists on abort."""
+ for path in itertools.chain(self._uncommitted_manifests,
self._written_manifests):
+ try:
+ self._io.delete(path)
+ except Exception:
+ logger.warning("Failed to delete uncommitted manifest: %s",
path, exc_info=True)
+ for path in self._written_manifest_lists:
+ try:
+ self._io.delete(path)
+ except Exception:
+ logger.warning("Failed to delete uncommitted manifest list:
%s", path, exc_info=True)
+ self._uncommitted_manifests.clear()
+ self._written_manifests.clear()
+ self._written_manifest_lists.clear()
+
+ def _refresh_for_retry(self) -> None:
+ """Reset state for a retry attempt with refreshed metadata."""
+ self._uncommitted_manifests.extend(self._written_manifests)
+ self._written_manifests.clear()
+ self._parent_snapshot_id = (
+ snapshot.snapshot_id if (snapshot :=
self._transaction.table_metadata.snapshot_by_name(self._target_branch)) else
None
+ )
+ self._snapshot_id = self._transaction.table_metadata.new_snapshot_id()
+ self._manifest_num_counter = itertools.count(0)
+ self.commit_uuid = uuid.uuid4()
+
+ def _validate_concurrency(self) -> None:
+ """Validate that concurrent changes do not conflict with this
operation.
+
+ Uses the CommitWindow to determine which catalog commits to validate
against.
+ The window spans from base (when the operation began) to head (latest
branch
+ HEAD), covering all external concurrent commits.
+
+ Subclasses that do not require validation (e.g. fast append) should
override
+ with a no-op.
+ """
+ from pyiceberg.table import TableProperties
+ from pyiceberg.table.snapshots import IsolationLevel
+ from pyiceberg.table.update.validate import (
+ _validate_added_data_files,
+ _validate_deleted_data_files,
+ _validate_no_new_delete_files,
+ _validate_no_new_deletes_for_data_files,
+ )
+
+ if self._commit_window is None or self._commit_window.is_empty():
+ return
+
+ catalog_head = self._commit_window.head
+ starting_snapshot = self._commit_window.base
+
+ if catalog_head is None or starting_snapshot is None:
+ return
+
+ table = self._transaction._table
+ isolation_level_str = table.metadata.properties.get(
+ self._isolation_level_property,
TableProperties.WRITE_ISOLATION_LEVEL_DEFAULT
+ )
+ isolation_level = IsolationLevel(isolation_level_str)
+ conflict_detection_filter = self._predicate if self._predicate !=
AlwaysFalse() else None
+
+ if isolation_level == IsolationLevel.SERIALIZABLE:
+ _validate_added_data_files(table, catalog_head,
conflict_detection_filter, starting_snapshot)
+
+ if conflict_detection_filter is not None:
Review Comment:
Why not simplify this by avoiding `conflict_detection_filter`:
```suggestion
if self._predicate != AlwaysFalse():
```
Makes it easier to read
##########
pyiceberg/table/__init__.py:
##########
@@ -1036,17 +1074,76 @@ def commit_transaction(self) -> Table:
The table with the updates applied.
"""
if len(self._updates) > 0:
- self._requirements +=
(AssertTableUUID(uuid=self.table_metadata.table_uuid),)
- self._table._do_commit( # pylint: disable=W0212
- updates=self._updates,
- requirements=self._requirements,
+ properties = self._table.metadata.properties
+ num_retries: int = property_as_int( # type: ignore # The default
is set with non-None value.
+ properties, TableProperties.COMMIT_NUM_RETRIES,
TableProperties.COMMIT_NUM_RETRIES_DEFAULT
+ )
+ min_wait_ms: int = property_as_int( # type: ignore # The default
is set with non-None value.
+ properties, TableProperties.COMMIT_MIN_RETRY_WAIT_MS,
TableProperties.COMMIT_MIN_RETRY_WAIT_MS_DEFAULT
+ )
+ max_wait_ms: int = property_as_int( # type: ignore # The default
is set with non-None value.
+ properties, TableProperties.COMMIT_MAX_RETRY_WAIT_MS,
TableProperties.COMMIT_MAX_RETRY_WAIT_MS_DEFAULT
)
+ total_timeout_ms: int = property_as_int( # type: ignore # The
default is set with non-None value.
+ properties, TableProperties.COMMIT_TOTAL_RETRY_TIME_MS,
TableProperties.COMMIT_TOTAL_RETRY_TIME_MS_DEFAULT
+ )
+ start_time = time.monotonic()
+ self._requirements +=
(AssertTableUUID(uuid=self.table_metadata.table_uuid),)
+
+ try:
+ for attempt in range(num_retries + 1):
+ try:
+ self._table._do_commit( # pylint: disable=W0212
+ updates=self._updates,
+ requirements=self._requirements,
+ )
+ self._cleanup_uncommitted_manifests()
+ break
+ except CommitFailedException:
+ elapsed_ms = (time.monotonic() - start_time) * 1000
Review Comment:
Should we emit a WARN here? So the user knows that the commit is being
retried? Otherwise the user might not know why it is taking its time.
##########
pyiceberg/table/update/snapshot.py:
##########
@@ -353,11 +397,116 @@ def new_manifest_output(self) -> OutputFile:
location_provider = self._transaction._table.location_provider()
file_name =
_new_manifest_file_name(num=next(self._manifest_num_counter),
commit_uuid=self.commit_uuid)
file_path = location_provider.new_metadata_location(file_name)
+ self._written_manifests.append(file_path)
return self._io.new_output(file_path)
def fetch_manifest_entry(self, manifest: ManifestFile, discard_deleted:
bool = True) -> list[ManifestEntry]:
return manifest.fetch_manifest_entry(io=self._io,
discard_deleted=discard_deleted)
+ def commit(self) -> None:
+ self._transaction._register_snapshot_producer(self)
+ super().commit()
+
+ def _cleanup_uncommitted(self) -> None:
+ """Delete manifest files and manifest lists from failed retry
attempts."""
+ for path in self._uncommitted_manifests:
+ try:
+ self._io.delete(path)
+ except Exception:
+ logger.warning("Failed to delete uncommitted manifest: %s",
path, exc_info=True)
+ self._uncommitted_manifests.clear()
+ # Delete all manifest lists except the last one (which the committed
snapshot references)
+ if len(self._written_manifest_lists) > 1:
+ for path in self._written_manifest_lists[:-1]:
+ try:
+ self._io.delete(path)
+ except Exception:
+ logger.warning("Failed to delete uncommitted manifest
list: %s", path, exc_info=True)
+ self._written_manifest_lists = self._written_manifest_lists[-1:]
+
+ def _clean_all_uncommitted(self) -> None:
+ """Clean up all manifests and manifest lists on abort."""
+ for path in itertools.chain(self._uncommitted_manifests,
self._written_manifests):
+ try:
+ self._io.delete(path)
+ except Exception:
+ logger.warning("Failed to delete uncommitted manifest: %s",
path, exc_info=True)
+ for path in self._written_manifest_lists:
+ try:
+ self._io.delete(path)
+ except Exception:
+ logger.warning("Failed to delete uncommitted manifest list:
%s", path, exc_info=True)
+ self._uncommitted_manifests.clear()
+ self._written_manifests.clear()
+ self._written_manifest_lists.clear()
+
+ def _refresh_for_retry(self) -> None:
+ """Reset state for a retry attempt with refreshed metadata."""
+ self._uncommitted_manifests.extend(self._written_manifests)
+ self._written_manifests.clear()
+ self._parent_snapshot_id = (
+ snapshot.snapshot_id if (snapshot :=
self._transaction.table_metadata.snapshot_by_name(self._target_branch)) else
None
+ )
+ self._snapshot_id = self._transaction.table_metadata.new_snapshot_id()
+ self._manifest_num_counter = itertools.count(0)
+ self.commit_uuid = uuid.uuid4()
+
+ def _validate_concurrency(self) -> None:
+ """Validate that concurrent changes do not conflict with this
operation.
+
+ Uses the CommitWindow to determine which catalog commits to validate
against.
+ The window spans from base (when the operation began) to head (latest
branch
+ HEAD), covering all external concurrent commits.
+
+ Subclasses that do not require validation (e.g. fast append) should
override
+ with a no-op.
+ """
+ from pyiceberg.table import TableProperties
+ from pyiceberg.table.snapshots import IsolationLevel
+ from pyiceberg.table.update.validate import (
+ _validate_added_data_files,
+ _validate_deleted_data_files,
+ _validate_no_new_delete_files,
+ _validate_no_new_deletes_for_data_files,
+ )
+
+ if self._commit_window is None or self._commit_window.is_empty():
+ return
+
+ catalog_head = self._commit_window.head
+ starting_snapshot = self._commit_window.base
+
+ if catalog_head is None or starting_snapshot is None:
+ return
+
+ table = self._transaction._table
+ isolation_level_str = table.metadata.properties.get(
+ self._isolation_level_property,
TableProperties.WRITE_ISOLATION_LEVEL_DEFAULT
+ )
+ isolation_level = IsolationLevel(isolation_level_str)
+ conflict_detection_filter = self._predicate if self._predicate !=
AlwaysFalse() else None
+
+ if isolation_level == IsolationLevel.SERIALIZABLE:
+ _validate_added_data_files(table, catalog_head,
conflict_detection_filter, starting_snapshot)
+
+ if conflict_detection_filter is not None:
+ _validate_no_new_delete_files(table, catalog_head,
conflict_detection_filter, None, starting_snapshot)
+ _validate_deleted_data_files(table, catalog_head,
conflict_detection_filter, starting_snapshot)
+
+ if self._deleted_data_files:
+ _validate_no_new_deletes_for_data_files(
+ table, catalog_head, conflict_detection_filter,
self._deleted_data_files, starting_snapshot
+ )
+
+ def _resolve_parent_snapshot(self) -> Snapshot | None:
Review Comment:
This method seems unused? Less is more 🚀
##########
pyiceberg/table/update/snapshot.py:
##########
@@ -132,8 +172,11 @@ def __init__(
self._parent_snapshot_id = (
snapshot.snapshot_id if (snapshot :=
self._transaction.table_metadata.snapshot_by_name(self._target_branch)) else
None
)
+ self._starting_snapshot_id = self._parent_snapshot_id
self._predicate = AlwaysFalse()
self._case_sensitive = True
+ self._commit_window = None
+ self._isolation_level_property: str =
TableProperties.WRITE_DELETE_ISOLATION_LEVEL
Review Comment:
I really dislike that we have to pass this in and around everywhere, it
really clutters up the code. Earlier I suggested to adding
`metadata.isolation_level(Operation)`, I think this would solve this and we can
just resolve it from the table.
Also, this will also take the latest properties into account. So, if in the
conflict, the isolation level has been updated, this will be taken into account
since the underlying properties will be refreshed 😄
--
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]
