-
Notifications
You must be signed in to change notification settings - Fork 28.3k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
[SPARK-24252][SQL] Add TableCatalog API #24246
Changes from 8 commits
f06c2e1
69ea102
94b7d74
43eb7ea
6a91081
cab35e9
c8edba0
8735b72
350cb1a
cdac39b
98c4622
5dfd9bf
06713bc
12796fb
File filter
Filter by extension
Conversations
Jump to
Diff view
Diff view
There are no files selected for viewing
Original file line number | Diff line number | Diff line change |
---|---|---|
@@ -0,0 +1,121 @@ | ||
/* | ||
* Licensed to the Apache Software Foundation (ASF) under one or more | ||
* contributor license agreements. See the NOTICE file distributed with | ||
* this work for additional information regarding copyright ownership. | ||
* The ASF licenses this file to You under the Apache License, Version 2.0 | ||
* (the "License"); you may not use this file except in compliance with | ||
* the License. You may obtain a copy of the License at | ||
* | ||
* http://www.apache.org/licenses/LICENSE-2.0 | ||
* | ||
* Unless required by applicable law or agreed to in writing, software | ||
* distributed under the License is distributed on an "AS IS" BASIS, | ||
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. | ||
* See the License for the specific language governing permissions and | ||
* limitations under the License. | ||
*/ | ||
|
||
package org.apache.spark.sql.catalog.v2; | ||
|
||
import org.apache.spark.sql.catalog.v2.expressions.Transform; | ||
import org.apache.spark.sql.catalyst.analysis.NoSuchNamespaceException; | ||
import org.apache.spark.sql.catalyst.analysis.NoSuchTableException; | ||
import org.apache.spark.sql.catalyst.analysis.TableAlreadyExistsException; | ||
import org.apache.spark.sql.sources.v2.Table; | ||
import org.apache.spark.sql.types.StructType; | ||
|
||
import java.util.Collections; | ||
import java.util.Map; | ||
|
||
/** | ||
* Catalog methods for working with Tables. | ||
*/ | ||
public interface TableCatalog extends CatalogPlugin { | ||
/** | ||
* List the tables in a namespace from the catalog. | ||
* | ||
* @param namespace a multi-part namespace | ||
* @return an array of Identifiers for tables | ||
* @throws NoSuchNamespaceException If the namespace does not exist (optional). | ||
*/ | ||
Identifier[] listTables(String[] namespace) throws NoSuchNamespaceException; | ||
|
||
/** | ||
* Load table metadata by {@link Identifier identifier} from the catalog. | ||
* | ||
* @param ident a table identifier | ||
* @return the table's metadata | ||
* @throws NoSuchTableException If the table doesn't exist. | ||
*/ | ||
Table loadTable(Identifier ident) throws NoSuchTableException; | ||
There was a problem hiding this comment. Choose a reason for hiding this commentThe reason will be displayed to describe this comment to others. Learn more. In There was a problem hiding this comment. Choose a reason for hiding this commentThe reason will be displayed to describe this comment to others. Learn more. Also, the word "load" implies that, the table will be loaded to somewhere. Will we cache the table metadata somewhere inside Spark? There was a problem hiding this comment. Choose a reason for hiding this commentThe reason will be displayed to describe this comment to others. Learn more. The table is loaded so Spark can use it. This usage not unusual for the word "load". The name I think that the fact that this takes an identifier and returns a There was a problem hiding this comment. Choose a reason for hiding this commentThe reason will be displayed to describe this comment to others. Learn more. will we support the Hive LOAD TABLE in the future? There was a problem hiding this comment. Choose a reason for hiding this commentThe reason will be displayed to describe this comment to others. Learn more. @cloud-fan,
|
||
|
||
/** | ||
* Refresh table metadata for {@link Identifier identifier}. | ||
* <p> | ||
* If the table is already loaded or cached, drop cached data and reload it. If the table is not | ||
There was a problem hiding this comment. Choose a reason for hiding this commentThe reason will be displayed to describe this comment to others. Learn more. Not sure if this is in the scope of this PR, but I was wondering if we could make the metadata/schema for a table un cacheable by default. For example, in the current V1 code if I reference a relation in the Hive catalog the session will cache the metadata associated with the table. This means that I can't update a user's view of the in catalog table without a manual user refresh. I was hoping we can signal via the catalog api that a table's metadata is dynamic and should not be cached. There was a problem hiding this comment. Choose a reason for hiding this commentThe reason will be displayed to describe this comment to others. Learn more. Right now, we've implemented caching in our catalogs. Spark always calls load and uses the version that is returned instead of caching. Our catalog then uses a weak reference in the cache so that the version is periodically updated when no more queries are running against that version. I'm not sure whether Spark should have some caching above the catalog or if the catalog should be responsible for this. Let's plan on talking about this at the next sync. There was a problem hiding this comment. Choose a reason for hiding this commentThe reason will be displayed to describe this comment to others. Learn more. sgtm |
||
* already loaded, load it and update caches with the new version. | ||
* | ||
* @param ident a table identifier | ||
* @return the table's current metadata | ||
* @throws NoSuchTableException If the table doesn't exist. | ||
*/ | ||
default Table refreshTable(Identifier ident) throws NoSuchTableException { | ||
There was a problem hiding this comment. Choose a reason for hiding this commentThe reason will be displayed to describe this comment to others. Learn more. This API design looks a bit weird, I think a more common approach is:
When Spark want to get a fresh table, just call There was a problem hiding this comment. Choose a reason for hiding this commentThe reason will be displayed to describe this comment to others. Learn more. Refresh table doesn't just require invalidating caches, it requires loading the table data at the time that refresh is called. Otherwise, it would be There was a problem hiding this comment. Choose a reason for hiding this commentThe reason will be displayed to describe this comment to others. Learn more. The API behavior also seems a bit weird to me. There was a problem hiding this comment. Choose a reason for hiding this commentThe reason will be displayed to describe this comment to others. Learn more. Do we have a precedent for how this is handled in other DBMS standards? There was a problem hiding this comment. Choose a reason for hiding this commentThe reason will be displayed to describe this comment to others. Learn more. I'm fine renaming this to There was a problem hiding this comment. Choose a reason for hiding this commentThe reason will be displayed to describe this comment to others. Learn more. I also think this API There was a problem hiding this comment. Choose a reason for hiding this commentThe reason will be displayed to describe this comment to others. Learn more. So you're saying that this should be There was a problem hiding this comment. Choose a reason for hiding this commentThe reason will be displayed to describe this comment to others. Learn more. Here's what /**
* Invalidate cached table metadata for an {@link Identifier identifier}.
* <p>
* If the table is already loaded or cached, drop cached data. If the table does not exist or is
* not cached, do nothing. Calling this method should not query remote services.
*
* @param ident a table identifier
*/
default void invalidateTable(Identifier ident) {
} Would you prefer those requirements over the semantics of There was a problem hiding this comment. Choose a reason for hiding this commentThe reason will be displayed to describe this comment to others. Learn more. In the DSv2 sync, no one had a strong opinion on this, but since since @cloud-fan, @gengliangwang, and @gatorsmile have all pointed this out, I went ahead and made this change. It is now Note that |
||
return loadTable(ident); | ||
} | ||
|
||
/** | ||
* Test whether a table exists using an {@link Identifier identifier} from the catalog. | ||
* | ||
* @param ident a table identifier | ||
* @return true if the table exists, false otherwise | ||
*/ | ||
default boolean tableExists(Identifier ident) { | ||
try { | ||
return loadTable(ident) != null; | ||
} catch (NoSuchTableException e) { | ||
return false; | ||
} | ||
} | ||
|
||
/** | ||
* Create a table in the catalog. | ||
* | ||
* @param ident a table identifier | ||
* @param schema the schema of the new table, as a struct type | ||
* @param partitions transforms to use for partitioning data in the table | ||
* @param properties a string map of table properties | ||
* @return metadata for the new table | ||
* @throws TableAlreadyExistsException If a table already exists for the identifier | ||
* @throws UnsupportedOperationException If a requested partition transform is not supported | ||
* @throws NoSuchNamespaceException If the identifier namespace does not exist (optional) | ||
*/ | ||
Table createTable( | ||
Identifier ident, | ||
StructType schema, | ||
Transform[] partitions, | ||
Map<String, String> properties) throws TableAlreadyExistsException, NoSuchNamespaceException; | ||
|
||
/** | ||
* Apply a set of {@link TableChange changes} to a table in the catalog. | ||
* <p> | ||
* Implementations may reject the requested changes. If any change is rejected, none of the | ||
* changes should be applied to the table. | ||
* | ||
* @param ident a table identifier | ||
* @param changes changes to apply to the table | ||
* @return updated metadata for the table | ||
* @throws NoSuchTableException If the table doesn't exist. | ||
* @throws IllegalArgumentException If any change is rejected by the implementation. | ||
*/ | ||
Table alterTable( | ||
Identifier ident, | ||
TableChange... changes) throws NoSuchTableException; | ||
|
||
/** | ||
* Drop a table in the catalog. | ||
There was a problem hiding this comment. Choose a reason for hiding this commentThe reason will be displayed to describe this comment to others. Learn more. We need to clearly define the behavior when the input is a view or a temp view. If we do not expect users to specify view names here, we need to document it. There was a problem hiding this comment. Choose a reason for hiding this commentThe reason will be displayed to describe this comment to others. Learn more. All the table metadata APIs are facing the same issue. There was a problem hiding this comment. Choose a reason for hiding this commentThe reason will be displayed to describe this comment to others. Learn more. I'll add a comment for how to handle views if they share the same namespace. There was a problem hiding this comment. Choose a reason for hiding this commentThe reason will be displayed to describe this comment to others. Learn more. I updated the javadoc for all of the methods to include how to handle views in the backing catalog. |
||
* | ||
* @param ident a table identifier | ||
* @return true if a table was deleted, false if no table exists for the identifier | ||
*/ | ||
boolean dropTable(Identifier ident); | ||
There was a problem hiding this comment. Choose a reason for hiding this commentThe reason will be displayed to describe this comment to others. Learn more. Hi @rdblue dropTable() here does not support to specify purge or not. Would you please share your idea about why it is designed like that? There was a problem hiding this comment. Choose a reason for hiding this commentThe reason will be displayed to describe this comment to others. Learn more. Purge is an option specific to some sources, like Hive. If you drop a table in most JDBC databases, you don't have an option to keep the data around somewhere. If an option can't be satisfied by sources, then it isn't a good candidate to be in the API. Even in Hive, purge is optional because managed tables will drop data automatically, but external tables will not. Using table configuration for sources that support "external" data is a better option, I think. There was a problem hiding this comment. Choose a reason for hiding this commentThe reason will be displayed to describe this comment to others. Learn more. for now maybe we can fail the analysis if PURGE is specified but we are dropping a v2 table. @waterlx would you like to send a PR to do it? |
||
} |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I am still not sure whether we should have a dedicated ViewCatalog for view management, like what the proposal suggests. For the catalog API developers, metadata management of views and tables are very similar. Do they need to implement two catalogs? cc @rxin @marmbrus
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I think the only similarity is that the two typically share the same namespace. The implementation is very different and will use different code paths. I think that it over-complicates the API to return the two using the same load method, when the alternative is to have two resolution rules instead.
That said, it would be easier to understand what you're asking for if you wrote up a proposed API. I can add this topic to the next v2 sync if you plan to propose an alternative.