Add table statistics #1285
Add table statistics #1285
Conversation
ndrluis
force-pushed
the
add-statistics
branch
from
6f0bee0
to
a70edb2
Compare
ndrluis
force-pushed
the
add-statistics
branch
from
a70edb2
to
384e229
Compare
|
I plan to move the set/remove statistics methods from the Transaction class to another class, such as ManageSnapshot. In the meantime, I’d like to confirm with everyone if I’m heading in the right direction with the current implementation. |
| statistics_path: str = Field(alias="statistics-path") | ||
| file_size_in_bytes: int = Field(alias="file-size-in-bytes") | ||
| file_footer_size_in_bytes: int = Field(alias="file-footer-size-in-bytes") | ||
| blob_metadata: List[BlobMetadata] = Field(alias="blob-metadata") |
There was a problem hiding this comment.
nit, missing key_metadata
https://iceberg.apache.org/spec/#table-statistics
pyiceberg/table/__init__.py
Outdated
| The alter table builder. | ||
| """ | ||
| updates = ( | ||
| RemoveStatisticsUpdate( |
There was a problem hiding this comment.
do you mind linking the java implementation? do we want to remove all stats?
There was a problem hiding this comment.
I understand that we want to remove the statistics of a specific snapshot, and I understand that we have one statistics file per snapshot.
The equivalent code would be the SetStatistics class, which follows the same pattern as our ManageSnapshot class. This is the scenario I want to double-check to ensure we follow the same pattern.
pyiceberg/table/update/__init__.py
Outdated
| if update.snapshot_id != update.statistics.snapshot_id: | ||
| raise ValueError("Snapshot id in statistics does not match the snapshot id in the update") | ||
|
|
||
| rest_statistics = [stat for stat in base_metadata.statistics if stat.snapshot_id != update.snapshot_id] |
There was a problem hiding this comment.
nit: this can be a helper function to filter on snapshot_id
| }, | ||
| { | ||
| "snapshot-id": 3055729675574597004, | ||
| "statistics-path": "s3://a/b/stats.puffin", |
There was a problem hiding this comment.
this does file need to exist on disk?
There was a problem hiding this comment.
No, there is no validation in place. This is only used for clients that support puffin files and for the expire snapshot procedure, which removes this information from the metadata. If the user wants, they can also remove the file as part of the expire snapshot procedure.
ndrluis
force-pushed
the
add-statistics
branch
2 times, most recently
from
9b15c86
to
d16ef47
Compare
|
@kevinjqliu could you please review it once more? |
| table.update_statistics() | ||
| .set_statistics(snapshot_id1, statistics_file1) | ||
| .remove_statistics(snapshot_id2) | ||
| # Operations are applied on commit. |
There was a problem hiding this comment.
nit: add .commit() instead of the comment
There was a problem hiding this comment.
I added the commit() and kept the comment consistent with the other examples.
mkdocs/docs/api.md
Outdated
| update.set_statistics(1, statistics_file) | ||
| update.remove_statistics(2) |
There was a problem hiding this comment.
nit: replace 1/2 with snapshot_id1/snapshot_id2 to show the input relation
pyiceberg/table/statistics.py
Outdated
| blob_metadata: List[BlobMetadata] = Field(alias="blob-metadata") | ||
|
|
||
|
|
||
| def reject_statistics( |
There was a problem hiding this comment.
nit: how about filter_statistics_by_snapshot_id?
@kevinjqliu As far as I know, only Trino can generate them. What kind of test would you like to have? I believe we are covering all relevant cases for this PR. If PyIceberg could generate or read puffin files, then I agree it would be useful to add tests to check compatibility between engines. However, I think it only makes sense to test puffin files during reading, as testing generation would mean verifying the implementation of something that isn’t our responsibility. In this case, it’s just a metadata update. What do you think? |
ndrluis
force-pushed
the
add-statistics
branch
from
d16ef47
to
11120bf
Compare
The Java expire snapshot process expires table statistics and partition statistics. I am implementing a statistics table to make our expire snapshot compatible with the Java implementation.